AI Enhanced Civil Infrastructure Safety

Topic Information

Dear Colleagues,

Due to the critical role of civil infrastructure in modern society, it should be able to remain safe and reliable under service environments or accident disasters, such as earthquakes, rockfalls, tsunamis, fires, blasts, etc. Maintaining the safety of civil infrastructure was, is and will continue to be a significant research topic. Although magnificent progress has been made, there are still critical challenges related to the demand for more accurate, efficient and pragmatic safety assessment and analysis of civil infrastructures under multiple scenes due to the intrinsic failure mechanism of materials and the large uncertainty within external effects. With more high-performance materials being introduced, this challenge becomes trickier. However, with the rapid development of the AI field, a brand-new opportunity has emerged to reveal these mechanisms and uncertainties and tackle the above challenge with AI's assistance. From this perspective, this topic aims to invite relevant scholars and collect the innovative outcomes of their research in civil and infrastructural safety via AI-enhanced, multi-disciplinary principles. We hope this Topic will be a platform for sharing novel knowledge and stimulating new ideas. The specific topics include, but are not limited to, new developments in the following:

• Data-driven material and component performance prediction
• AI-enhanced structural behavior analysis
• Structural design upgraded by AI
• AI-aided structure construction techniques
• Structure maintenance with smart sensing
• Structural damage inspection based on AI
• AI applications in structural health monitoring
• Smart structural maintenance management
• AI-aided optimization of conformation and structure.

Dr. Shizhi Chen
Dr. Jingfeng Zhang
Dr. Ekin Ozer
Dr. Zilong Ti
Dr. Xiaoming Lei
Topic Editors

Keywords

Participating Journals

Journal Name	Impact Factor	CiteScore	Launched Year	First Decision (median)	APC
Applied Sciences applsci	2.5	5.3	2011	17.8 Days	CHF 2400
Buildings buildings	3.1	3.4	2011	17.2 Days	CHF 2600
Infrastructures infrastructures	2.7	5.2	2016	16.8 Days	CHF 1800
Materials materials	3.1	5.8	2008	15.5 Days	CHF 2600
Sensors sensors	3.4	7.3	2001	16.8 Days	CHF 2600
Remote Sensing remotesensing	4.2	8.3	2009	24.7 Days	CHF 2700
Inventions inventions	2.1	4.8	2016	21.2 Days	CHF 1800

MDPI Topics is cooperating with Preprints.org and has built a direct connection between MDPI journals and Preprints.org. Authors are encouraged to enjoy the benefits by posting a preprint at Preprints.org prior to publication:

1. Immediately share your ideas ahead of publication and establish your research priority;
2. Protect your idea from being stolen with this time-stamped preprint article;
3. Enhance the exposure and impact of your research;
4. Receive feedback from your peers in advance;
5. Have it indexed in Web of Science (Preprint Citation Index), Google Scholar, Crossref, SHARE, PrePubMed, Scilit and Europe PMC.

Published Papers (33 papers)

Order results

Content type Publication Date

Result details

Normal Extended Compact

Journals

All Applied Sciences Buildings Infrastructures Materials Sensors Remote Sensing Inventions

Show export options expand_more Show export options expand_less

Select all

Export citation of selected articles as:

Plain Text BibTeX BibTeX (without abstracts) Endnote Endnote (without abstracts) Tab-delimited RIS

Error

Oops... you haven't selected anything for export.

Ok

clear

23 pages, 3848 KiB  

Open AccessArticle

SC-YOLOv8 Network with Soft-Pooling and Attention for Elevator Passenger Detection

by Zhiheng Wang, Jiayan Chen, Ping Yu, Bin Feng and Da Feng

Appl. Sci. 2024, 14(8), 3321; https://doi.org/10.3390/app14083321 - 15 Apr 2024

Cited by 1 | Viewed by 1506

Abstract

This paper concentrates on the elevator passenger detection task, a pivotal element for subsequent elevator passenger tracking and behavior recognition, crucial for ensuring passenger safety. To enhance the accuracy of detecting passenger positions inside elevators, we improved the YOLOv8 network and proposed the [...] Read more.

This paper concentrates on the elevator passenger detection task, a pivotal element for subsequent elevator passenger tracking and behavior recognition, crucial for ensuring passenger safety. To enhance the accuracy of detecting passenger positions inside elevators, we improved the YOLOv8 network and proposed the SC-YOLOv8 elevator passenger detection network with soft-pooling and attention mechanisms. The main improvements in this paper encompass the following aspects: Firstly, we transformed the convolution module (ConvModule) of the YOLOv8 backbone network by introducing spatial and channel reconstruction convolution (SCConv). This improvement aims to reduce spatial and channel redundancy in the feature extraction process of the backbone network, thereby improving the overall efficiency and performance of the detection network. Secondly, we propose a dual-branch SPP-Fast module by incorporating a soft-pooling branch into the YOLOv8 network’s SPP-Fast module. This dual-branch SPP-Fast module can preserve essential information while reducing the impact of noise. Finally, we propose a soft-pooling and multi-scale convolution CBAM module to further enhance the network’s performance. This module enhances the network’s focus on key regions, allowing for more targeted feature extraction, thereby further improving the accuracy of object detection. Additionally, the attention module enhances the network’s robustness in handling complex backgrounds. We conducted experiments on an elevator passenger dataset. The results show that the precision, recall, and mAP of our improved YOLOv8 network are 94.32%, 91.17%, and 92.95%, respectively, all surpassing those of the original YOLOv8 network. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Structure of SC-YOLOv8 network with soft-pooling and attention.</p> Full article ">Figure 2
<p>The architecture of SCConv.</p> Full article ">Figure 3
<p>The architecture of the spatial reconstruction unit.</p> Full article ">Figure 4
<p>The architecture of the channel reconstruction unit.</p> Full article ">Figure 5
<p>SC-ConvModule. In SC-ConvModule, the feature map first passes through an SCConv layer for feature extraction and transformation. Immediately after that, the BatchNorm submodule performs batch normalization on the output of the SCConv layer to adjust the distribution of the data, making the model training more stable and efficient. Subsequently, the normalized data enter the SiLU activation function to perform a nonlinear transformation to further increase the expressive power of the model.</p> Full article ">Figure 6
<p>Structure of ELAN.</p> Full article ">Figure 7
<p>SC-CSP_2Conv module. The feature maps fed into the SC-CSP_2Conv module are first processed by the SC-ConvModule for feature extraction. Subsequently, the feature map processed by SC-ConvModule enters the Split layer, which is divided into multiple parallel channels. The data in these channels are processed by four consecutive SC-Bottleneck layers, and all the output feature maps from SC-Bottleneck are recombined in the Concat layer to form a feature map with multi-channel information. Finally, this merged feature map passes through an SC-ConvModule again for final feature extraction and transformation to obtain the output of the SC-CSP_2Conv module.</p> Full article ">Figure 8
<p>Bottleneck and SC-Bottleneck.</p> Full article ">Figure 9
<p>Structure of dual-branch SPP-Fast.</p> Full article ">Figure 10
<p>CBAM attention.</p> Full article ">Figure 11
<p>Channel attention module with soft-pooling.</p> Full article ">Figure 12
<p>Spatial attention module with soft-pooling and multi-scale depth-wise separable convolutions.</p> Full article ">Figure 13
<p>Partial example of elevator passenger dataset.</p> Full article ">Figure 14
<p>LabelImg marking interface.</p> Full article ">Figure 15
<p>Tag file format of xml.</p> Full article ">Figure 16
<p>Example of data augmentation.</p> Full article ">Figure 17
<p>Training results for YOLOv8 and our network.</p> Full article ">Figure 18
<p>Loss curve.</p> Full article ">Figure 19
<p>YOLOv 8 network visualization results.</p> Full article ">Figure 20
<p>Our network visualization results.</p> Full article ">

25 pages, 13415 KiB  

Open AccessArticle

The Use of Lidar and Artificial Intelligence Algorithms for Detection and Size Estimation of Potholes

by Sk Abu Talha, Dmitry Manasreh and Munir D. Nazzal

Buildings 2024, 14(4), 1078; https://doi.org/10.3390/buildings14041078 - 12 Apr 2024

Cited by 4 | Viewed by 2537

Abstract

Road potholes have a well-known impact on driving quality and safety. Therefore, timely mitigation of potholes is critical for the safety of road users. However, efficient and timely maintenance relies on the presence of an effective process for pothole detection. Currently, transportation agencies [...] Read more.

Road potholes have a well-known impact on driving quality and safety. Therefore, timely mitigation of potholes is critical for the safety of road users. However, efficient and timely maintenance relies on the presence of an effective process for pothole detection. Currently, transportation agencies primarily rely on manual inspection and road user reporting. These methods are subjective, prone to inaccuracy, and some are also laborious and time-consuming. An ideal pothole detection system would be accurate, objective, automated, and relatively inexpensive. In this context, accuracy encompasses three distinct performance areas: detection, localization, and size estimation. This study explores the potential of utilizing a mobile light detection and ranging (LiDAR) for accurate detection and size estimation, along with a global navigation satellite system (GNSS) receiver for localization, to develop an effective pothole surveillance system. To achieve this objective, the study proposes a four-step framework. Firstly, the LiDAR data are processed to generate ring-wise cross-sectional images. Secondly, a deep learning object detection network is trained to predict the presence and size of potholes. Thirdly, the ring-wise inferences are aggregated to produce a final decision. Lastly, the aggregated inferences are synchronized with GNSS locations to generate inspection maps. The system’s performance was validated using multiple road strips, never seen by the model, containing potholes of different sizes and shapes. The results demonstrated the effectiveness and accuracy of the proposed system. Overall, this research contributes to the research on LiDAR-based pothole inspection by proposing a novel four-step framework and incorporating it into an end-to-end pothole detection system, which can greatly improve the efficiency of pothole maintenance and enhance the safety of road users. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>The data collection system: (<b>a</b>) top view, (<b>b</b>) side view.</p> Full article ">Figure 2
<p>Implementation of the proposed algorithm in ROS.</p> Full article ">Figure 3
<p>Flowchart of the point cloud data processing.</p> Full article ">Figure 4
<p>Schematic diagram of the overlap between two subsequent LiDAR scans at a speed of 70 mph.</p> Full article ">Figure 5
<p>Rotation and trimming of the point cloud.</p> Full article ">Figure 6
<p>Conversion of point cloud to 2D histogram.</p> Full article ">Figure 7
<p>Pothole detection from the 2D histograms.</p> Full article ">Figure 8
<p>Consecutive LiDAR rings corresponding to the same pothole (overlap).</p> Full article ">Figure 9
<p>Subsequent LiDAR rings corresponding to different potholes (non-overlap).</p> Full article ">Figure 10
<p>Implementation of the pothole counting algorithm.</p> Full article ">Figure 11
<p>Prediction bounding box.</p> Full article ">Figure 12
<p>Point-click procedure to obtain the actual dimension of the pothole.</p> Full article ">Figure 13
<p>(<b>a</b>) Correlation between actual and predicted width, and (<b>b</b>) correlation between actual and predicted depth.</p> Full article ">Figure 14
<p>Determination of length.</p> Full article ">Figure 15
<p>Correlation between actual length and predicted length.</p> Full article ">Figure 16
<p>Confusion matrix: (<b>a</b>) YOLO v5s, and (<b>b</b>) YOLO v5n.</p> Full article ">Figure 17
<p>Accuracy comparison between YOLO v5s and YOLO v5n.</p> Full article ">Figure 18
<p>Inference speed comparison between YOLO v5s and YOLO v5n.</p> Full article ">Figure 19
<p>Effect of background image percentage on false negative detection.</p> Full article ">Figure 20
<p>Verification of the pothole count algorithm: (<b>a</b>) image of captured by camera (<b>b</b>) LiDAR scan corresponding to the image.</p> Full article ">Figure 21
<p>(<b>a</b>) Location of the testing strip on State Route 126, and (<b>b</b>) detected potholes.</p> Full article ">Figure 22
<p>Measured vs. predicted dimension: (<b>a</b>) depth, (<b>b</b>) width, and (<b>c</b>) length.</p> Full article ">Figure 23
<p>Test strip (blue line) at Interstate 71 North.</p> Full article ">Figure 24
<p>Comparison between actual and predicted dimensions for Run 1 and Run 2 at 55 mph: (<b>a</b>) depth, (<b>b</b>) width, and (<b>c</b>) length.</p> Full article ">Figure 25
<p>Comparison between actual and predicted dimensions for Run 2 (55 mph) and run at 65 mph: (<b>a</b>) depth, (<b>b</b>) width, and (<b>c</b>) length.</p> Full article ">

14 pages, 11517 KiB  

Open AccessArticle

Analysis of Arch Bridge Condition Data to Identify Network-Wide Controls and Trends

by Kristopher Campbell, Myra Lydon, Nicola-Ann Stevens and Su Taylor

Infrastructures 2024, 9(4), 70; https://doi.org/10.3390/infrastructures9040070 - 4 Apr 2024

Viewed by 1627

Abstract

This paper outlines an initial analysis of 20 years of data held on an electronic bridge management database for approximately 3500 arch bridges across Northern Ireland (NI) by the Department for Infrastructure. Arch bridges represent the largest group of bridge types, making up [...] Read more.

This paper outlines an initial analysis of 20 years of data held on an electronic bridge management database for approximately 3500 arch bridges across Northern Ireland (NI) by the Department for Infrastructure. Arch bridges represent the largest group of bridge types, making up nearly 56% of the total bridge stock in NI. This initial analysis aims to identify trends that might help inform maintenance decisions in the future. Consideration of the Bridge Condition Indicator (BCI) average value for the overall arch bridge stock indicates the potential for regional variations in the overall condition and the potential for human bias in inspections. The paper presents the most prevalent structural elements and associated defects recorded in the inspections of arch bridges. This indicated a link to scour and undermining for the worst-conditioned arch bridges. An Analysis of Variance (ANOVA) analysis identified function, number of spans, and deck width as significant factors during the various deterioration stages in a bridge’s lifecycle. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Grouped BCI average scores for all current arch bridge inspections in NI.</p> Full article ">Figure 2
<p>Four geographical divisional boundaries in NI (northern, eastern, southern, and western).</p> Full article ">Figure 3
<p>BCI Average scores for all current arch bridge inspections in NI are plotted in ascending order per divisional area.</p> Full article ">Figure 4
<p>Probabilistic distribution of BCI scores by region. Eastern (ED), Northern (ND), Southern (SD) and Western (WD).</p> Full article ">Figure 5
<p>Distribution of cumulative spans for all arch bridges in NI.</p> Full article ">Figure 6
<p>Comparison between defect prevalence for all arch bridges (All) and State 4 arch bridges.</p> Full article ">Figure 7
<p>Comparison between defect prevalence for all arch bridges (All) and State 4 arch bridges, with adjusted <span class="html-italic">y</span>-axis for clarity.</p> Full article ">Figure 8
<p>Typical defect photo shows scour to abutments, missing masonry, pointing loss, and cracking.</p> Full article ">Figure 9
<p>Typical defect photo shows scour to abutments and vegetation cover including trees.</p> Full article ">Figure 10
<p>Comparison of component defect breakdown for all arch bridges (All) and State 4 arch bridges for selected defects and bridge elements.</p> Full article ">

15 pages, 7399 KiB  

Open AccessArticle

Energy-Efficient Mixtures Suitable for 3D Technologies

by Leonid Dvorkin, Vitaliy Marchuk, Katarzyna Mróz, Marcin Maroszek and Izabela Hager

Appl. Sci. 2024, 14(7), 3038; https://doi.org/10.3390/app14073038 - 4 Apr 2024

Cited by 2 | Viewed by 1313

Abstract

Compositions of fine-grained concrete mixtures that provide the minimum required strength values in 1 day (7.5 MPa) have been developed. A comparison was made of the test results of the properties of samples printed on a 3D printer with samples made according to [...] Read more.

Compositions of fine-grained concrete mixtures that provide the minimum required strength values in 1 day (7.5 MPa) have been developed. A comparison was made of the test results of the properties of samples printed on a 3D printer with samples made according to the same recipes on a vibrating platform. A laboratory printer was designed and constructed to study the properties of extruded mixtures. The method was also proposed for measuring concrete mixes’ structural strength. Analysis of experimental data allowed the establishment of the features of the influence of the mineral additives and slag–alkaline binders for a comparison of basic physical and mechanical properties of concretes for 3D printing. It has been experimentally shown that possible undercompaction of the fine-grained mixtures formed on a 3D printer and decrease of properties are compensated by the introduction of hardening activator and superplasticizer additives. The novelty of this work lies in determining the comparative effect of various products of technogenic origin on the properties of mixtures for 3D printing. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Photograph showing the defects when applying layers. Bonding zone—in red.</p> Full article ">Figure 2
<p>Photograph showing the laboratory 3D printer: 1—electric motor of the extruder; 2—hopper of building mixture; 3—auger; 4—replaceable nozzle; 5—control panel; 6—frequency converter of electricity; 7—reverse motor moving the extruder in the horizontal direction; 8—manual drive moving the extruder in the vertical direction; 9—frame; 10—power cable of electric motors.</p> Full article ">Figure 3
<p>Photograph showing the device for determining the structural strength.</p> Full article ">Figure 4
<p>An example of determining the structural strength of extruded concrete: (<b>a</b>) the sample withstands the load (structural strength is provided) structural strength &gt; 4500 Pa; (<b>b</b>) the sample is destroyed.</p> Full article ">Figure 5
<p>Photograph showing the test samples were made using a 3D printer.</p> Full article ">Figure 6
<p>Graphic depictions of the mobility of mixtures for a 3D printer.</p> Full article ">Figure 7
<p>Graphic depictions of the initial setting time of mixtures for a 3D printer.</p> Full article ">Figure 8
<p>Graphic depictions of the structural strength of mixtures for a 3D printer (<b>a</b>) 20 min after mixing and (<b>b</b>) 40 min after mixing.</p> Full article ">Figure 9
<p>Graphic depictions of the average density of mixtures for a 3D printer.</p> Full article ">Figure 10
<p>Graphic depictions of bending strength of mixtures for a 3D printer.</p> Full article ">Figure 11
<p>Graphic depictions of compressive strength of mixtures for a 3D printer.</p> Full article ">Figure 12
<p>Graphic depictions of the splitting strength of compacted concrete on a vibrating platform and with the help of a 3D printer.</p> Full article ">Figure 13
<p>Graphic depictions of the bending strength of mixtures produced on a vibrating platform and with the help of a 3D printer.</p> Full article ">Figure 14
<p>Destruction of samples during the determination of bending strength for slag–alkaline (<b>a</b>) and the cement–ash binders (<b>b</b>).</p> Full article ">Figure 15
<p>Destruction of samples during the determination of splitting strength: (<b>a</b>) destruction along the boundary of superimposed layers (slag–alkaline binder, composition No. 5, <a href="#applsci-14-03038-t006" class="html-table">Table 6</a>); (<b>b</b>) destruction beyond the boundaries of superimposed layers (cement–ash binder, composition No. 2, <a href="#applsci-14-03038-t006" class="html-table">Table 6</a>).</p> Full article ">Figure 16
<p>Graphic depictions of the compressive strength of samples made on a vibrating platform and with the help of a 3D printer.</p> Full article ">

23 pages, 41214 KiB  

Open AccessArticle

A Deep Learning Approach for Surface Crack Classification and Segmentation in Unmanned Aerial Vehicle Assisted Infrastructure Inspections

by Shamendra Egodawela, Amirali Khodadadian Gostar, H. A. D. Samith Buddika, A. J. Dammika, Nalin Harischandra, Satheeskumar Navaratnam and Mojtaba Mahmoodian

Sensors 2024, 24(6), 1936; https://doi.org/10.3390/s24061936 - 18 Mar 2024

Cited by 4 | Viewed by 2155

Abstract

Surface crack detection is an integral part of infrastructure health surveys. This work presents a transformative shift towards rapid and reliable data collection capabilities, dramatically reducing the time spent on inspecting infrastructures. Two unmanned aerial vehicles (UAVs) were deployed, enabling the capturing of [...] Read more.

Surface crack detection is an integral part of infrastructure health surveys. This work presents a transformative shift towards rapid and reliable data collection capabilities, dramatically reducing the time spent on inspecting infrastructures. Two unmanned aerial vehicles (UAVs) were deployed, enabling the capturing of images simultaneously for efficient coverage of the structure. The suggested drone hardware is especially suitable for the inspection of infrastructure with confined spaces that UAVs with a broader footprint are incapable of accessing due to a lack of safe access or positioning data. The collected image data were analyzed using a binary classification convolutional neural network (CNN), effectively filtering out images containing cracks. A comparison of state-of-the-art CNN architectures against a novel CNN layout “CrackClassCNN” was investigated to obtain the optimal layout for classification. A Segment Anything Model (SAM) was employed to segment defect areas, and its performance was benchmarked against manually annotated images. The suggested “CrackClassCNN” achieved an accuracy rate of 95.02%, and the SAM segmentation process yielded a mean Intersection over Union (IoU) score of 0.778 and an F1 score of 0.735. It was concluded that the selected UAV platform, the communication network, and the suggested processing techniques were highly effective in surface crack detection. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Ultra-lightweight UAV used in the study—DJI Ryze Tech Tello. The Tello is a compact UAV with small dimensions, designed to be easily maneuverable in confined spaces.</p> Full article ">Figure 2
<p>Methodology for data acquisition and the crack-detection process. This flowchart illustrates the step-by-step process for acquiring data and performing crack detection.</p> Full article ">Figure 3
<p>Communication network for data collection through UAVs. Shows the work involved in coordinating the movements and positions of the UAVs. User-defined waypoints and trajectories guided the drones during the mission, ensuring comprehensive coverage of the target structure.</p> Full article ">Figure 4
<p>Image data collected of the structure under inspection using the UAVs. The Subfigures (<b>a</b>–<b>f</b>) shows the surface conditions with clearly visible cracks (<b>a</b>–<b>e</b>), exhibiting a combination of horizontal and vertical fractures, multiple cracks, and evident masonry plaster deterioration and barely visible defects (<b>f</b>).</p> Full article ">Figure 4 Cont.
<p>Image data collected of the structure under inspection using the UAVs. The Subfigures (<b>a</b>–<b>f</b>) shows the surface conditions with clearly visible cracks (<b>a</b>–<b>e</b>), exhibiting a combination of horizontal and vertical fractures, multiple cracks, and evident masonry plaster deterioration and barely visible defects (<b>f</b>).</p> Full article ">Figure 5
<p>Comparison of original blurry image (<b>a</b>) versus deblurred image (<b>b</b>) achieved using the MPRNet CNN architecture. While the original image exhibits extreme blurriness, the deblurred image showcases remarkably sharp and well-defined crack edges.</p> Full article ">Figure 6
<p>Visual representation of outputs of different denoising algorithms. (<b>a</b>) Gaussian denoising; (<b>b</b>) bilateral denoising; (<b>c</b>) wavelet denoising; (<b>d</b>) total variation denoising; (<b>e</b>) non-local means denoising; (<b>f</b>) shift invariant wavelet denoising; (<b>g</b>) anisotropic diffusion denoising; (<b>h</b>) block-matching denoising.</p> Full article ">Figure 7
<p>The CrackClassCNN architecture employs stacked convolutional layers for local feature extraction, followed by max-pooling layers to reduce dimensions. A densely connected layer culminates in a single-neuron output layer, optimizing speed and accuracy for crack detection.</p> Full article ">Figure 8
<p>Variation in training (solid line) and validation accuracy (dashed line) of the different CNN architectures.</p> Full article ">Figure 9
<p>The confusion matrices before and after the denoising and blurring processing steps show the clear favorable bias of the classifier due to denoising. The figure showcases the classifier’s performance in distinguishing between positive and negative instances. The main diagonal contains true negatives where the classifier correctly identified the class. The transposed main diagonal contains the classifier’s incorrect predictions. (<b>a</b>) Before pre-processing; (<b>b</b>) after pre-processing.</p> Full article ">Figure 10
<p>Visualization of feature maps of CrackClassCNN: The subfigures (<b>a</b>–<b>d</b>) illustrates the feature maps extracted from layers of the CNN. Each subfigure represents the output of a specific layer, showcasing the network’s ability to capture progressively abstract and complex patterns in the input image. The learned features in the subfigures show the hierarchical representation learned by the CNN.</p> Full article ">Figure 11
<p>SAM layout, courtesy of Kirillov et al. [<a href="#B31-sensors-24-01936" class="html-bibr">31</a>].</p> Full article ">Figure 12
<p>Classifier failure analysis. The image classifier encounters two main issues: false positives and false negatives. False positives include misidentifying non-existent cracks, especially confusing spalling with cracks due to similar appearances as shown in (<b>a</b>–<b>d</b>). False negatives occur when cracks, particularly barely visible ones, are missed as shown in (<b>e</b>). Scale relative to the camera’s view is critical, potentially causing smaller cracks to be overlooked (<b>f</b>). Extreme exposure issues causes both False positives and True negatives the cracks not being detected as shown in (<b>g</b>,<b>h</b>).</p> Full article ">Figure 12 Cont.
<p>Classifier failure analysis. The image classifier encounters two main issues: false positives and false negatives. False positives include misidentifying non-existent cracks, especially confusing spalling with cracks due to similar appearances as shown in (<b>a</b>–<b>d</b>). False negatives occur when cracks, particularly barely visible ones, are missed as shown in (<b>e</b>). Scale relative to the camera’s view is critical, potentially causing smaller cracks to be overlooked (<b>f</b>). Extreme exposure issues causes both False positives and True negatives the cracks not being detected as shown in (<b>g</b>,<b>h</b>).</p> Full article ">

19 pages, 2383 KiB  

Open AccessArticle

Analyzing the Factors Influencing Time Delays in Korean Railroad Accidents

by Ji-Myong Kim and Kwang-Kyun Lim

Appl. Sci. 2024, 14(5), 1697; https://doi.org/10.3390/app14051697 - 20 Feb 2024

Viewed by 1254

Abstract

Railroads play a pivotal role in the Korean national economy, necessitating a thorough understanding of factors influencing accidents for effective mitigation strategies. Unlike prior research focused on accident frequency and severity, this study delves into the often-overlooked aspect of time delays resulting from [...] Read more.

Railroads play a pivotal role in the Korean national economy, necessitating a thorough understanding of factors influencing accidents for effective mitigation strategies. Unlike prior research focused on accident frequency and severity, this study delves into the often-overlooked aspect of time delays resulting from railroad accidents. Analyzing 15 years of nationwide data (2008–2022), encompassing 3244 human-related and 3350 technical events, this research identifies key factors influencing delay likelihood and duration. Factors considered include event type, season, train type, location, operator size, person type involved, facility type, and causes. Despite an overall decrease in events, variable delay times highlight the need to comprehend specific contributing factors. To address excess zeros, the study employs a two-stage model and a zero-inflated negative binomial (ZINB) model, alongside artificial neural networks (ANNs) for non-linear pattern recognition. Human-related delays are influenced by event types, seasons, and passenger categories, exhibit nuanced impacts. Technical-related delays are influenced by incident types and facility involvement. Regarding model performance, the ANN models outperform regression-based models consistently in all cases. This study emphasizes the importance of considering both human and technical factors in predicting and understanding railroad accident delays, offering valuable insights for formulating strategies to mitigate service disruptions associated with these incidents. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Trends in the number of accidents, incidents, and fatalities per accident.</p> Full article ">Figure 2
<p>Ratio of railroad accidents and incidents with time delay records.</p> Full article ">Figure 3
<p>Number of significant accidents of UIC members in 32 different countries (see page 27 of the UIC Safety Report 2023).</p> Full article ">Figure 4
<p>Causes of accidents of UIC member in 32 different countries (see page 28 of the UIC Safety Report 2023).</p> Full article ">Figure 5
<p>Distribution of delay time due to human and technical factors.</p> Full article ">Figure 6
<p>The architecture of artificial neural networks.</p> Full article ">Figure 7
<p>Flow of computational processes within an artificial neural network model.</p> Full article ">

18 pages, 5299 KiB  

Open AccessArticle

A Machine-Learning-Based Failure Mode Classification Model for Reinforced Concrete Columns Using Simple Structural Information

by Subin Kim, Heejin Hwang, Keunyeong Oh and Jiuk Shin

Appl. Sci. 2024, 14(3), 1243; https://doi.org/10.3390/app14031243 - 2 Feb 2024

Cited by 3 | Viewed by 1426

Abstract

The seismically deficient column details in existing reinforced concrete buildings affect the overall behavior of the building depending on the failure type of the column. The purpose of this study is to develop and validate a machine-learning-based prediction model for the column failure [...] Read more.

The seismically deficient column details in existing reinforced concrete buildings affect the overall behavior of the building depending on the failure type of the column. The purpose of this study is to develop and validate a machine-learning-based prediction model for the column failure modes (shear, flexure–shear, and flexure failure modes). For this purpose, artificial neural network (ANN), K-nearest neighbor (KNN), decision tree (DT), and random forest (RF) models were used considering previously collected experimental data. Using four machine learning methodologies, we developed a classification learning model that can predict the column failure modes in terms of the input variables using the concrete compressive strength, steel yield strength, axial load ratio, height-to-dept aspect ratio, longitudinal reinforcement ratio, and transverse reinforcement ratio. The performance of each machine learning model was compared and verified by calculating the accuracy, precision, recall, F1-Score, and ROC. Based on the performance measurements of the classification model, the RF model has the highest average value for the classification model performance measurements among the considered learning methods and can conservatively predict the shear failure mode. Thus, the RF model can rapidly predict the column failure modes with the simple column details. Additionally, it was demonstrated that the predicted failure modes from the selected model were exactly same as the failure mode determined from a code-defined equation (traditional method). Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Constituents of the reinforced concrete column database.</p> Full article ">Figure 2
<p>Frequency of input parameters.</p> Full article ">Figure 3
<p>Confusion matrix of each machine learning methodology for the testing dataset.</p> Full article ">Figure 4
<p>ROC curve for all machine learning models.</p> Full article ">Figure 5
<p>Relative importance of input variables.</p> Full article ">Figure 6
<p>Flexure-governed column modeling method.</p> Full article ">Figure 7
<p>Shear-governed column modeling method.</p> Full article ">Figure 8
<p>Validation of modeling method (experiment vs. simulation) [<a href="#B1-applsci-14-01243" class="html-bibr">1</a>,<a href="#B2-applsci-14-01243" class="html-bibr">2</a>,<a href="#B3-applsci-14-01243" class="html-bibr">3</a>].</p> Full article ">Figure 9
<p>Traditional method predicting column failure mode.</p> Full article ">Figure 10
<p>Relationships between shear demand curve and shear capacity curve [<a href="#B1-applsci-14-01243" class="html-bibr">1</a>,<a href="#B2-applsci-14-01243" class="html-bibr">2</a>,<a href="#B3-applsci-14-01243" class="html-bibr">3</a>].</p> Full article ">

18 pages, 9114 KiB  

Open AccessArticle

Experimental Study on Measuring and Tracking Structural Displacement Based on Surveillance Video Image Analysis

by Tongyuan Ni, Liuqi Wang, Xufeng Yin, Ziyang Cai, Yang Yang, Deyu Kong and Jintao Liu

Sensors 2024, 24(2), 601; https://doi.org/10.3390/s24020601 - 17 Jan 2024

Cited by 1 | Viewed by 1493

Abstract

The digital image method of monitoring structural displacement is receiving more attention today, especially in non-contact structure health monitoring. Some obvious advantages of this method, such as economy and convenience, were shown while it was used to monitor the deformation of the bridge [...] Read more.

The digital image method of monitoring structural displacement is receiving more attention today, especially in non-contact structure health monitoring. Some obvious advantages of this method, such as economy and convenience, were shown while it was used to monitor the deformation of the bridge structure during the service period. The image processing technology was used to extract structural deformation feature information from surveillance video images containing structural displacement in order to realize a new non-contact online monitoring method in this paper. The influence of different imaging distances and angles on the conversion coefficient (η) that converts the pixel coordinates to the actual displacement was first studied experimentally. Then, the measuring and tracking of bridge structural displacement based on surveillance video images was investigated by laboratory-scale experiments under idealized conditions. The results showed that the video imaging accuracy can be affected by changes in the relative position of the imaging device and measured structure, which is embodied in the change in η (actual size of individual pixel) on the structured image. The increase in distance between the measured structure and the monitoring equipment will have a significant effect on the change in the η value. The value of η varies linearly with the change in shooting distance. The value of η will be affected by the changes in shooting angle. The millimeter-level online monitoring of the structure displacement can be realized using images based on surveillance video images. The feasibility of measuring and tracking structural displacement based on surveillance video images was confirmed by a laboratory-scale experiment. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Road surveillance equipment and value video image information acquisition process.</p> Full article ">Figure 2
<p>The relationship between perspective and location.</p> Full article ">Figure 3
<p>The calibration experiment using a checkerboard pattern.</p> Full article ">Figure 4
<p>Schematic of the correction principle and correction effect comparison.</p> Full article ">Figure 5
<p>Structural displacement measurement experiment scheme using surveillance cameras.</p> Full article ">Figure 6
<p>The relationship between <span class="html-italic">η</span> and <b><span class="html-italic">d</span></b>.</p> Full article ">Figure 7
<p>The relationship between <span class="html-italic">η</span> and <b><span class="html-italic">θ</span></b>. (<b>a</b>) a distance of 0.5 m; (<b>b</b>) a distance of 1.0 m; (<b>c</b>) a distance of 1.5 m; (<b>d</b>) a distance of 2.0 m; (<b>e</b>) a distance of 2.5; (<b>f</b>) a distance of 3.0 m; (<b>g</b>) a distance of 3.5 m; (<b>h</b>) a distance of 4.0 m; (<b>i</b>) a distance of 4.5 m; and (<b>j</b>) a distance of 5.0 m.</p> Full article ">Figure 7 Cont.
<p>The relationship between <span class="html-italic">η</span> and <b><span class="html-italic">θ</span></b>. (<b>a</b>) a distance of 0.5 m; (<b>b</b>) a distance of 1.0 m; (<b>c</b>) a distance of 1.5 m; (<b>d</b>) a distance of 2.0 m; (<b>e</b>) a distance of 2.5; (<b>f</b>) a distance of 3.0 m; (<b>g</b>) a distance of 3.5 m; (<b>h</b>) a distance of 4.0 m; (<b>i</b>) a distance of 4.5 m; and (<b>j</b>) a distance of 5.0 m.</p> Full article ">Figure 7 Cont.
<p>The relationship between <span class="html-italic">η</span> and <b><span class="html-italic">θ</span></b>. (<b>a</b>) a distance of 0.5 m; (<b>b</b>) a distance of 1.0 m; (<b>c</b>) a distance of 1.5 m; (<b>d</b>) a distance of 2.0 m; (<b>e</b>) a distance of 2.5; (<b>f</b>) a distance of 3.0 m; (<b>g</b>) a distance of 3.5 m; (<b>h</b>) a distance of 4.0 m; (<b>i</b>) a distance of 4.5 m; and (<b>j</b>) a distance of 5.0 m.</p> Full article ">Figure 8
<p>The relative relationship between <span class="html-italic">η</span> and structure position.</p> Full article ">Figure 9
<p>The relationship between the <span class="html-italic">η</span> value and imaging posture.</p> Full article ">Figure 10
<p>Three-dimensional fitting between <span class="html-italic">η</span> and the location of the structure (<b><span class="html-italic">d</span></b>, <b><span class="html-italic">θ</span></b>).</p> Full article ">Figure 11
<p>Time registration for tracking and monitoring (<b>a</b>) without eliminating time error; (<b>b</b>) eliminating time error.</p> Full article ">Figure 12
<p>Displacement meter monitoring and digital image tracking and monitoring under static load.</p> Full article ">Figure 13
<p>Displacement meter monitoring and image tracking and monitoring under moving load.</p> Full article ">Figure 14
<p>Tracking and monitoring after the secondary time registration.</p> Full article ">Figure 15
<p>Absolute error of structural deformation monitoring experiments with (<b>a</b>) static load; (<b>b</b>) moving load.</p> Full article ">Figure 16
<p>Relative error of structural deformation monitoring experiments.</p> Full article ">

20 pages, 2272 KiB  

Open AccessArticle

Modeling the Optimal Maintenance Strategy for Bridge Elements Based on Agent Sequential Decision Making

by Gongfeng Xin, Zhiqiang Liang, Yerong Hu, Guanxu Long, Yang Zhang and Peng Liang

Appl. Sci. 2024, 14(1), 14; https://doi.org/10.3390/app14010014 - 19 Dec 2023

Cited by 1 | Viewed by 1031

Abstract

In addressing the issues of low efficiency in bridge maintenance decision making, the inaccurate estimation of maintenance costs, and the lack of specificity in decision making regarding maintenance measures for specific defects, this study utilizes data from regular bridge inspections. It employs a [...] Read more.

In addressing the issues of low efficiency in bridge maintenance decision making, the inaccurate estimation of maintenance costs, and the lack of specificity in decision making regarding maintenance measures for specific defects, this study utilizes data from regular bridge inspections. It employs a two-parameter Weibull distribution to model the duration variables of the states of bridge elements, thereby enabling the prediction of the duration time of bridge elements in various states. Referring to existing bridge maintenance and repair regulations, the estimation process of maintenance costs is streamlined. Taking into account the specific types and development state of bridge defects, as well as considering the adequacy of maintenance and the restorative effects of maintenance measures, an intelligent agent sequential decision-making model for bridge maintenance decisions is established. The model utilizes dynamic programming algorithms to determine the optimal maintenance and repair measures for elements in various states. The decision results are precise, all the way down to the specific bridge elements and maintenance measures for individual defects. In using the case of the regular inspection data of 222 bridges along a highway loop, this study further validates the effectiveness of the proposed research methods. By constructing an intelligent agent sequential decision-making model for bridge element maintenance, the optimal maintenance measures for 21 bridge elements in different states are obtained, thereby significantly enhancing the efficiency of actual bridge maintenance and the practicality of decision results. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Decision flow chart of bridge project-level maintenance.</p> Full article ">Figure 2
<p>Flow chart of the policy iteration method.</p> Full article ">Figure 3
<p>The route bridges’ distribution of a highway loop.</p> Full article ">Figure 4
<p>Selected bridge inspection photos. (<b>a</b>) Side view and (<b>b</b>) bottom view.</p> Full article ">Figure 5
<p>Diagram of the maintenance plans under different parameters in the sixth year.</p> Full article ">Figure 6
<p>Diagram of the maintenance plans under different parameters in the second year.</p> Full article ">

16 pages, 4270 KiB  

Open AccessArticle

A Practical Data Extraction, Cleaning, and Integration Method for Structural Condition Assessment of Highway Bridges

by Gongfeng Xin, Fidel Lozano Galant, Wenwu Zhang, Ye Xia and Guoquan Zhang

Infrastructures 2023, 8(12), 183; https://doi.org/10.3390/infrastructures8120183 - 18 Dec 2023

Cited by 1 | Viewed by 2278

Abstract

The success of regional bridge condition assessment, a crucial component of systematic maintenance strategies, relies heavily on comprehensive, well-structured regional bridge databases. This study proposes the data extraction, cleaning, and integration method for the construction of such databases. First, this research proposes an [...] Read more.

The success of regional bridge condition assessment, a crucial component of systematic maintenance strategies, relies heavily on comprehensive, well-structured regional bridge databases. This study proposes the data extraction, cleaning, and integration method for the construction of such databases. First, this research proposes an extraction method tailored for unstructured data often present in inspection reports. Additionally, this paper meticulously outlines a cleaning procedure designed to rectify two distinct categories of typical errors that are present within the inspection data. Subsequently, this study takes a holistic approach by establishing integration rules that harmonize data from various sources, including inspection records, monitoring data, traffic statistics, as well as design and construction blueprints. The architectural framework of the regional bridge information database is then meticulously laid out. To validate and demonstrate the effectiveness of the method, this study applies them to a set of representative highway bridges situated within Shandong Province. The results show that this approach can be used to successfully establish a functional regional bridge database. The database plays a pivotal role in harnessing the latent potential of an extensive range of multi-source information and propels the field of bridge condition assessment forward by providing a solid basis for informed decision making and strategic planning in the realm of infrastructure maintenance. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Data Integration Rules and Structures.</p> Full article ">Figure 2
<p>Sample Structural Condition Evaluation Forms from Bridge Inspection Reports.</p> Full article ">Figure 3
<p>Continuous and Discrete Structural Condition Level Functions Under Different Scoring Systems.</p> Full article ">Figure 4
<p>Example Demonstrating Score Conversion Between Scoring Systems.</p> Full article ">Figure 5
<p>Distribution of Deviation Coefficients for Converted Scores After Data Cleaning.</p> Full article ">Figure 6
<p>Distribution of Span Lengths for Bridges in the Dataset.</p> Full article ">Figure 7
<p>Distribution of Girder Section Types Among Beam Bridges in the Dataset.</p> Full article ">Figure 8
<p>Changes in Distribution of Bridge Condition Scores Over Time.</p> Full article ">Figure 9
<p>Changes in Distribution of Bridge Condition Levels Over Time.</p> Full article ">

18 pages, 6083 KiB  

Open AccessArticle

Predicting Trajectories of Plate-Type Wind-Borne Debris in Turbulent Wind Flow with Uncertainties

by Feng Wang, Peng Huang, Rongxin Zhao, Huayong Wu, Mengjin Sun, Zijie Zhou and Yun Xing

Infrastructures 2023, 8(12), 180; https://doi.org/10.3390/infrastructures8120180 - 15 Dec 2023

Viewed by 1998

Abstract

Debris poses multifaceted risks and jeopardizes various aspects of the environment, human health, safety, and infrastructure. The debris trajectory in turbulent wind flow is more dispersed due to the inherent randomness of the turbulent winds. This paper investigates the three-dimensional trajectories of plate-type [...] Read more.

Debris poses multifaceted risks and jeopardizes various aspects of the environment, human health, safety, and infrastructure. The debris trajectory in turbulent wind flow is more dispersed due to the inherent randomness of the turbulent winds. This paper investigates the three-dimensional trajectories of plate-type wind-borne debris in turbulent wind fields via the method of numerical simulation. A 3D probabilistic trajectory model of plate-type wind-borne debris is developed. The debris trajectories are numerically calculated by solving the governing equation of debris motion and by introducing turbulent wind flows based on the near-ground wind field measured in the wind tunnel to account for the probability characteristics of the debris trajectory. The dimensionless velocities and displacements of the debris trajectory show good agreement with the experimental data in wind tunnel tests, confirming the rationality of the probabilistic trajectory model. Based on the validated trajectory model, the probability characteristics of the debris impact position, impact velocity, and kinetic energy, debris angular displacement, and angular velocity are analyzed in detail under five different wind attack angles. The proposed probabilistic model of plate-type debris in turbulent wind flow provides an accurate and effective method for predicting debris trajectory in three-dimensional space. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>(<b>a</b>) Body-fixed coordinate system (<span class="html-italic">Xp</span>, <span class="html-italic">Yp</span>, <span class="html-italic">Zp</span>) and (<b>b</b>) global inertial reference frame (<span class="html-italic">Xg</span>, <span class="html-italic">Yg</span>, <span class="html-italic">Zg</span>) and transnational coordinate (<span class="html-italic">Xt</span>, <span class="html-italic">Yt</span>, <span class="html-italic">Zt</span>).</p> Full article ">Figure 2
<p>Normal force coefficients for plates with side length ratios equal to 2 [<a href="#B18-infrastructures-08-00180" class="html-bibr">18</a>].</p> Full article ">Figure 3
<p>Illustration showing the model building and definition of wind directions and initial debris position.</p> Full article ">Figure 4
<p>Velocity profile and turbulence intensity profile of the approaching wind flow.</p> Full article ">Figure 5
<p>An example of wind speed history measured at the initial debris release position in the wind tunnel.</p> Full article ">Figure 6
<p>(<b>a</b>) Longitudinal and (<b>b</b>) lateral and vertical spectra of the fluctuating wind velocity at the debris release point.</p> Full article ">Figure 7
<p>Nine examples of (<b>a</b>) dimensionless displacement vs. dimensionless velocity and (<b>b</b>) dimensionless flight time vs. dimensionless displacement (compare with Lin et al., 2006) [<a href="#B11-infrastructures-08-00180" class="html-bibr">11</a>,<a href="#B14-infrastructures-08-00180" class="html-bibr">14</a>].</p> Full article ">Figure 8
<p>Probability density distribution of (<b>a</b>) mean wind speed, (<b>b</b>) turbulence intensity, and (<b>c</b>) turbulence integral scale of the one hundred 1 s wind speed time histories.</p> Full article ">Figure 8 Cont.
<p>Probability density distribution of (<b>a</b>) mean wind speed, (<b>b</b>) turbulence intensity, and (<b>c</b>) turbulence integral scale of the one hundred 1 s wind speed time histories.</p> Full article ">Figure 9
<p>One hundred examples of positions of debris center and the dimensional horizontal flight velocity of the debris at wind attack angles of (<b>a</b>) 0°, (<b>b</b>) 15°, (<b>c</b>) 30°, (<b>d</b>) 45°, and (<b>e</b>) 60°.</p> Full article ">Figure 9 Cont.
<p>One hundred examples of positions of debris center and the dimensional horizontal flight velocity of the debris at wind attack angles of (<b>a</b>) 0°, (<b>b</b>) 15°, (<b>c</b>) 30°, (<b>d</b>) 45°, and (<b>e</b>) 60°.</p> Full article ">Figure 10
<p>One hundred examples of debris landing positions varying with the lateral integral scale (color bar, unit: m) at wind attack angles of (<b>a</b>) 0°, (<b>b</b>) 15°, (<b>c</b>) 30°, (<b>d</b>) 45°, and (<b>e</b>) 60°.</p> Full article ">Figure 11
<p>One hundred examples of debris landing position varying with dimensionless impact kinetic energy (color bar) at wind attack angles of (<b>a</b>) 0°, (<b>b</b>) 15°, (<b>c</b>) 30°, (<b>d</b>) 45°, and (<b>e</b>) 60°.</p> Full article ">Figure 12
<p>(<b>a</b>) Probability density function and (<b>b</b>) cumulative density function of the dimensionless impact kinetic energy of debris at five wind attack angles.</p> Full article ">Figure 13
<p>Effect of wind attack angle on debris rotation angles expressed by Euler angles: (<b>a</b>) <span class="html-italic">φ</span>, (<b>b</b>) <span class="html-italic">θ</span>, and (<b>c</b>) <span class="html-italic">ψ</span>.</p> Full article ">Figure 14
<p>Effect of wind attack angle on debris rotation angular velocity: (<b>a</b>) <span class="html-italic">ω</span>_X, (<b>b</b>) <span class="html-italic">ω</span>_Y, and (<b>c</b>) <span class="html-italic">ω</span>_Z.</p> Full article ">Figure 15
<p>Debris rotation angles expressed by the Euler angles <span class="html-italic">φ</span>, <span class="html-italic">θ</span>, and <span class="html-italic">ψ</span>: (<b>a</b>–<b>c</b>) 0° and (<b>d</b>–<b>f</b>) 45° wind attack angle (the colored lines stand for the single simulation examples).</p> Full article ">Figure 16
<p>Debris rotation angular velocity expressed by <span class="html-italic">ω</span>_X, <span class="html-italic">ω</span>_Y, and <span class="html-italic">ω</span>_Z: (<b>a</b>–<b>c</b>) 0° and (<b>d</b>–<b>f</b>) 45° wind attack angles (the colored lines stand for the single simulation examples).</p> Full article ">

18 pages, 13626 KiB  

Open AccessArticle

Internal Tree Trunk Decay Detection Using Close-Range Remote Sensing Data and the PointNet Deep Learning Method

by Marek Hrdina and Peter Surový

Remote Sens. 2023, 15(24), 5712; https://doi.org/10.3390/rs15245712 - 13 Dec 2023

Cited by 1 | Viewed by 1338

Abstract

The health and stability of trees are essential information for the safety of people and property in urban greenery, parks or along roads. The stability of the trees is linked to root stability but essentially also to trunk decay. Currently used internal tree [...] Read more.

The health and stability of trees are essential information for the safety of people and property in urban greenery, parks or along roads. The stability of the trees is linked to root stability but essentially also to trunk decay. Currently used internal tree stem decay assessment methods, such as tomography and penetrometry, are reliable but usually time-consuming and unsuitable for large-scale surveys. Therefore, a new method based on close-range remotely sensed data, specifically close-range photogrammetry and iPhone LiDAR, was tested to detect decayed standing tree trunks automatically. The proposed study used the PointNet deep learning algorithm for 3D data classification. It was verified in three different datasets consisting of pure coniferous trees, pure deciduous trees, and mixed data to eliminate the influence of the detectable symptoms for each group and species itself. The mean achieved validation accuracies of the models were 65.5% for Coniferous trees, 58.4% for Deciduous trees and 57.7% for Mixed data classification. The accuracies indicate promising data, which can be either used by practitioners for preliminary surveys or for other researchers to acquire more input data and create more robust classification models. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Locations of study plots south of the city of Karlovy Vary and north of the town of Mašťov.</p> Full article ">Figure 2
<p>iPhone Lidar point cloud. Colours depict Verticality: 0 (Blue); 0.3 (Green); 0.6 (Yellow) 1 (Red).</p> Full article ">Figure 3
<p>(<b>a</b>) Acoustic tomograph Arbor Sonic 3D; (<b>b</b>) Tomograms of a healthy and decayed stem. Numerical values with colour–scale correspond to the speed of sound wave propagation.</p> Full article ">Figure 4
<p>Basic workflow of the data processing. The asterisk indicates steps that had to be done manually.</p> Full article ">Figure 5
<p>(<b>a</b>) Tomogram before adjustment; (<b>b</b>) Tomogram after adjustment by calculating more accurate sensor distances. Sensor locations are represented by the black dots with numerical values. Colours are explained in the <a href="#remotesensing-15-05712-f003" class="html-fig">Figure 3</a>b.</p> Full article ">Figure 6
<p>(<b>a</b>) Original CRP Point Cloud (<b>b</b>) Automatically separated stem (<b>c</b>) 3D Mesh of the stem.</p> Full article ">Figure 7
<p>The representation of validation accuracy changes with the type of data used. Other parameters are kept as described in the text.</p> Full article ">Figure 8
<p>The representation of validation accuracy changes with the number of epochs on the Coniferous dataset. Other parameters are kept as described in the text.</p> Full article ">Figure 9
<p>Representation of the validation accuracy change with changing number of epochs on the Deciduous dataset. Other parameters are kept as described in the text.</p> Full article ">Figure 10
<p>Representation of the validation accuracy change with changing number of epochs on the Mixed dataset. Other parameters are kept as described in the text.</p> Full article ">Figure 11
<p>Effect of varying Learning Rate on the Validation Accuracy of Mixed Data classifiers.</p> Full article ">

13 pages, 4772 KiB  

Open AccessArticle

Performance Comparison of Deep Learning Models for Damage Identification of Aging Bridges

by Su-Wan Chung, Sung-Sam Hong and Byung-Kon Kim

Appl. Sci. 2023, 13(24), 13204; https://doi.org/10.3390/app132413204 - 12 Dec 2023

Viewed by 1103

Abstract

Currently, damage in aging bridges is assessed visually, leading to significant personnel, time, and cost expenditures. Moreover, the results depend on the subjective judgment of the inspector. Machine-learning-based approaches, such as deep learning, can solve these problems. In particular, instance-segmentation models have been [...] Read more.

Currently, damage in aging bridges is assessed visually, leading to significant personnel, time, and cost expenditures. Moreover, the results depend on the subjective judgment of the inspector. Machine-learning-based approaches, such as deep learning, can solve these problems. In particular, instance-segmentation models have been used to identify different types of bridge damage. However, the value of deep-learning-based damage identification may be reduced by insufficient training data, class imbalance, and model-reliability issues. To overcome these limitations, this study utilized photographic data from real bridge-management systems for the inspection and assessment of bridges as the training dataset. Six types of damage were considered. Moreover, the performances of three representative deep learning models—Mask R-CNN, BlendMask, and SWIN—were compared in terms of loss–function values. SWIN showed the best performance, achieving a loss value of 0.000005 after 269,939 training iterations. This shows that bridge-damage-identification performance can be maximized by setting an appropriate learning rate and using a deep learning model with a minimal loss value. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Object-detection framework for identifying bridge damage by integrating deep learning combination–modules [<a href="#B12-applsci-13-13204" class="html-bibr">12</a>].</p> Full article ">Figure 2
<p>Mask R-CNN architecture.</p> Full article ">Figure 3
<p>BlendMask pipeline.</p> Full article ">Figure 4
<p>SWIN architecture.</p> Full article ">Figure 5
<p>Loss values for Mask R-CNN (learning rate: 0.01).</p> Full article ">Figure 6
<p>Loss values for BlendMask (learning rate: 0.01).</p> Full article ">Figure 7
<p>Loss values for Mask R-CNN (learning rate: 0.005).</p> Full article ">Figure 8
<p>Loss values for SWIN (learning rate: 0.005).</p> Full article ">Figure 9
<p>Loss values for Mask R-CNN (learning rate: 0.0001).</p> Full article ">Figure 10
<p>Loss values for BlendMask (learning rate: 0.0001).</p> Full article ">Figure 11
<p>Loss values for SWIN (learning rate: 0.0001).</p> Full article ">

23 pages, 9044 KiB  

Open AccessArticle

Seismic Performance Evaluation and Analysis of Vertical Hydrogen Storage Vessels Based on Shaking Table Testing

by Sangmoon Lee, Bubgyu Jeon and Wooyoung Jung

Appl. Sci. 2023, 13(24), 13190; https://doi.org/10.3390/app132413190 - 12 Dec 2023

Cited by 1 | Viewed by 1321

Abstract

In this study, the structural integrity of a system installed on protrusion concrete, considering the usability of a vertical hydrogen storage vessel, was verified. To achieve this, a site survey was conducted to select the target structure, and analytical validation was performed to [...] Read more.

In this study, the structural integrity of a system installed on protrusion concrete, considering the usability of a vertical hydrogen storage vessel, was verified. To achieve this, a site survey was conducted to select the target structure, and analytical validation was performed to design specimens for shaking table tests. Subsequently, dynamic behavior characteristics were analyzed using an artificial earthquake simulated according to the procedures outlined in ICC-ES AC 156, which is the seismic design criterion. As a result, it was observed that the seismic motion was amplified by approximately 10 times compared to the original load magnitude, based on the acceleration response of the test specimen. It is inferred that the seismic motion occurring during an earthquake could cause significant damage to both the internal and external aspects of the structure, depending on the structure’s form and the composition of materials. Through analytical verification and testing, it was revealed that the main structure of the test specimen and the anchor bolts for installation met the seismic performance criteria. However, the protrusion concrete area exhibited damage, indicating a structural vulnerability when subjected to external forces such as earthquakes. Consequently, on-site measures to address this structural risk need to be explored. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Damage to protrusion concrete caused by overturning under earthquake conditions.</p> Full article ">Figure 2
<p>Outline of the on-site investigation.</p> Full article ">Figure 3
<p>The configuration and support system of the target structures installed in each region (unit: mm): (<b>a</b>) Gangneung, (<b>b</b>) Samcheok, and (<b>c</b>) Ulsan.</p> Full article ">Figure 4
<p>Summary diagram of specimen design for shaking table testing (unit: mm).</p> Full article ">Figure 5
<p>Various components of FE modeling.</p> Full article ">Figure 6
<p>Friction coefficient of the contact for each material: (<b>a</b>) steel-to-steel and (<b>b</b>) steel-to-concrete.</p> Full article ">Figure 7
<p>Response of concrete to a uniaxial loading condition: (<b>a</b>) compression behavior and (<b>b</b>) tension behavior.</p> Full article ">Figure 8
<p>Seismic input applied in the numerical analysis (AC 156 Amp. 100%).</p> Full article ">Figure 9
<p>View of the first mode shape (scale: 1:200): (<b>a</b>) isometric and (<b>b</b>) axis X-Y: side.</p> Full article ">Figure 10
<p>Acceleration response by location of target structure according to seismic load: (<b>a</b>) footing concrete (A1), (<b>b</b>) support column (A2), (<b>c</b>) middle of the hydrogen storage vessel (A3), and (<b>d</b>) top of the hydrogen storage vessel (A4).</p> Full article ">Figure 11
<p>Results of comparison of areas where stress is concentrated: (<b>a</b>) connecting point (S1), (<b>b</b>) support column (S2), and (<b>c</b>) anchor bolts (S3).</p> Full article ">Figure 11 Cont.
<p>Results of comparison of areas where stress is concentrated: (<b>a</b>) connecting point (S1), (<b>b</b>) support column (S2), and (<b>c</b>) anchor bolts (S3).</p> Full article ">Figure 12
<p>Review area of protrusion concrete according to seismic loading.</p> Full article ">Figure 13
<p>Safety review of the protrusion concrete: (<b>a</b>) upper part (C1) and (<b>b</b>) connecting part (C2).</p> Full article ">Figure 14
<p>Required Response Spectrum (RRS) and Test Response Spectrum (TRS) based on ICC-ES AC 156 with Amp. 150%.</p> Full article ">Figure 15
<p>Test set-up and sensor location.</p> Full article ">Figure 16
<p>Failure mode of the test specimen.</p> Full article ">Figure 17
<p>Acceleration response history at each location based on shaking table testing and comparison between tests and analytical results: (<b>a</b>) footing concrete (A1), (<b>b</b>) support column (A2), (<b>c</b>) middle of the hydrogen storage vessel (A3), and (<b>d</b>) top of the hydrogen storage vessel (A4).</p> Full article ">Figure 17 Cont.
<p>Acceleration response history at each location based on shaking table testing and comparison between tests and analytical results: (<b>a</b>) footing concrete (A1), (<b>b</b>) support column (A2), (<b>c</b>) middle of the hydrogen storage vessel (A3), and (<b>d</b>) top of the hydrogen storage vessel (A4).</p> Full article ">Figure 18
<p>The Von Mises stress response results of the supporting column.</p> Full article ">

17 pages, 3118 KiB  

Open AccessArticle

Risk Assessment in the Design of Railroad Control Command and Signaling Devices Using Fuzzy Sets

by Przemysław Ilczuk and Magdalena Kycko

Appl. Sci. 2023, 13(22), 12460; https://doi.org/10.3390/app132212460 - 17 Nov 2023

Cited by 4 | Viewed by 1050

Abstract

Risk assessment in the design of control command and signaling devices (CCS) is one of the elements required by law. These analyses should be carried out at many stages of investment with the participation of various teams. This article presents a risk analysis [...] Read more.

Risk assessment in the design of control command and signaling devices (CCS) is one of the elements required by law. These analyses should be carried out at many stages of investment with the participation of various teams. This article presents a risk analysis method based on fuzzy sets, which can support and increase the safety of investment processes involving the railroad traffic control industry. The article analyzes hazards identified in CCS design. These risks were identified using a survey method based on a set of questions prepared by the authors and by conducting interviews among experts from design offices. As part of the survey, responses were obtained from 28 respondents who are specialists in the railway traffic control industry. Workshop meetings were held in six different design offices and at manufacturing plants of motion control systems. The identified risks were analyzed using the FMEA (failure mode and effect analysis) method and the fuzzy set method, as well as various methods of fuzzification and defuzzification. The results of all of the methods were compared with each other. The best solution from the analyzed ones was proposed. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>View of the window of the application created in MatLab R2022a (source: own elaboration).</p> Full article ">Figure 2
<p>Gaussian urgency membership function graph (source: own elaboration).</p> Full article ">Figure 3
<p>Trapezoidal urgency membership function graph (source: own elaboration).</p> Full article ">Figure 4
<p>Triangular urgency membership function graph (source: own elaboration).</p> Full article ">Figure 5
<p>Window of rules created in MatLab (source: own elaboration).</p> Full article ">Figure 6
<p>The process of applying fuzzy logic.</p> Full article ">Figure 7
<p>The comparison of different fuzzy logic methods.</p> Full article ">

16 pages, 5701 KiB  

Open AccessArticle

Deep Learning Based Fire Risk Detection on Construction Sites

by Hojune Ann and Ki Young Koo

Sensors 2023, 23(22), 9095; https://doi.org/10.3390/s23229095 - 10 Nov 2023

Cited by 4 | Viewed by 2095

Abstract

The recent large-scale fire incidents on construction sites in South Korea have highlighted the need for computer vision technology to detect fire risks before an actual occurrence of fire. This study developed a proactive fire risk detection system by detecting the coexistence of [...] Read more.

The recent large-scale fire incidents on construction sites in South Korea have highlighted the need for computer vision technology to detect fire risks before an actual occurrence of fire. This study developed a proactive fire risk detection system by detecting the coexistence of an ignition source (sparks) and a combustible material (urethane foam or Styrofoam) using object detection on images from a surveillance camera. Statistical analysis was carried out on fire incidences on construction sites in South Korea to provide insight into the cause of the large-scale fire incidents. Labeling approaches were discussed to improve the performance of the object detectors for sparks and urethane foams. Detecting ignition sources and combustible materials at a distance was discussed in order to improve the performance for long-distance objects. Two candidate deep learning models, Yolov5 and EfficientDet, were compared in their performance. It was found that Yolov5 showed slightly higher mAP performances: Yolov5 models showed mAPs from 87% to 90% and EfficientDet models showed mAPs from 82% to 87%, depending on the complexity of the model. However, Yolov5 showed distinctive advantages over EfficientDet in terms of easiness and speed of learning. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Fire accident cases on construction sites (<b>a</b>) Icheon Refrigerated Warehouse site 2008. (<b>b</b>) Goyang Bus Terminal site 2014.</p> Full article ">Figure 2
<p>35-ft rule for cutting or welding operation in NFPA 51b [<a href="#B1-sensors-23-09095" class="html-bibr">1</a>].</p> Full article ">Figure 3
<p>Ignition sources in fire incidents on construction sites in South Korea.</p> Full article ">Figure 4
<p>Combustible materials in fire incidents on construction sites.</p> Full article ">Figure 5
<p>Performance of Yolov5 and EfficietDet [<a href="#B35-sensors-23-09095" class="html-bibr">35</a>].</p> Full article ">Figure 6
<p>Two labeling approaches and their performance on sparks. (<b>a</b>) Individual labeling. (<b>b</b>) Whole labeling. (<b>c</b>) AP.</p> Full article ">Figure 7
<p>Two labeling approaches and their performance on urethane foam. (<b>a</b>) Individual labeling. (<b>b</b>) Whole labeling. (<b>c</b>) AP.</p> Full article ">Figure 8
<p>Labeling approach and its performance on Styrofoam. (<b>a</b>) Labeling. (<b>b</b>) AP.</p> Full article ">Figure 9
<p>Short- and long-distance labeling approaches used for sparks. (<b>a</b>) Short-distance labeling. (<b>b</b>) Long-distance labeling. (<b>c</b>) Performance.</p> Full article ">Figure 10
<p>Short- and long-distance labeling approaches used for urethane foam. (<b>a</b>) Short-distance labeling. (<b>b</b>) Long-distance labeling. (<b>c</b>) Performance.</p> Full article ">Figure 11
<p>Short- and long-distance labeling approaches used for Styrofoam. (<b>a</b>) Short-distance labeling. (<b>b</b>) Long-distance labeling. (<b>c</b>) Performance.</p> Full article ">Figure 12
<p>Performance comparison of Yolov5 and EfficientDet networks on sparks, urethane foam, and Styrofoam.</p> Full article ">Figure 13
<p>Typical detection result of the trained Yolov5 or EfficientDet. (<b>a</b>) Sparks detection (Yolov5s). (<b>b</b>) Sparks detection (EfficientDet-d1). (<b>c</b>) Urethane foam detection (Yolov5s). (<b>d</b>) Urethane foam detection (EfficientDet-d1). (<b>e</b>) Styrofoam detection (Yolov5s). (<b>f</b>) Styrofoam detection (EfficientDet-d1).</p> Full article ">Figure 13 Cont.
<p>Typical detection result of the trained Yolov5 or EfficientDet. (<b>a</b>) Sparks detection (Yolov5s). (<b>b</b>) Sparks detection (EfficientDet-d1). (<b>c</b>) Urethane foam detection (Yolov5s). (<b>d</b>) Urethane foam detection (EfficientDet-d1). (<b>e</b>) Styrofoam detection (Yolov5s). (<b>f</b>) Styrofoam detection (EfficientDet-d1).</p> Full article ">Figure 14
<p>Example of fire risk detection on the construction site.</p> Full article ">

27 pages, 13491 KiB  

Open AccessArticle

Safety Evaluation of Reinforced Concrete Structures Using Multi-Source Fusion Uncertainty Cloud Inference and Experimental Study

by Zhao Liu, Huiyong Guo and Bo Zhang

Sensors 2023, 23(20), 8638; https://doi.org/10.3390/s23208638 - 22 Oct 2023

Cited by 1 | Viewed by 1392

Abstract

Structural damage detection and safety evaluations have emerged as a core driving force in structural health monitoring (SHM). Focusing on the multi-source monitoring data in sensing systems and the uncertainty caused by initial defects and monitoring errors, in this study, we develop a [...] Read more.

Structural damage detection and safety evaluations have emerged as a core driving force in structural health monitoring (SHM). Focusing on the multi-source monitoring data in sensing systems and the uncertainty caused by initial defects and monitoring errors, in this study, we develop a comprehensive method for evaluating structural safety, named multi-source fusion uncertainty cloud inference (MFUCI), that focuses on characterizing the relationship between condition indexes and structural performance in order to quantify the structural health status. Firstly, based on cloud theory, the cloud numerical characteristics of the condition index cloud drops are used to establish the qualitative rule base. Next, the proposed multi-source fusion generator yields a multi-source joint certainty degree, which is then transformed into cloud drops with certainty degree information. Lastly, a quantitative structural health evaluation is performed through precision processing. This study focuses on the numerical simulation of an RC frame at the structural level and an RC T-beam damage test at the component level, based on the stiffness degradation process. The results show that the proposed method is effective at evaluating the health of components and structures in a quantitative manner. It demonstrates reliability and robustness by incorporating uncertainty information through noise immunity and cross-domain inference, outperforming baseline models such as Bayesian neural network (BNN) in uncertainty estimations and LSTM in point estimations. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Principles of antecedent and consequent cloud generators.</p> Full article ">Figure 2
<p>Multi-source fusion generator.</p> Full article ">Figure 3
<p>The basic architecture of the proposed MFUCI.</p> Full article ">Figure 4
<p>RC frame reinforcement diagram.</p> Full article ">Figure 5
<p>Comparison between simulation and experimental load–displacement skeleton curves and performance point segmentation.</p> Full article ">Figure 6
<p>The health degree quantization values corresponding to each working condition of the RC frame.</p> Full article ">Figure 7
<p>The monitoring index evaluation system of the RC frame.</p> Full article ">Figure 8
<p>The antecedent and consequent affiliation cloud maps for the RC frame.</p> Full article ">Figure 9
<p>Trends in concrete strain signal curves at different noise levels.</p> Full article ">Figure 10
<p>Comparison of the degradation curves of the proposed method and the baseline models in the RC frame for (<b>a</b>) MFUCI, (<b>b</b>) BNN, (<b>c</b>) LSTM, (<b>d</b>) rebar strain-based, (<b>e</b>) displacement angle-based, and (<b>f</b>) concrete strain-based.</p> Full article ">Figure 11
<p>The RC T-beam components and the reinforcement diagrams.</p> Full article ">Figure 12
<p>The RC T-beam configuration details.</p> Full article ">Figure 13
<p>Analysis of load–displacement curves for SJ-T-1 and SJ-T-2.</p> Full article ">Figure 14
<p>Analysis of load–strain curves for rebars at bottom of SJ-T-1 and SJ-T-2.</p> Full article ">Figure 15
<p>Analysis of load–strain curves for concrete in the pure bending section of SJ-T-1 and SJ-T-2.</p> Full article ">Figure 16
<p>The antecedent and consequent affiliation cloud maps for the RC T-beams.</p> Full article ">Figure 17
<p>Comparison of the (<b>a</b>) degradation curves and (<b>b</b>) accuracy between the proposed method and the baseline models with samples of SJ-T-1 as the inference sets.</p> Full article ">

16 pages, 1301 KiB  

Open AccessArticle

Probabilistic Shear Strength Prediction for Deep Beams Based on Bayesian-Optimized Data-Driven Approach

by Mao-Yi Liu, Zheng Li and Hang Zhang

Buildings 2023, 13(10), 2471; https://doi.org/10.3390/buildings13102471 - 28 Sep 2023

Cited by 6 | Viewed by 1127

Abstract

To ensure the safety of buildings, accurate and robust prediction of a reinforced concrete deep beam’s shear capacity is necessary to avoid unpredictable accidents caused by brittle failure. However, the failure mechanism of reinforced concrete deep beams is very complicated, has not been [...] Read more.

To ensure the safety of buildings, accurate and robust prediction of a reinforced concrete deep beam’s shear capacity is necessary to avoid unpredictable accidents caused by brittle failure. However, the failure mechanism of reinforced concrete deep beams is very complicated, has not been fully elucidated, and cannot be accurately described by simple equations. To solve this issue, machine learning techniques have been utilized and corresponding prediction models have been developed. Nevertheless, these models can only provide deterministic prediction results of the scalar type, and the confidence level is uncertain. Thus, these prediction results cannot be used for the design and assessment of deep beams. Therefore, in this paper, a probabilistic prediction approach of the shear strength of reinforced concrete deep beams is proposed based on the natural gradient boosting algorithm trained on a collected database. A database of 267 deep beam experiments was utilized, with 14 key parameters identified as the inputs related to the beam geometry, material properties, and reinforcement details. The proposed NGBoost model was compared to empirical formulas from design codes and other machine learning methods. The results showed that the NGBoost model achieved higher accuracy in mean shear strength prediction, with an R² of 0.9045 and an RMSE of 38.8 kN, outperforming existing formulas by over 50%. Additionally, the NGBoost model provided probabilistic predictions of shear strength as probability density functions, enabling reliable confidence intervals. This demonstrated the capability of the data-driven NGBoost approach for robust shear strength evaluation of RC deep beams. Overall, the results illustrated that the proposed probabilistic prediction approach dramatically surpassed the current formulas adopted in design codes and machine learning models in both prediction accuracy and robustness. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Illustration of a typical RC deep beam: (<b>a</b>) A typical deep beam. (<b>b</b>) Key impact factors for a deep beam.</p> Full article ">Figure 2
<p>Marginal distributions of the input factors in the deep beam shear database: (<b>a</b>) <span class="html-italic">L</span>. (<b>b</b>) <span class="html-italic">a</span>. (<b>c</b>) <span class="html-italic">h</span>. (<b>d</b>) <math display="inline"><semantics> <msub> <mi>h</mi> <mn>0</mn> </msub> </semantics></math>. (<b>e</b>) <span class="html-italic">b</span>. (<b>f</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mi>c</mi> </msub> </semantics></math>. (<b>g</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mi>y</mi> </msub> </semantics></math>. (<b>h</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mrow> <mi>y</mi> <mi>h</mi> </mrow> </msub> </semantics></math>. (<b>i</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mrow> <mi>y</mi> <mi>v</mi> </mrow> </msub> </semantics></math>. (<b>j</b>) <math display="inline"><semantics> <msub> <mi>ρ</mi> <mi>l</mi> </msub> </semantics></math>. (<b>k</b>) <math display="inline"><semantics> <msub> <mi>ρ</mi> <mrow> <mi>w</mi> <mi>h</mi> </mrow> </msub> </semantics></math>. (<b>l</b>) <math display="inline"><semantics> <msub> <mi>ρ</mi> <mrow> <mi>w</mi> <mi>v</mi> </mrow> </msub> </semantics></math>. (<b>m</b>) <math display="inline"><semantics> <msub> <mi>S</mi> <mrow> <mi>w</mi> <mi>h</mi> </mrow> </msub> </semantics></math>. (<b>n</b>) <math display="inline"><semantics> <msub> <mi>S</mi> <mrow> <mi>w</mi> <mi>v</mi> </mrow> </msub> </semantics></math>.</p> Full article ">Figure 2 Cont.
<p>Marginal distributions of the input factors in the deep beam shear database: (<b>a</b>) <span class="html-italic">L</span>. (<b>b</b>) <span class="html-italic">a</span>. (<b>c</b>) <span class="html-italic">h</span>. (<b>d</b>) <math display="inline"><semantics> <msub> <mi>h</mi> <mn>0</mn> </msub> </semantics></math>. (<b>e</b>) <span class="html-italic">b</span>. (<b>f</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mi>c</mi> </msub> </semantics></math>. (<b>g</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mi>y</mi> </msub> </semantics></math>. (<b>h</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mrow> <mi>y</mi> <mi>h</mi> </mrow> </msub> </semantics></math>. (<b>i</b>) <math display="inline"><semantics> <msub> <mi>f</mi> <mrow> <mi>y</mi> <mi>v</mi> </mrow> </msub> </semantics></math>. (<b>j</b>) <math display="inline"><semantics> <msub> <mi>ρ</mi> <mi>l</mi> </msub> </semantics></math>. (<b>k</b>) <math display="inline"><semantics> <msub> <mi>ρ</mi> <mrow> <mi>w</mi> <mi>h</mi> </mrow> </msub> </semantics></math>. (<b>l</b>) <math display="inline"><semantics> <msub> <mi>ρ</mi> <mrow> <mi>w</mi> <mi>v</mi> </mrow> </msub> </semantics></math>. (<b>m</b>) <math display="inline"><semantics> <msub> <mi>S</mi> <mrow> <mi>w</mi> <mi>h</mi> </mrow> </msub> </semantics></math>. (<b>n</b>) <math display="inline"><semantics> <msub> <mi>S</mi> <mrow> <mi>w</mi> <mi>v</mi> </mrow> </msub> </semantics></math>.</p> Full article ">Figure 3
<p>Bayesian optimization process.</p> Full article ">Figure 4
<p>Implementation workflow for NGBoost.</p> Full article ">Figure 5
<p>Optimization process of NGBoost’s hyper-parameters: (<b>a</b>) Optimization history. (<b>b</b>) Optimization contour of two basic hyper-parameters.</p> Full article ">Figure 6
<p>Scalar prediction performance of models on deep beam shear database: (<b>a</b>) Linear regression. (<b>b</b>) Support vector regression. (<b>c</b>) Neural network. (<b>d</b>) Random forest. (<b>e</b>) XGBoost. (<b>f</b>) NGBoost.</p> Full article ">Figure 7
<p>Scalar prediction performance of empirical formulas on testing database of deep beam shear experiment: (<b>a</b>) GB. (<b>b</b>) ACI. (<b>c</b>) CSA. (<b>d</b>) EU. (<b>e</b>) NGBoost.</p> Full article ">Figure 8
<p>Probabilistic prediction result of NGBoost on deep beam’s shear capacity.</p> Full article ">Figure 9
<p>Probabilistic prediction of shear capacity on four deep beam samples: (<b>a</b>) Sample 1. (<b>b</b>) Sample 2. (<b>c</b>) Sample 3. (<b>d</b>) Sample 4.</p> Full article ">

17 pages, 5765 KiB  

Open AccessArticle

LCA-YOLOv8-Seg: An Improved Lightweight YOLOv8-Seg for Real-Time Pixel-Level Crack Detection of Dams and Bridges

by Yang Wu, Qingbang Han, Qilin Jin, Jian Li and Yujing Zhang

Appl. Sci. 2023, 13(19), 10583; https://doi.org/10.3390/app131910583 - 22 Sep 2023

Cited by 17 | Viewed by 5064

Abstract

Remotely operated vehicles (ROVs) and unmanned aerial vehicles (UAVs) provide a solution for dam and bridges structural health information acquisition, but problems like effective damage-related information extraction also occur. Vision-based crack detection methods can replace traditional manual inspection and achieve fast and accurate [...] Read more.

Remotely operated vehicles (ROVs) and unmanned aerial vehicles (UAVs) provide a solution for dam and bridges structural health information acquisition, but problems like effective damage-related information extraction also occur. Vision-based crack detection methods can replace traditional manual inspection and achieve fast and accurate crack detection. This paper thereby proposes a lightweight, real-time, pixel-level crack detection method based on an improved instance segmentation model. A lightweight backbone and a novel efficient prototype mask branch are proposed to decrease the complexity of the model and maintain the accuracy of the model. The proposed method attains an accuracy of 0.945 at 129 frames per second (FPS). Moreover, our model has smaller volume, lower computational requirements and is suitable for low-performance devices. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>A remotely operated vehicle (ROV).</p> Full article ">Figure 2
<p>The framework of the proposed method.</p> Full article ">Figure 3
<p>The structure of LCA-YOLOv8n-seg.</p> Full article ">Figure 4
<p>The structure of DWConv block.</p> Full article ">Figure 5
<p>The implementation of standard convolution.</p> Full article ">Figure 6
<p>The implementation of depthwise separable convolution.</p> Full article ">Figure 7
<p>The specific structure of standard Proto and ProtoC1. (<b>a</b>) The specific structure of standard Proto; (<b>b</b>) the specific structure of ProtoC1.</p> Full article ">Figure 8
<p>The flowchart of the TL method.</p> Full article ">Figure 9
<p>Examples of underwater dam concrete crack.</p> Full article ">Figure 10
<p>Examples of crack images.</p> Full article ">Figure 11
<p>Flowchart of the labelling process.</p> Full article ">Figure 12
<p>Example images of data augmentation.</p> Full article ">Figure 13
<p>Curves for the precision metric and <math display="inline"><semantics> <mrow> <mi>m</mi> <mi>A</mi> <msub> <mi>P</mi> <mrow> <mn>0.5</mn> </mrow> </msub> </mrow> </semantics></math> metric of the training. (<b>a</b>) The curve of box precision and <math display="inline"><semantics> <mrow> <mi>m</mi> <mi>A</mi> <msub> <mi>P</mi> <mrow> <mn>0.5</mn> </mrow> </msub> </mrow> </semantics></math>; (<b>b</b>) the curve of mask precision and <math display="inline"><semantics> <mrow> <mi>m</mi> <mi>A</mi> <msub> <mi>P</mi> <mrow> <mn>0.5</mn> </mrow> </msub> </mrow> </semantics></math>.</p> Full article ">Figure 14
<p>Crack detection performance of our model. (<b>a</b>–<b>d</b>) are crack pictures with different background and different crack size.</p> Full article ">Figure 15
<p>Comparison of crack detection results of different algorithms. (<b>a</b>–<b>c</b>) are different crack pictures.</p> Full article ">Figure 16
<p>Comparison of crack detection mask of YOLOv8n-seg model using different Proto modules. (<b>a</b>,<b>b</b>) are crack pictures with different crack shape.</p> Full article ">

14 pages, 5420 KiB  

Open AccessArticle

Probabilistic Seismic Hazard Analysis of a Back Propagation Neural Network Predicting the Peak Ground Acceleration

by Xin Guo, Hongnan Li, Hao Zhang, Qi Wang and Jiran Xu

Appl. Sci. 2023, 13(17), 9790; https://doi.org/10.3390/app13179790 - 30 Aug 2023

Viewed by 1498

Abstract

Probabilistic seismic hazard analysis (PSHA) has been recognized as a reasonable method for quantifying seismic threats. Traditionally, this method ignores the effect of the focal depth, in which the ground motion prediction equations (GMPEs) are applied to estimate the probability distribution associated with [...] Read more.

Probabilistic seismic hazard analysis (PSHA) has been recognized as a reasonable method for quantifying seismic threats. Traditionally, this method ignores the effect of the focal depth, in which the ground motion prediction equations (GMPEs) are applied to estimate the probability distribution associated with the possible motion levels induced by the site earthquakes, but it is limited by the unclear geological conditions, which makes it difficult to provide a uniform equation, and these equations cannot express the non-linear relationship under geological conditions. Hence, this paper proposed a method to consider the seismic focal depth for the PSHA with the example of California and used a back propagation neural network (BPNN) to predict the peak ground acceleration (PGA) instead of the GMPEs. Firstly, the measured PGA and unknown PGA seismic data applicable to this method were collected separately. Secondly, the unknown PGA data were supplemented by applying the BPNN based on the measured PGA data. Lastly, based on the full-probability equation, PSHA considering the focal depth was completed and compared with the current California seismic zoning results. The results showed that using the BPNN in the PSHA can ensure computational accuracy and universality, making it more suitable for regions with unclear geological structures and providing the possibility of adding other parameters to be considered for the influence of the PSHA. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Steps underlying PSHA.</p> Full article ">Figure 2
<p>California strong motion database map.</p> Full article ">Figure 3
<p>Known PGA data.</p> Full article ">Figure 4
<p>The California earthquake database map (<a href="https://earthquake.usgs.gov/earthquakes/map" target="_blank">https://earthquake.usgs.gov/earthquakes/map</a> (accessed on 30 August 2022)).</p> Full article ">Figure 5
<p>Histogram of magnitude frequency distribution.</p> Full article ">Figure 6
<p>The BPNN.</p> Full article ">Figure 7
<p>Regression results.</p> Full article ">Figure 8
<p>Cumulative distribution of the PGA. (<b>a</b>) Sample A. (<b>b</b>) Sample B.</p> Full article ">Figure 9
<p>Logic tree used for PSHA calculations.</p> Full article ">Figure 10
<p>Hazard curves.</p> Full article ">

22 pages, 7303 KiB  

Open AccessArticle

Study on the Interaction Behaviors Identification of Construction Workers Based on ST-GCN and YOLO

by Peilin Li, Fan Wu, Shuhua Xue and Liangjie Guo

Sensors 2023, 23(14), 6318; https://doi.org/10.3390/s23146318 - 11 Jul 2023

Cited by 7 | Viewed by 2810

Abstract

The construction industry is accident-prone, and unsafe behaviors of construction workers have been identified as a leading cause of accidents. One important countermeasure to prevent accidents is monitoring and managing those unsafe behaviors. The most popular way of detecting and identifying workers’ unsafe [...] Read more.

The construction industry is accident-prone, and unsafe behaviors of construction workers have been identified as a leading cause of accidents. One important countermeasure to prevent accidents is monitoring and managing those unsafe behaviors. The most popular way of detecting and identifying workers’ unsafe behaviors is the computer vision-based intelligent monitoring system. However, most of the existing research or products focused only on the workers’ behaviors (i.e., motions) recognition, limited studies considered the interaction between man-machine, man-material or man-environments. Those interactions are very important for judging whether the workers’ behaviors are safe or not, from the standpoint of safety management. This study aims to develop a new method of identifying construction workers’ unsafe behaviors, i.e., unsafe interaction between man-machine/material, based on ST-GCN (Spatial Temporal Graph Convolutional Networks) and YOLO (You Only Look Once), which could provide more direct and valuable information for safety management. In this study, two trained YOLO-based models were, respectively, used to detect safety signs in the workplace, and objects that interacted with workers. Then, an ST-GCN model was trained to detect and identify workers’ behaviors. Lastly, a decision algorithm was developed considering interactions between man-machine/material, based on YOLO and ST-GCN results. Results show good performance of the developed method, compared to only using ST-GCN, the accuracy was significantly improved from 51.79% to 85.71%, 61.61% to 99.11%, and 58.04% to 100.00%, respectively, in the identification of the following three kinds of behaviors, throwing (throwing hammer, throwing bottle), operating (turning on switch, putting bottle), and crossing (crossing railing and crossing obstacle). The findings of the study have some practical implications for safety management, especially workers’ behavior monitoring and management. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Representations of the selected behaviors.</p> Full article ">Figure 2
<p>OpenPose COCO model (<b>A</b>) and sequence of skeleton keypoints (<b>B</b>).</p> Full article ">Figure 3
<p>ST-GCN network structure.</p> Full article ">Figure 4
<p>YOLO network structure.</p> Full article ">Figure 5
<p>Safety signs recognized in this study.</p> Full article ">Figure 6
<p>Experimental apparatus and settings.</p> Full article ">Figure 7
<p>Confusion matrix of Type I behaviors identification only based on ST-GCN (TH: Throwing Hammer, TB: Throwing Bottle, TS: Turing on Switch, PB: Putting Bottle).</p> Full article ">Figure 8
<p>Confusion matrix of Type II behaviors identification only based on ST-GCN (CR: Crossing Railing, CO: Crossing Obstacle).</p> Full article ">Figure 9
<p>Confusion matrix of Type I behaviors identification based on YOLO-ST-GCN (TH: Throwing Hammer, TB: Throwing Bottle, TS: Turing on Switch, and PB: Putting Bottle).</p> Full article ">Figure 10
<p>Confusion matrix of Type II behaviors identification based on YOLO-ST-GCN (CR: Crossing Railing, CO: Crossing Obstacle).</p> Full article ">Figure 11
<p>Confusion matrix of behaviors risk evaluation considering safety signs identification (UB: unsafe behavior and SB: safe behavior). (<b>a</b>) No Throwing. (<b>b</b>) No Operating. (<b>c</b>) No Crossing.</p> Full article ">Figure 12
<p>Confusion matrix of behaviors identification based on YOLO-ST-GCN based on YOLO-NAS: (<b>A</b>) Type I behaviors, (<b>B</b>) Type II behaviors. (TH: Throwing Hammer, TB: Throwing Bottle, TS: Turing on Switch, and PB: Putting Bottle).</p> Full article ">Figure 13
<p>Behaviors of throwing hammer (<b>A</b>) and hammering nail (<b>B</b>).</p> Full article ">Figure 14
<p>Confusion matrix of throwing hammer and hammering nail.</p> Full article ">

18 pages, 7420 KiB  

Open AccessArticle

Research on a Ship Deflection Anti-Collision Method Based on a Water-Jet Interference Flow Field

by Kui Yu, Hongming Wang, Xianqing Liu and Bingli Peng

Appl. Sci. 2023, 13(13), 7354; https://doi.org/10.3390/app13137354 - 21 Jun 2023

Cited by 1 | Viewed by 1378

Abstract

Currently, water jets are mainly used in the fields of mechanical processing and mining collection. This paper creatively introduces them to the field of safety assurance for inland navigation. Compared with the traditional bridge anti-striking methods such as intelligent early warning and passive [...] Read more.

Currently, water jets are mainly used in the fields of mechanical processing and mining collection. This paper creatively introduces them to the field of safety assurance for inland navigation. Compared with the traditional bridge anti-striking methods such as intelligent early warning and passive anti-striking, this method can form an “interference zone” by changing the water flow conditions in the local bridge water areas, causing the yawing moment of the yaw ship to change, thereby causing the ship’s course to change, and thus guiding the ship to move away from the bridge pier to realize active anti-striking of the ship. In this paper, a combination of generalized model testing and numerical simulation was used to study the effects of different nozzle pressures and different ship pier distances of the water-jet generator on the trajectory and drift angle of the stalled ship. The results showed that the numerical simulation was in good agreement with the model test results. Within the interference zone, the distance between the ship and the pier increased rapidly after the action of the disturbance zone to 9.1, 5.8, and 6.2 times the ship’s width, respectively, reaching a safe distance. During the process of being affected by the interference zone, the maximum drift angle of the yaw ship was less than 20°, the course of the ship was generally stable, and the drift angle comparison error was a maximum of 10.6%, a minimum of 3.5%, and an average error of 6.7%. A negative peak and a positive peak of four times the absolute value of the negative peak occurred in the yaw-moment ephemeral curve during the ship’s passage through the interference area. The method had a notable effect on the anti-striking of stalled ships and two invention patents applied for in the course of research. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Scene of the ship striking accident at the Fengkai Bridge.</p> Full article ">Figure 2
<p>Working principle of the water-jet generator.</p> Full article ">Figure 3
<p>Regional division.</p> Full article ">Figure 4
<p>Elevation of the bridge.</p> Full article ">Figure 5
<p>Bridge piers scale.</p> Full article ">Figure 6
<p>Test ship model.</p> Full article ">Figure 7
<p>Model of bridge piers and anti-striking devices: (<b>a</b>) Model drawing of the anti-striking device; (<b>b</b>) Bridge pier model drawing.</p> Full article ">Figure 8
<p>The overall structure of the water-jet generator.</p> Full article ">Figure 9
<p>Generalized model test layout.</p> Full article ">Figure 10
<p>Calculate domain layout.</p> Full article ">Figure 11
<p>Grid of the test area.</p> Full article ">Figure 12
<p>Comparison of ship deflection effects: (<b>a</b>) Comparison of the deflection effects in moment one; (<b>b</b>) Comparison of the deflection effects in moment two; (<b>c</b>) Comparison of the deflection effects in moment three.</p> Full article ">Figure 13
<p>Pressure of 0.005 MPa test tracks line.</p> Full article ">Figure 14
<p>Pressure of 0.010 MPa test tracks line.</p> Full article ">Figure 15
<p>Pressure of 0.015 MPa test tracks line.</p> Full article ">Figure 16
<p>Comparison of T1 and T10 tracks line.</p> Full article ">Figure 17
<p>Comparison of T2 and T11 tracks line.</p> Full article ">Figure 18
<p>Comparison of T3 and T12 tracks line.</p> Full article ">Figure 19
<p>Model test track line maximum drift angle.</p> Full article ">Figure 20
<p>Comparison of the maximum drift angles.</p> Full article ">Figure 21
<p>Comparison of T10, T11, and T12 ship yaw moment.</p> Full article ">Figure 22
<p>Comparison of T13, T14, and T15 ship yaw moment.</p> Full article ">Figure 23
<p>Comparison of T16, T17, and T18 ship yaw moment.</p> Full article ">Figure 24
<p>Example 1: Four stages in the process of moving from a dangerous yawing course to a safe course when a ship is subject to the action of the interference zone. (<b>a</b>) Stage 1; (<b>b</b>) Stage 2; (<b>c</b>) Stage 3; (<b>d</b>) Stage 4.</p> Full article ">Figure 25
<p>Example 2: Four stages in the process of moving from a dangerous yawing course to a safe course when a ship is subject to the action of the interference zone. (<b>a</b>) Stage 1; (<b>b</b>) Stage 2; (<b>c</b>) Stage 3; (<b>d</b>) Stage 4.</p> Full article ">Figure 26
<p>Example 3: Four stages in the process of moving from a dangerous yawing course to a safe course when a ship is subject to the action of the interference zone. (<b>a</b>) Stage 1; (<b>b</b>) Stage 2; (<b>c</b>) Stage 3; (<b>d</b>) Stage 4.</p> Full article ">

30 pages, 7933 KiB  

Open AccessArticle

Distributed Compressive Sensing for Wireless Signal Transmission in Structural Health Monitoring: An Adaptive Hierarchical Bayesian Model-Based Approach

by Zhiwen Wang, Shouwang Sun, Yiwei Li, Zixiang Yue and Youliang Ding

Sensors 2023, 23(12), 5661; https://doi.org/10.3390/s23125661 - 17 Jun 2023

Cited by 2 | Viewed by 1691

Abstract

Signal transmission plays an important role in the daily operation of structural health monitoring (SHM) systems. In wireless sensor networks, transmission loss often occurs and threatens reliable data delivery. The massive amount of data monitoring also leads to a high signal transmission and [...] Read more.

Signal transmission plays an important role in the daily operation of structural health monitoring (SHM) systems. In wireless sensor networks, transmission loss often occurs and threatens reliable data delivery. The massive amount of data monitoring also leads to a high signal transmission and storage cost throughout the system’s service life. Compressive Sensing (CS) provides a novel perspective on alleviating these problems. Based on the sparsity of vibration signals in the frequency domain, CS can reconstruct a nearly complete signal from just a few measurements. This can improve the robustness of data loss while facilitating data compression to reduce transmission demands. Extended from CS methods, distributed compressive sensing (DCS) can exploit the correlation across multiple measurement vectors (MMV) to jointly recover the multi-channel signals with similar sparse patterns, which can effectively enhance the reconstruction quality. In this paper, a comprehensive DCS framework for wireless signal transmission in SHM is constructed, incorporating the process of data compression and transmission loss together. Unlike the basic DCS formulation, the proposed framework not only activates the inter-correlation among channels but also provides flexibility and independence to single-channel transmission. To promote signal sparsity, a hierarchical Bayesian model using Laplace priors is built and further improved as the fast iterative DCS-Laplace algorithm for large-scale reconstruction tasks. Vibration signals (e.g., dynamic displacement and accelerations) acquired from real-life SHM systems are used to simulate the whole process of wireless transmission and test the algorithm’s performance. The results demonstrate that (1) DCS-Laplace is an adaptative algorithm that can actively adapt to signals with various sparsity by adjusting the penalty term to achieve optimal performance; (2) compared with CS methods, DCS methods can effectively improve the reconstruction quality of multi-channel signals; (3) the Laplace method has advantages over the OMP method in terms of reconstruction performance and applicability, which is a better choice in SHM wireless signal transmission. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Workflow of the proposed DCS framework for wireless signal transmission in SHM.</p> Full article ">Figure 2
<p>Graphical model of the Bayesian CS formulation using Laplace Priors.</p> Full article ">Figure 3
<p>Hierarchical Bayesian representation of the DCS-Laplace.</p> Full article ">Figure 4
<p>Lieshihe Highway Bridge: (<b>a</b>,<b>b</b>) the actual view; (<b>c</b>) site installation of sensor ZK4W; (<b>d</b>) sensor placement on the box girder.</p> Full article ">Figure 5
<p>Dynamic displacement signals with 5 channels under vehicle events in the time and frequency domain, respectively: (<b>a</b>) the measured signals within 100 s; (<b>b</b>) the Fourier spectrum of the measured signals.</p> Full article ">Figure 6
<p>Received measurement vectors under different data loss patterns in Case 1: (<b>a</b>) uniform random loss, Channel 1 and 2 with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> <mo>,</mo> <mo> </mo> <mi>L</mi> <mi>R</mi> <mo>=</mo> <mn>20</mn> <mo>%</mo> </mrow> </semantics></math>; (<b>b</b>) non-uniform random loss, Channel 1 with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> <mo>,</mo> <mo> </mo> <mi>L</mi> <mi>R</mi> <mo>=</mo> <mn>20</mn> <mo>%</mo> </mrow> </semantics></math> and Channel 2 with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> <mo>,</mo> <mo> </mo> <mi>L</mi> <mi>R</mi> <mo>=</mo> <mn>40</mn> <mo>%</mo> </mrow> </semantics></math>.</p> Full article ">Figure 6 Cont.
<p>Received measurement vectors under different data loss patterns in Case 1: (<b>a</b>) uniform random loss, Channel 1 and 2 with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> <mo>,</mo> <mo> </mo> <mi>L</mi> <mi>R</mi> <mo>=</mo> <mn>20</mn> <mo>%</mo> </mrow> </semantics></math>; (<b>b</b>) non-uniform random loss, Channel 1 with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> <mo>,</mo> <mo> </mo> <mi>L</mi> <mi>R</mi> <mo>=</mo> <mn>20</mn> <mo>%</mo> </mrow> </semantics></math> and Channel 2 with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> <mo>,</mo> <mo> </mo> <mi>L</mi> <mi>R</mi> <mo>=</mo> <mn>40</mn> <mo>%</mo> </mrow> </semantics></math>.</p> Full article ">Figure 7
<p>The SNR values of the recovered signal from channels 1–3 under different CR and LR with DCS-Laplace: (<b>a</b>) channels 1–3 from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>1</mn> </mrow> </semantics></math>; (<b>b</b>) channels 1–3 from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> </mrow> </semantics></math>; (<b>c</b>) channels 1–3 from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> </mrow> </semantics></math>.</p> Full article ">Figure 8
<p>Reconstructed signals of channel 1 with AE when SNR values are close to 40 and 20, respectively: (<b>a</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>39.8814</mn> </mrow> </semantics></math>; (<b>b</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>20.3442</mn> </mrow> </semantics></math>.</p> Full article ">Figure 8 Cont.
<p>Reconstructed signals of channel 1 with AE when SNR values are close to 40 and 20, respectively: (<b>a</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>39.8814</mn> </mrow> </semantics></math>; (<b>b</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>20.3442</mn> </mrow> </semantics></math>.</p> Full article ">Figure 9
<p>The SNR values of the recovered signal from channel 1 under different CR and LR with CS and DCS algorithms: (<b>a</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>1</mn> </mrow> </semantics></math>; (<b>b</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> </mrow> </semantics></math>; (<b>c</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> </mrow> </semantics></math>.</p> Full article ">Figure 9 Cont.
<p>The SNR values of the recovered signal from channel 1 under different CR and LR with CS and DCS algorithms: (<b>a</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>1</mn> </mrow> </semantics></math>; (<b>b</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> </mrow> </semantics></math>; (<b>c</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>4</mn> </mrow> </semantics></math>.</p> Full article ">Figure 10
<p>Dashengguan High-speed Railway Bridge: (<b>a</b>) the actual view; (<b>b</b>) accelerometer deployment on the bridge.</p> Full article ">Figure 11
<p>Acceleration signals from sensor 1 under 3 different train events in the time and frequency domain, respectively: (<b>a</b>) the measured signals within 45 s; (<b>b</b>) the Fourier spectrum of train-induced accelerations.</p> Full article ">Figure 11 Cont.
<p>Acceleration signals from sensor 1 under 3 different train events in the time and frequency domain, respectively: (<b>a</b>) the measured signals within 45 s; (<b>b</b>) the Fourier spectrum of train-induced accelerations.</p> Full article ">Figure 12
<p>The SNR values of recovered signals from channels 1–3 under different CR and LR with DCS-Laplace: (<b>a</b>) channels 1–3 from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>1</mn> </mrow> </semantics></math>; (<b>b</b>) channels 1–3 from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> </mrow> </semantics></math>.</p> Full article ">Figure 13
<p>Reconstructed signals of channel 1 with AE when SNR values are close to 40 and 20, respectively: (<b>a</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>40.5403</mn> </mrow> </semantics></math>; (<b>b</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>20.0039</mn> </mrow> </semantics></math>.</p> Full article ">Figure 13 Cont.
<p>Reconstructed signals of channel 1 with AE when SNR values are close to 40 and 20, respectively: (<b>a</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>40.5403</mn> </mrow> </semantics></math>; (<b>b</b>) <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>N</mi> <mi>R</mi> <mo>=</mo> <mn>20.0039</mn> </mrow> </semantics></math>.</p> Full article ">Figure 14
<p>The SNR values of recovered signals from channel 2 under different CR and LR with CS and DCS algorithms: (<b>a</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>1</mn> </mrow> </semantics></math>; (<b>b</b>) DCS vs. CS from left to right with <math display="inline"><semantics> <mrow> <mi>C</mi> <mi>R</mi> <mo>=</mo> <mn>2</mn> </mrow> </semantics></math>.</p> Full article ">

13 pages, 3012 KiB  

Open AccessArticle

Void Detection inside Duct of Prestressed Concrete Bridges Based on Deep Support Vector Data Description

by Byoung-Doo Oh, Hyung Choi, Won-Jong Chin, Chan-Young Park and Yu-Seop Kim

Appl. Sci. 2023, 13(10), 5981; https://doi.org/10.3390/app13105981 - 12 May 2023

Cited by 1 | Viewed by 1493

Abstract

The tendon that is inserted into the duct is a crucial component of prestressed concrete (PSC) bridges and, when exposed to air, can quickly corrode, and cause structural collapse. It can interpret the signal measured by non-destructive testing (NDT) to determine the condition [...] Read more.

The tendon that is inserted into the duct is a crucial component of prestressed concrete (PSC) bridges and, when exposed to air, can quickly corrode, and cause structural collapse. It can interpret the signal measured by non-destructive testing (NDT) to determine the condition (normal or void) inside the duct. However, it requires the use of expensive NDT equipment such as ultrasonic waves or the hiring of experts. In this paper, we proposed an impact–echo (IE) method based on deep support vector data description (Deep SVDD) for economical void detection inside a duct. Because the pattern of IE changes for various reasons such as difference of specimen or bridge, supervised learning is not suitable. Deep SVDD is classified as normal and defective, which is a broad distribution as a hypersphere that encloses a multi-dimensional feature space for normal data represented by an autoencoder. Here, an autoencoder was developed based on the ELMo (embeddings from language model)-like structure to obtain an effective representation for IE. In the experiment, we evaluated the performance of the IE data measured in different specimens. Thus, our proposed model showed an accuracy of about 77.84% which is an improvement of up to about 47% compared to the supervised learning approach. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Structure of PSC girder bridge.</p> Full article ">Figure 2
<p>IE equipment used in this paper.</p> Full article ">Figure 3
<p>(<b>a</b>) Basic principle of IE; (<b>b</b>) examples of IEs that may contain information of internal defect.</p> Full article ">Figure 4
<p>Example of IE data in Specimen−1 (top: normal, bottom: defect).</p> Full article ">Figure 5
<p>IE data in different specimens show different patterns. (<b>a</b>) Normal IE data (Specimen−1); (<b>b</b>) normal IE data (Specimen−2); (<b>c</b>) void IE data (Specimen−1); (<b>d</b>) void IE data (Specimen−2).</p> Full article ">Figure 6
<p>The overall flow of our proposed model for void detection inside duct in PSC bridge.</p> Full article ">Figure 7
<p>Basic process of Deep SVDD.</p> Full article ">Figure 8
<p>(<b>a</b>) Example of Bi-LSTM’s structure; (<b>b</b>) example of ELMo’s structure.</p> Full article ">Figure 9
<p>Architecture of the autoencoder used in Deep SVDD.</p> Full article ">

19 pages, 1991 KiB  

Open AccessArticle

Improved YOLOv5-Based Lightweight Object Detection Algorithm for People with Visual Impairment to Detect Buses

by Rio Arifando, Shinji Eto and Chikamune Wada

Appl. Sci. 2023, 13(9), 5802; https://doi.org/10.3390/app13095802 - 8 May 2023

Cited by 34 | Viewed by 8035

Abstract

Object detection is crucial for individuals with visual impairment, especially when waiting for a bus. In this study, we propose a lightweight and highly accurate bus detection model based on an improved version of the YOLOv5 model. We propose integrating the GhostConv and [...] Read more.

Object detection is crucial for individuals with visual impairment, especially when waiting for a bus. In this study, we propose a lightweight and highly accurate bus detection model based on an improved version of the YOLOv5 model. We propose integrating the GhostConv and C3Ghost Modules into the YOLOv5 network to reduce the number of parameters and floating-point operations per second (FLOPs), ensuring detection accuracy while reducing the model parameters. Following that, we added the SimSPPF module to replace the SPPF in the YOLOv5 backbone for increased computational efficiency and accurate object detection capabilities. Finally, we developed a Slim scale detection model by modifying the original YOLOv5 structure in order to make the model more efficient and faster, which is critical for real-time object detection applications. According to the experimental results, the Improved-YOLOv5 outperforms the original YOLOv5 in terms of the precision, recall, and [email protected]. Further analysis of the model complexity reveals that the Improved-YOLOv5 is more efficient due to fewer FLOPS, with fewer parameters, less memory usage, and faster inference time capabilities. The proposed model is smaller and more feasible to implement in resource-constrained mobile devices and a promising option for bus detection systems. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>The YOLOv5 model structure.</p> Full article ">Figure 2
<p>Structure of the SimSPPF Module.</p> Full article ">Figure 3
<p>Schematic diagram of the Ghost module.</p> Full article ">Figure 4
<p>The structure of YOLOv5 with GhostNet. (<b>a</b>) Ghost Module. (<b>b</b>) Ghost Bottleneck with stride = 1. (<b>c</b>) Ghost Bottleneck with stride = 2. (<b>d</b>) C3Ghost.</p> Full article ">Figure 5
<p>The structure of the Improved-YOLOv5.</p> Full article ">Figure 6
<p>Example image of the dataset.</p> Full article ">Figure 7
<p>Visualization of the dataset. (<b>a</b>) Number of annotations per class. (<b>b</b>) Visualization of the location and size of each bounding box. (<b>c</b>) The statistical distribution of the bounding box position. (<b>d</b>) The statistical distribution of the bounding box sizes.</p> Full article ">Figure 8
<p>Comparison of the detection results of different models. (<b>a</b>) YOLOv5n model. (<b>b</b>) Improved-YOLOv5.</p> Full article ">Figure 9
<p>Comparison of the models’ accuracy. (<b>a</b>) Precision. (<b>b</b>) Recall. (<b>c</b>) mAP@0.5. (<b>d</b>) mAP@0.5:0.9. (<b>e</b>) Train/box_loss. (<b>f</b>) Train/obj_loss. (<b>g</b>) Val/box_loss. (<b>h</b>) Val/obj_loss.</p> Full article ">Figure 10
<p>Comparison of the models’ accuracy.</p> Full article ">Figure 11
<p>Performance comparison of different models in graphical form.</p> Full article ">

28 pages, 10901 KiB  

Open AccessArticle

Three-Dimensional Temperature Field Simulation and Analysis of a Concrete Bridge Tower Considering the Influence of Sunshine Shadow

by Shuai Zou, Jun Xiao, Jianping Xian, Yongshui Zhang and Jingfeng Zhang

Appl. Sci. 2023, 13(8), 4769; https://doi.org/10.3390/app13084769 - 10 Apr 2023

Viewed by 1791

Abstract

This paper forms a set of three-dimensional temperature field simulation methods considering the influence of sunshine shadow based on the DFLUX subroutine and FILM subroutine interface provided by the Abaqus platform to simulate the three-dimensional temperature field of concrete bridge towers and study [...] Read more.

This paper forms a set of three-dimensional temperature field simulation methods considering the influence of sunshine shadow based on the DFLUX subroutine and FILM subroutine interface provided by the Abaqus platform to simulate the three-dimensional temperature field of concrete bridge towers and study its distribution law. The results show that the method has high accuracy for shadow recognition and temperature field calculation. The maximum difference between the shadow recognition results and the theoretical calculation value was only 19.1 mm, and the maximum difference between the simulated temperature and the measured temperature was 3.3 °C. The results of analyzing the temperature field of the concrete bridge tower using this algorithm show that the temperature difference between the opposite external surface of the tower column can reach 11.6 °C, which is significantly greater than the recommended temperature difference value of 5 °C in the specifications. For the concrete bridge tower, in the thickness direction of the tower wall, the temperature change was obvious only at a range of 0.3 m from the external surface of the tower wall, and the temperature change in the remaining range was small. In addition, the temperature gradient distribution of the sunshine temperature field in the direction of wall thickness conformed to the exponential function T(x) = T₀e^−αx + C. Additionally, the data fitting results indicate that using the temperature data at a distance of 0.8 m from the external surface as the calculation parameter in the function can achieve the ideal fitting result. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Schematic diagram of heat exchange between structural surfaces and the environment. Note: In the figure, <span class="html-italic">I</span> is the heat flow density and <span class="html-italic">h</span> is the heat transfer coefficient.</p> Full article ">Figure 2
<p>The relative position relationship between the sun and the inclined plane.</p> Full article ">Figure 3
<p>Relationship between structural surface shadow occlusion and solar radiation.</p> Full article ">Figure 4
<p>Daily temperature model.</p> Full article ">Figure 5
<p>Shadow occlusion classification.</p> Full article ">Figure 6
<p>Sunshine shadow recognition technology implementation process.</p> Full article ">Figure 7
<p>The schematic diagram of the projection mode of the node to be detected and the structural surface mesh.</p> Full article ">Figure 8
<p>The principle of projection point inclusion detection.</p> Full article ">Figure 9
<p>The triangular barycentric coordinate method.</p> Full article ">Figure 10
<p>Three-dimensional sunshine temperature field analysis process.</p> Full article ">Figure 11
<p>Sunshine shadow display operation process.</p> Full article ">Figure 12
<p>Section measuring point arrangement and size diagram.</p> Full article ">Figure 13
<p>The finite element model and mesh division diagram.</p> Full article ">Figure 14
<p>The shadow occlusion relationship diagram of the box-girder web.</p> Full article ">Figure 15
<p>Comparison of Shadow Occlusion Results.</p> Full article ">Figure 16
<p>Temperature change curve of floor and roof.</p> Full article ">Figure 17
<p>Temperature change curve of the web on the sunward side and shaded side.</p> Full article ">Figure 18
<p>Schematic diagram of the bridge tower structure. (<b>a</b>) The elevation of the bridge tower; (<b>b</b>) Section A; (<b>c</b>) Section B.</p> Full article ">Figure 19
<p>Shadow occlusion state and temperature field cloud of the bridge tower at 10:00 a.m. (<b>a</b>) The shaded state of the bridge tower surface; (<b>b</b>) the temperature field cloud map.</p> Full article ">Figure 20
<p>The temperature change of the bridge tower surface with height at 10:00 a.m.</p> Full article ">Figure 21
<p>The temperature change curve of the outer surface of the tower wall at a height of 125 m and the temperature difference curve between the opposite faces of the tower column.</p> Full article ">Figure 22
<p>The temperature change curve in the thickness direction of the east tower wall at a 125 m height in the early morning.</p> Full article ">Figure 23
<p>The temperature variation curve at night in the thickness direction of the east tower wall at a 125 m height.</p> Full article ">Figure 24
<p>The temperature variation curve in the thickness direction of the east tower wall at a 125 m height during the day.</p> Full article ">Figure 25
<p>The maximum variation of temperature at the same distance from the external surface of the tower wall at a height of 125 m.</p> Full article ">Figure 26
<p>Two-dimensional temperature field cloud of tower column section at a 125 m height.</p> Full article ">Figure 27
<p>The fitting results of discrete temperature data. (<b>a</b>) The east tower wall; (<b>b</b>) the south tower wall; (<b>c</b>) the west tower wall; (<b>d</b>) the north tower wall; (<b>e</b>) the inner east tower wall; (<b>f</b>) the inner west tower wall. Note: The temperature fitting curve 1 in the figure is the result of fitting the temperature value at 0.8 m from the external surface as the parameter <span class="html-italic">C</span> in Formula (33); the temperature fitting curve 2 is the result of fitting the temperature value of the internal surface as the parameter <span class="html-italic">C</span> in Formula (33).</p> Full article ">Figure 27 Cont.
<p>The fitting results of discrete temperature data. (<b>a</b>) The east tower wall; (<b>b</b>) the south tower wall; (<b>c</b>) the west tower wall; (<b>d</b>) the north tower wall; (<b>e</b>) the inner east tower wall; (<b>f</b>) the inner west tower wall. Note: The temperature fitting curve 1 in the figure is the result of fitting the temperature value at 0.8 m from the external surface as the parameter <span class="html-italic">C</span> in Formula (33); the temperature fitting curve 2 is the result of fitting the temperature value of the internal surface as the parameter <span class="html-italic">C</span> in Formula (33).</p> Full article ">Figure 28
<p>Temperature load loading mode. Note: <span class="html-italic">T</span>₁(<span class="html-italic">x</span>), <span class="html-italic">T</span>₂(<span class="html-italic">x</span>), <span class="html-italic">T</span>₃(<span class="html-italic">x</span>), and <span class="html-italic">T</span>₄(<span class="html-italic">x</span>) in the diagram are all the calculation relations of temperature in the thickness direction of each tower wall fitted by Formula (33).</p> Full article ">

17 pages, 3908 KiB  

Open AccessArticle

Measuring the Distance between Trees and Power Lines under Wind Loads to Assess the Heightened Potential Risk of Wildfire

by Seulbi Lee and Youngjib Ham

Remote Sens. 2023, 15(6), 1485; https://doi.org/10.3390/rs15061485 - 7 Mar 2023

Viewed by 3271

Abstract

The incidence of wildfires caused by tree contact with high-voltage power lines has become an increasingly pressing issue in the United States. To prevent such incidents, local safety councils have established minimum clearance regulations between trees and power lines. While most studies have [...] Read more.

The incidence of wildfires caused by tree contact with high-voltage power lines has become an increasingly pressing issue in the United States. To prevent such incidents, local safety councils have established minimum clearance regulations between trees and power lines. While most studies have focused on the tree encroachment around power lines during normal weather conditions, recent catastrophic fires have been caused by strong winds. To address this gap in knowledge, we investigated the critical wind speed that heightens the risk of wildfires by calculating the distance between trees and wires. To conduct this study, we used airborne LiDAR data collected from Sonoma County in northern California and analyzed the behavior of a sample tree having a height of 19.2 m under wind loads. Our analysis showed that the main factor determining tree deflection is the ratio of the tree height to the trunk diameter. We also found that, although the probability of fire ignition is typically low under normal conditions, it is likely to increase at a wind speed of approximately 40.3 m/s. In conclusion, this research demonstrates the utility of point cloud data in identifying potentially dangerous trees and reducing the risk of fires. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>The study area in Sonoma County in northern California.</p> Full article ">Figure 2
<p>Overview of the proposed method.</p> Full article ">Figure 3
<p>(<b>a</b>) Results of the first preprocessing: point cloud image colored by class. (<b>b</b>) Results of the second preprocessing: point cloud image colored by elevation.</p> Full article ">Figure 4
<p>(<b>a</b>) Google Street View image of the trees and the wires in the study area. (<b>b</b>) Raster image of the trees colored by elevation. (<b>c</b>) Three-dimensional view of the point cloud data colored by individual tree. (<b>d</b>) Individual segmented tree colored by crown projection area.</p> Full article ">Figure 5
<p>Three-dimensional geometric shape and parameters of the sample tree.</p> Full article ">Figure 6
<p>The relationships between tree deflection and the wind speed.</p> Full article ">Figure 7
<p>The relationships between tree deflection and the height of each point of the tree at a wind speed of 30 m/s.</p> Full article ">Figure 8
<p>The relationships between the deflected angle of the tree and the minimum distance between trees and wires.</p> Full article ">Figure 9
<p>Schematic diagram of how the tree is deflected toward the wire by wind loading.</p> Full article ">Figure 10
<p>The comparison results for the ratio between the maximum deflection and tree height under varying wind speeds.</p> Full article ">

20 pages, 5296 KiB  

Open AccessArticle

AMSCN: A Novel Dual-Task Model for Automatic Modulation Classification and Specific Emitter Identification

by Shanchuan Ying, Sai Huang, Shuo Chang, Jiashuo He and Zhiyong Feng

Sensors 2023, 23(5), 2476; https://doi.org/10.3390/s23052476 - 23 Feb 2023

Cited by 6 | Viewed by 2094

Abstract

Specific emitter identification (SEI) and automatic modulation classification (AMC) are generally two separate tasks in the field of radio monitoring. Both tasks have similarities in terms of their application scenarios, signal modeling, feature engineering, and classifier design. It is feasible and promising to [...] Read more.

Specific emitter identification (SEI) and automatic modulation classification (AMC) are generally two separate tasks in the field of radio monitoring. Both tasks have similarities in terms of their application scenarios, signal modeling, feature engineering, and classifier design. It is feasible and promising to integrate these two tasks, with the benefit of reducing the overall computational complexity and improving the classification accuracy of each task. In this paper, we propose a dual-task neural network named AMSCN that simultaneously classifies the modulation and the transmitter of the received signal. In the AMSCN, we first use a combination of DenseNet and Transformer as the backbone network to extract the distinguishable features; then, we design a mask-based dual-head classifier (MDHC) to reinforce the joint learning of the two tasks. To train the AMSCN, a multitask cross-entropy loss is proposed, which is the sum of the cross-entropy loss of the AMC and the cross-entropy loss of the SEI. Experimental results show that our method achieves performance gains for the SEI task with the aid of additional information from the AMC task. Compared with the traditional single-task model, our classification accuracy of the AMC is generally consistent with the state-of-the-art performance, while the classification accuracy of the SEI is improved from 52.2% to 54.7%, which demonstrates the effectiveness of the AMSCN. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>The modulation and transmitter joint-identification framework.</p> Full article ">Figure 2
<p>The process of digital modulation and up-conversion.</p> Full article ">Figure 3
<p>Offline training process of the proposed AMSCN.</p> Full article ">Figure 4
<p>Schematic diagram of forward and backward propagation of AMSCN.</p> Full article ">Figure 5
<p>Schematic diagram of the Transformer part of the AMSCN. Before the features are fed into the first transformer block, preprocessing is required, i.e., adding the classification token and position embedding.</p> Full article ">Figure 6
<p>The structure of the self-attention part a transformer block. (<b>a</b>) Details of the multi-head attention structure. (<b>b</b>) Scaled dot-product attention structure.</p> Full article ">Figure 7
<p>The calculation process of the SEI feature vectors. The mask matrix is generated from the AMC feature vector. The subscript of the largest element of the AMC feature vector determines which row of the mask matrix is 1. The colored part of this figure indicates the maximum value in the vector and the row of the matrix that takes 1.</p> Full article ">Figure 8
<p>Adding a mask between the two task headers improves the performance on both tasks. (<b>a</b>) Comparison of the mask effects in the AMC tasks. (<b>b</b>) Comparison of the mask effects in the SEI tasks.</p> Full article ">Figure 9
<p>For the AMSCN models, multitask training is more effective than single-task training. (<b>a</b>) Comparison of the fusion method and the single-head method in the AMC tasks. (<b>b</b>) Comparison of the fusion method and the single-head method in the SEI tasks.</p> Full article ">Figure 10
<p>Performance comparison between the AMSCN and other popular models. (<b>a</b>) Model performance curves in the AMC tasks. (<b>b</b>) Model performance curves in the SEI tasks.</p> Full article ">

16 pages, 15011 KiB  

Open AccessArticle

Research on Tower Mechanical Fault Classification Method Based on Multiclass Central Segmentation Hyperplane Support Vector Machine Improvement Algorithm

by Shunjie Han, Heran Wang, Xueyan Hu, Huan Yang and Hanye Wu

Appl. Sci. 2023, 13(3), 1331; https://doi.org/10.3390/app13031331 - 19 Jan 2023

Cited by 2 | Viewed by 1333

Abstract

In this paper, a classification recognition algorithm for tower mechanical faults is proposed, and a multiclass central segmentation hyperplane support vector machine (CSH-SVM) is proposed to improve the existing multiclass support vector machine for problems in which a certain sample satisfies multiple hyperplanes [...] Read more.

In this paper, a classification recognition algorithm for tower mechanical faults is proposed, and a multiclass central segmentation hyperplane support vector machine (CSH-SVM) is proposed to improve the existing multiclass support vector machine for problems in which a certain sample satisfies multiple hyperplanes at the same time. The tilt angle change and wind direction data were extracted using the tilt sensors and anemometers attached to the tower, and the temperature and humidity sensors, as well as real-time rainfall and water accumulation information, were combined to construct a sample of the original dataset during the operation of the tower. The unbalanced samples were improved using the synthetic minority oversampling technique (SMOTE) algorithm to construct a balanced dataset suitable for machine learning and improve the prediction accuracy of machine learning. At the same time, the support vector machine hyperplane under the one-vs-all classification principle was additionally computed, and the new hyperplane was computed via the existing hyperplane not only to solve the classification problem of the transition area under the one-vs-all classification so that the samples located in this area no longer meet two hyperplane equations at the same time, but also to reduce the probability of incorrect classification to a certain extent. Through verification, CSH-SVM can classify 15 out of 77 misclassified samples into the correct category with slightly higher computational power than the traditional one-vs-all classification SVM, which can improve the classification prediction accuracy for unbalanced tower mechanical failure datasets and make an accurate judgment on the current state of the tower through the tower data as to when the tower may generate mechanical failure, thus reducing economic loss and personal safety threats. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>The representation of an unbalanced 2D dataset in 2D coordinates.</p> Full article ">Figure 2
<p>(<b>a</b>) SMOTE algorithm random selection process; (<b>b</b>) SMOTE algorithm random generation process.</p> Full article ">Figure 3
<p>(<b>a</b>) One-vs-one method of classifying hyperplanes; (<b>b</b>) one-vs-all method of classifying hyperplanes.</p> Full article ">Figure 4
<p>Equivalent force analysis results.</p> Full article ">Figure 5
<p>(<b>a</b>) Comparison of echocardiogram dataset before applying SMOTE algorithm; (<b>b</b>) comparison of echocardiogram dataset after applying SMOTE algorithm.</p> Full article ">Figure 6
<p>Echocardiogram raw dataset validation accuracy.</p> Full article ">Figure 7
<p>SMOTE echocardiogram dataset validated accurately.</p> Full article ">Figure 8
<p>(<b>a</b>) Relationship between samples in the transition area and the corresponding category hyperplane; (<b>b</b>) Relationship between the improved hyperplane and the corresponding category samples.</p> Full article ">Figure 9
<p>Schemes follow the same formatting.</p> Full article ">Figure 10
<p>Calculated results of four types of samples under the original hyperplane.</p> Full article ">Figure 11
<p>Calculation results of four types of samples under improved hyperplane.</p> Full article ">Figure 12
<p>Classification results of misclassified samples in one-vs-all classification SVM.</p> Full article ">Figure 13
<p>Classification results of improved CSH-SVM algorithm.</p> Full article ">Figure 14
<p>SMOTE and CSH-SVM improved algorithm classification results.</p> Full article ">

18 pages, 19254 KiB  

Open AccessArticle

Multiple Defects Inspection of Dam Spillway Surface Using Deep Learning and 3D Reconstruction Techniques

by Kunlong Hong, Hongguang Wang, Bingbing Yuan and Tianfu Wang

Buildings 2023, 13(2), 285; https://doi.org/10.3390/buildings13020285 - 18 Jan 2023

Cited by 6 | Viewed by 2238

Abstract

After a lengthy period of scouring, the reinforced concrete surface of the dam spillway (i.e., drift spillways and flood discharge spillways) will suffer from deterioration and damage. Regular manual inspection is time-consuming and dangerous. This paper presents a robotic solution to detect automatically, [...] Read more.

After a lengthy period of scouring, the reinforced concrete surface of the dam spillway (i.e., drift spillways and flood discharge spillways) will suffer from deterioration and damage. Regular manual inspection is time-consuming and dangerous. This paper presents a robotic solution to detect automatically, count defect instance numbers, and reconstruct the surface of dam spillways by incorporating the deep learning method with a visual 3D reconstruction method. The lack of a real dam defect dataset and incomplete registration of minor defects on the 3D mesh model in fusion step are two challenges addressed in the paper. We created a multi-class semantic segmentation dataset of 1711 images (with resolutions of 848 × 480 and 1280 × 720 pixels) acquired by a wall-climbing robot, including cracks, erosion, spots, patched areas, and power safety cable. Then, the architecture of the U-net is modified with pixel-adaptive convolution (PAC) and conditional random field (CRF) to segment different scales of defects, trained, validated, and tested using this dataset. The reconstruction and recovery of minor defect instances in the flow surface and sidewall are facilitated using a keyframe back-projection method. By generating an instance adjacency matrix within the class, the intersection over union (IoU) of 3D voxels is calculated to fuse multiple instances. Our segmentation model achieves an average IoU of 60% for five defect class. For the surface model’s semantic recovery and instance statistics, our method achieves accurate statistics of patched area and erosion instances in an environment of 200 m${}^2$, and the average absolute error of the number of spots and cracks has reduced from the original 13.5 to 3.5. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Graphical abstract

Graphical abstract
Full article ">Figure 1
<p>Field testing environment at Three Gorges Dam. (<b>a</b>) deep spillway, (<b>b</b>) drift spillway, (<b>c</b>) robot scanning on a vertical surface, (<b>d</b>) climbing robot.</p> Full article ">Figure 2
<p>Sample images from DSI dataset combined with CSSC dataset. The first two rows are the dam surface multi-defect dataset we collected. Contains crack, patched area, erosion, spot, rope categories, and the two rows in the lower left corner are CSSC datasets, including crack and spalling defect data.</p> Full article ">Figure 3
<p>Pipeline of our algorithm.The algorithm is divided into two parts. The sequence images collected by the wall-climbing robot are selected to establish the DSI dataset, and then the deep neural network model is trained. The 3D reconstruction uses the keyframe and pose data of the sequence images to obtain an initial set of surface voxels through the TSDF method. Then, using the principle of connectivity, the semantic image output is initialized as a collection of instance images of different defects. Finally, the back-projection method and the adjacency matrix are applied to restore and fuse the instance voxels.</p> Full article ">Figure 4
<p>Inspection side-out CRF architecture. We improve Inspection-Net by pooling the output of the HED to the corresponding level size of the U-Net. Then, each pooling feature is the input of the guide layer of the PAC, and each level of the decoder is guided to output the segmentation results. Next, interpolate the results of each level and perform learnable weighting as the initial output result. Finally, through the CRF layer, the original image and the initial result are iteratively optimized, and the final segmentation result is output.</p> Full article ">Figure 5
<p>PAC Module. PAC achieves spatial adaptability through the guide layer because each feature of the guide layer is not limited to a fixed kernel.</p> Full article ">Figure 6
<p>Keyframes back-projection method. The left side shows why the original TSDF method loses semantic voxels during the sparse keyframe reconstruction process. Due to the limitation of the truncated value, the sparse defect semantics will be removed as outliers during the fusion process. On the right, the semantic pixels are restored to the initialized 3D grid using the keyframe instance image pixel back-projection method.</p> Full article ">Figure 7
<p>Instances fusion illustration in one class. (Note: There are three instance examples in color orange, light green, and light blue. We present their overlapping in 2D grid, the green grid represent the overlapping between instance 1 and instance 2, the blue grid represent the overlapping between instance 1 and instance 3. The number of green and blue grids are 4 and 6 which are converted to overlapping matrix element <math display="inline"><semantics> <msubsup> <mi>M</mi> <mrow> <mn>12</mn> </mrow> <mi>c</mi> </msubsup> </semantics></math> and <math display="inline"><semantics> <msubsup> <mi>M</mi> <mrow> <mn>13</mn> </mrow> <mi>c</mi> </msubsup> </semantics></math>. Instance 2 and instance 3 are not overlapping which leads to <math display="inline"><semantics> <mrow> <msubsup> <mi>M</mi> <mrow> <mn>23</mn> </mrow> <mi>c</mi> </msubsup> <mo>=</mo> <mn>0</mn> </mrow> </semantics></math>. The diagonal of <math display="inline"><semantics> <msup> <mi>M</mi> <mi>c</mi> </msup> </semantics></math> represent residual instance voxels of smaller instance after being overlapped by bigger instances (e.g., <math display="inline"><semantics> <mrow> <msubsup> <mi>M</mi> <mrow> <mn>11</mn> </mrow> <mi>c</mi> </msubsup> <mo>=</mo> <mn>5</mn> </mrow> </semantics></math>). Row sum up can obtain each instance original number, and by using Equation (<a href="#FD8-buildings-13-00285" class="html-disp-formula">8</a>), we can calculate the IoU matrix for class <span class="html-italic">c</span>.</p> Full article ">Figure 8
<p>Training result comparison of different models. (<b>a</b>) The training Lovasz loss value (directly related to IoU), (<b>b</b>) defect mIoU, and (<b>c</b>) spot IoU changes during training are compared.</p> Full article ">Figure 9
<p>Segment results on different dataset and images. We selected some representative images from the segmentation results of the DSI and CSSC datasets, and the orange ellipses circle the parts with obvious contrast in the segmentation results.</p> Full article ">Figure 10
<p>3D instance reconstruction comparison. Keyframes GT using manual label to obtain defect segmentation; second and third columns compare naive distance TSDF and our back-projection with adjacency matrix approach, where the second and third rows show the details of instance recovery and reconstruction, the different instance of a class was in a different color. Red dotted circles show the recovery of class voxels, and blue dotted circles show the instance fusion of different classes. Our method shows better consistency.</p> Full article ">

11 pages, 1988 KiB  

Open AccessArticle

Investigation into Recognition Technology of Helmet Wearing Based on HBSYOLOX-s

by Teng Gao and Xianwu Zhang

Appl. Sci. 2022, 12(24), 12997; https://doi.org/10.3390/app122412997 - 18 Dec 2022

Cited by 2 | Viewed by 1449

Abstract

This work proposes a new approach based on YOLOX model enhancement for the helmet-wearing real-time detection task, which is plagued by low detection accuracy, incorrect detection, and missing detection. First, in the backbone network, recursive gated convolution (gⁿConv) is utilized instead [...] Read more.

This work proposes a new approach based on YOLOX model enhancement for the helmet-wearing real-time detection task, which is plagued by low detection accuracy, incorrect detection, and missing detection. First, in the backbone network, recursive gated convolution (gⁿConv) is utilized instead of traditional convolution, hence addressing the issue of extracting many worthless features due to excessive redundancy in the process of feature extraction using conventional convolution. Replace the original FPN layer in the Neck network with the EfficientNet-BiFPN layer. Realize top–down and bottom–up bidirectional fusion of deep and shallow features to improve the flow of feature data between network layers. Lastly, the SIOU cross-entropy loss function is implemented to address the issue of missed detections in crowded environments and further increase the model’s detection precision. Experiments and data comparisons indicate that the modified model’s average detection accuracy is 95.5%, which is 5.4% higher than that of the original network model, and that the detection speed has been dramatically increased to suit actual production requirements. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Network architecture of HorNet.</p> Full article ">Figure 2
<p>BiFPN structural diagram.</p> Full article ">Figure 3
<p>Structure of improved YOLOX-s.</p> Full article ">Figure 4
<p>mAP curves of different improvement methods.</p> Full article ">Figure 5
<p>(<b>a</b>) Hand-held helmet prone to wrong inspection, (<b>b</b>) long-range small target phenomenon, (<b>c</b>) severe obstruction problem, (<b>d</b>) crowding phenomenon.</p> Full article ">

15 pages, 8225 KiB  

Open AccessArticle

Crack Segmentation on Earthen Heritage Site Surfaces

by Yuan Zhang, Zhiyong Zhang, Wu Zhao and Qiang Li

Appl. Sci. 2022, 12(24), 12830; https://doi.org/10.3390/app122412830 - 14 Dec 2022

Cited by 7 | Viewed by 2013

Abstract

Earthen heritage sites are historical relics left by ancient human activity, with earthen as the primary building material, and have significant historical, scientific, and artistic value. However, many sites have experienced extensive deterioration caused by environmental forces and human factors. A crack is [...] Read more.

Earthen heritage sites are historical relics left by ancient human activity, with earthen as the primary building material, and have significant historical, scientific, and artistic value. However, many sites have experienced extensive deterioration caused by environmental forces and human factors. A crack is a kind of typical damage to the walls of earthen heritage sites. Studies of the crack-formation process can effectively predict trends in damage, which will play a critical role in the maintenance of earthen heritage sites. This study is the first of its kind to propose a deep learning method to study the cracks on earthen heritage sites at the pixel-level, adopt the idea of transfer learning, and employ a mixed-crack image dataset for training three deep learning models. The precision, recall, IoU, and F1 metrics were used to evaluate the performance of the trained models. The experimental results showed that FPN-vgg16 appeared to have the highest level of applicability to detect cracks on earthen heritage sites among all networks, due to the highest F1 score of 84.40% and the highest IoU score of 73.11%. The results illustrated that the proposed method in this paper can effectively be used to analyze the rammed earth surface crack images, with great potential in related research fields. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Suoyang Ancient City. (<b>a</b>) Google Maps image of Suoyang Ancient City. The photos show (<b>b</b>) some typical damage to wall surfaces, including (<b>c</b>) scouring, (<b>d</b>) flaking, and (<b>e</b>) cracking.</p> Full article ">Figure 2
<p>Workflow of the proposed method for detecting and predicting cracks in rammed earthen heritage sites.</p> Full article ">Figure 3
<p>Samples of cracks in the training dataset.</p> Full article ">Figure 4
<p>The metrics Precision, Recall, IoU, and F1 score as obtained from U-Net for different backbones: (<b>a</b>) vgg16, (<b>b</b>) resnet152, (<b>c</b>) densenet201, (<b>d</b>) inceptionv3.</p> Full article ">Figure 5
<p>The metrics Precision, Recall, IoU, and F1 score as obtained from Linknet for different backbones: (<b>a</b>) vgg16, (<b>b</b>) resnet152, (<b>c</b>) densenet201, (<b>d</b>) inceptionv3.</p> Full article ">Figure 5 Cont.
<p>The metrics Precision, Recall, IoU, and F1 score as obtained from Linknet for different backbones: (<b>a</b>) vgg16, (<b>b</b>) resnet152, (<b>c</b>) densenet201, (<b>d</b>) inceptionv3.</p> Full article ">Figure 6
<p>The metrics Precision, Recall, IoU, and F1 score as obtained from FPN for different backbones: (<b>a</b>) vgg16, (<b>b</b>) resnet152, (<b>c</b>) densenet201, (<b>d</b>) inceptionv3.</p> Full article ">Figure 7
<p>The original image, ground truth, and predictions with all models for different images from the validation dataset.</p> Full article ">Figure 7 Cont.
<p>The original image, ground truth, and predictions with all models for different images from the validation dataset.</p> Full article ">

15 pages, 3701 KiB  

Open AccessArticle

Long Short-Term Memory-Based Methodology for Predicting Carbonation Models of Reinforced Concrete Slab Bridges: Case Study in South Korea

by Tae Ho Kwon, Jaehwan Kim, Ki-Tae Park and Kyu-San Jung

Appl. Sci. 2022, 12(23), 12470; https://doi.org/10.3390/app122312470 - 6 Dec 2022

Cited by 4 | Viewed by 2218

Abstract

Reinforced concrete slab (RCS) bridges deteriorate because of exposure to environmental factors over time, resulting in reduced durability. Particularly, the carbonation of RCS bridges corrodes the rebars and reduces the strength. However, carbonation models derived from short-term experiments exhibit low reliability with respect [...] Read more.

Reinforced concrete slab (RCS) bridges deteriorate because of exposure to environmental factors over time, resulting in reduced durability. Particularly, the carbonation of RCS bridges corrodes the rebars and reduces the strength. However, carbonation models derived from short-term experiments exhibit low reliability with respect to existing bridges. Therefore, a long short-term memory (LSTM)-based methodology was developed in this study for generating carbonation models using existing bridge inspection reports. The proposed methodology trains the LSTM model by combining data extracted from reports and local environmental data. The learning process uses padding and masking methods to consider the history of environmental data. A case study was performed to validate the proposed method in three different regions of Korea. The results verified that the coefficient of determination of the proposed method was higher than those of the existing carbonation models and other regression analyses. Therefore, the developed methodology can be used for predicting regional carbonation models using the data from existing bridges. Full article

(This article belongs to the Topic AI Enhanced Civil Infrastructure Safety)

►▼ Show Figures

Figure 1

Figure 1
<p>Conceptual structure of the long short-term memory (LSTM).</p> Full article ">Figure 2
<p>Conceptual structure of the padding method.</p> Full article ">Figure 3
<p>Example of weight matrix of the masking method.</p> Full article ">Figure 4
<p>Carbonation depth according to regions: (<b>a</b>) Region A; (<b>b</b>) Region B; and (<b>c</b>) Region C.</p> Full article ">Figure 5
<p>History of environmental conditions according to regions: (<b>a</b>) average temperature; (<b>b</b>) difference in daily temperature; (<b>c</b>) relative humidity; (<b>d</b>) carbon dioxide concentration; (<b>e</b>) precipitation; and (<b>f</b>) number of snowy days.</p> Full article ">Figure 6
<p>Methodology for generating a carbonation model for actual bridges.</p> Full article ">Figure 7
<p>Proposed LSTM model for generating the carbonation models.</p> Full article ">Figure 8
<p>Carbonation prediction results of the LSTM model for the (<b>a</b>) training and (<b>b</b>) validation datasets.</p> Full article ">Figure 9
<p>The k-fold cross-validation test results of the proposed model.</p> Full article ">

Show export options expand_more Show export options expand_less

Select all

Export citation of selected articles as:

Plain Text BibTeX BibTeX (without abstracts) Endnote Endnote (without abstracts) Tab-delimited RIS

 

Error

Oops... you haven't selected anything for export.

Ok

clear

Displaying articles 1-33