[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN118378195B - Screw air compressor fault prediction method based on multi-source data fusion - Google Patents

Screw air compressor fault prediction method based on multi-source data fusion Download PDF

Info

Publication number
CN118378195B
CN118378195B CN202410805946.9A CN202410805946A CN118378195B CN 118378195 B CN118378195 B CN 118378195B CN 202410805946 A CN202410805946 A CN 202410805946A CN 118378195 B CN118378195 B CN 118378195B
Authority
CN
China
Prior art keywords
data
fusion
feature
source
time sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202410805946.9A
Other languages
Chinese (zh)
Other versions
CN118378195A (en
Inventor
窦文强
赵迎普
黄华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Denair Energy Saving Technology Shanghai Co ltd
Denair Energy Equipment Co ltd
Original Assignee
Denair Energy Saving Technology Shanghai Co ltd
Denair Energy Equipment Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Denair Energy Saving Technology Shanghai Co ltd, Denair Energy Equipment Co ltd filed Critical Denair Energy Saving Technology Shanghai Co ltd
Priority to CN202410805946.9A priority Critical patent/CN118378195B/en
Publication of CN118378195A publication Critical patent/CN118378195A/en
Application granted granted Critical
Publication of CN118378195B publication Critical patent/CN118378195B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F04—POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
    • F04B—POSITIVE-DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS
    • F04B51/00—Testing machines, pumps, or pumping installations
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D21/00—Measuring or testing not otherwise provided for
    • G01D21/02—Measuring two or more variables by means not covered by a single other subclass
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/251—Fusion techniques of input or preprocessed data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2123/00—Data types
    • G06F2123/02—Data types in the time domain, e.g. time-series data

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mechanical Engineering (AREA)
  • Probability & Statistics with Applications (AREA)
  • Testing And Monitoring For Control Systems (AREA)
  • Applications Or Details Of Rotary Compressors (AREA)

Abstract

The invention discloses a screw air compressor fault prediction method based on multi-source data fusion, which relates to the field of data processing, and comprises the following steps: acquiring multi-source data; preprocessing and extracting the characteristics of the monitoring data; based on the source of the multi-source data, carrying out characteristic correlation evaluation by combining the monitored data characteristics, and screening target characteristics; performing feature fusion, determining a multi-source feature fusion rule, and constructing a training data fusion channel for performing feature fusion on multi-source data to generate a training data set; constructing a time sequence training data set and training a time sequence prediction model; the method comprises the steps of deploying the method into a monitoring system, and predicting fault time sequences according to multisource monitoring data acquired in real time. The technical problem of the poor prediction accuracy of current screw air compressor fault prediction existence has been solved, the technical effect of the accurate prediction of realization to the fault emergence has been reached.

Description

Screw air compressor fault prediction method based on multi-source data fusion
Technical Field
The application relates to the field of data processing, in particular to a screw air compressor fault prediction method based on multi-source data fusion.
Background
With the rapid development of industrial automation, the screw air compressor is used as an important power device, and the stability and the reliability of the screw air compressor are important for guaranteeing the continuity and the safety of the production process. The screw air compressor is affected by various factors in the operation process, such as equipment aging, improper operation, external environment change and the like, and the factors can cause equipment failure. At present, a monitoring method based on a single sensor is generally adopted for fault prediction of the screw air compressor, the method mainly depends on sensor data of equipment, whether the equipment is in a normal state or not is judged by setting a threshold value or a simple statistical model, and the method cannot accurately predict the fault occurrence time and the fault type of the equipment.
In the related art at the present stage, the screw air compressor fault prediction has the technical problem of poor prediction accuracy.
Disclosure of Invention
The application provides a screw air compressor fault prediction method based on multi-source data fusion, which adopts a plurality of sensor data integrated on the screw air compressor to comprehensively monitor the running state of equipment, screen out target characteristics, perform characteristic fusion on the target characteristics, train a time sequence prediction model and other technical means, thereby achieving the technical effect of realizing accurate prediction of fault occurrence.
The application provides a screw air compressor fault prediction method based on multi-source data fusion, which comprises the following steps:
Acquiring multi-source data, wherein the multi-source data is acquired by a sensor arranged on a screw air compressor and comprises exhaust seat vibration data, cylinder body vibration data, rotor axial displacement data, oil inlet pressure data, exhaust port pressure data and exhaust port temperature data; preprocessing the multi-source data, and extracting monitoring data characteristics of the multi-source data; based on the source of the multi-source data, carrying out feature correlation evaluation by combining the monitored data features, and screening target features; performing feature fusion according to the target features, determining a multi-source feature fusion rule, and constructing a training data fusion channel, wherein the training data fusion channel is used for performing feature fusion on multi-source data to generate a training data set; based on the training data set, constructing a time sequence training data set and training a time sequence prediction model, wherein the time sequence prediction model is used for predicting fault occurrence time information of the screw air compressor; and deploying the trained time sequence prediction model into a monitoring system, and performing fault time sequence prediction according to the multisource monitoring data acquired in real time.
In a possible implementation manner, preprocessing is performed on the multi-source data, monitoring data characteristics of the multi-source data are extracted, and the following processing is performed:
Carrying out normalization processing on the multi-source data, and carrying out multi-source data synchronous alignment according to the monitoring time to obtain preprocessed multi-source data; mapping the associated twin network branches and the monitoring sensors, and inputting monitoring data corresponding to the monitoring sensors into the twin network branches according to the mapping association relation to extract data characteristics; and respectively obtaining the monitoring data characteristics of each sensor, and combining the monitoring data characteristics of the multi-source data to obtain the monitoring data characteristics of the multi-source data.
In a possible implementation manner, the mapping associates twin network branches and monitoring sensors, and performs the following processing:
Constructing a feature extraction network model; performing network replication according to the configuration quantity based on the feature extraction network model to construct a plurality of network branches, wherein the network branches have the same network structure and share weights to construct a twin network model; and constructing a connecting channel according to the sensors arranged on the screw air compressor, and establishing a mapping association relation between the monitoring sensor and each network branch of the twin network model.
In a possible implementation manner, based on the source of the multi-source data, performing feature correlation evaluation in combination with the monitored data features, screening target features, and performing the following processing:
According to the source of the multi-source data, a monitoring data characteristic pair is established, wherein the monitoring data characteristic pair consists of two monitoring data characteristics with different data sources and the same acquisition time; calculating the correlation coefficient of the monitoring data feature pair; and screening and combining the multiple co-linear data according to the correlation coefficient to obtain the target feature.
In a possible implementation manner, feature fusion is performed according to the target features, a multi-source feature fusion rule is determined, a training data fusion channel is constructed, and the following processing is performed:
According to the monitoring data characteristics of the exhaust seat vibration data, the cylinder body vibration data, the rotor axial displacement data, the oil inlet pressure data, the exhaust port pressure data and the exhaust port temperature data, configuring data characteristic classes, wherein the data characteristic classes comprise data directions and data fluctuation amplitudes; classifying the target features based on the data feature classes to construct a multi-feature class set; based on the correlation coefficient, carrying out simultaneous screening on the multi-feature class set according to a preset screening threshold value to determine a fusion feature group, wherein the fusion feature group comprises two or more multi-feature classes; taking the fusion relation of a plurality of different fusion feature groups as the multi-source feature fusion rule; and constructing a multi-path fusion module according to the multi-source characteristic fusion rule, integrating a screw air compressor state acquisition channel, and constructing the training data fusion channel.
In a possible implementation, based on the training data set, a time-series training data set is constructed, and the following processing is performed:
Clustering the training data sets according to the fusion feature groups to obtain training data sets of the fusion feature groups, wherein the training data sets comprise monitoring time identifiers; performing fault sample screening according to the state information of the screw air compressor in the training data set to obtain a target training data set; the time sequence training data sets are constructed by performing time sequence arrangement on the fusion characteristic sets based on the monitoring time identification of the target training data set, wherein the time sequence training data sets comprise a plurality of groups of training data with time sequence relation, and the plurality of groups of training data have the same fusion characteristic set and screw air compressor fault information; judging whether the time sequence training data set reaches a preset number, and when the time sequence training data set does not reach the preset number, performing mutation derivatization or simulation test based on the time sequence training data set to expand the time sequence training data set.
In a possible implementation manner, the training time sequence prediction model performs the following processing:
Determining a time sequence module framework, wherein the time sequence module framework is a hidden Markov model; training the time sequence module frame by using the time sequence training data set, taking the fault information of the screw air compressor as a target state, determining the probability of each time point transferring to the target state, and training and converging by using a forward-backward algorithm to obtain model parameters; and obtaining the time sequence prediction model based on the model parameters.
In a possible implementation manner, the fault time sequence prediction is performed according to the multi-source monitoring data collected in real time, and the following processing is performed:
Acquiring target characteristics according to the multisource monitoring data acquired in real time; matching the target features with a multisource feature fusion rule, and obtaining fusion features through a training data fusion channel; carrying out state probability prediction on the fusion characteristics through the time sequence prediction model to obtain fault probability and corresponding time nodes; acquiring a trusted coefficient of a time node, and determining a self-adaptive early warning strategy based on the trusted coefficient and the fault probability; generating a fault time sequence early warning result according to the self-adaptive early warning strategy, the fault probability and the corresponding time node.
According to the screw air compressor fault prediction method based on multi-source data fusion, multi-source data are firstly acquired, the multi-source data are collected for sensors arranged on the screw air compressor, the sensor comprises exhaust seat vibration data, cylinder body vibration data, rotor axial displacement data, oil inlet pressure data, exhaust port pressure data and exhaust port temperature data, then the multi-source data are preprocessed, monitoring data characteristics of the multi-source data are extracted, further, based on sources of the multi-source data, characteristic correlation evaluation is conducted by combining the monitoring data characteristics, target characteristics are screened, characteristic fusion is conducted according to the target characteristics, a multi-source characteristic fusion rule is determined, a training data fusion channel is constructed, the multi-source data are subjected to characteristic fusion to generate a training data set, then a time sequence training data set is constructed based on the training data set, a training time sequence prediction model is used for predicting time information of the screw air compressor fault occurrence, finally the time sequence prediction model after training is deployed into a monitoring system, and the technical effect of accurate prediction of fault occurrence is achieved according to the multi-source monitoring data collected in real time.
Drawings
In order to more clearly illustrate the technical solution of the embodiments of the present application, the following description will briefly explain the drawings of the embodiments of the present application, in which flowcharts are used to illustrate operations performed by the method according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in order precisely. Rather, the various steps may be processed in reverse order or simultaneously, as desired. Also, other operations may be added to or removed from these processes.
Fig. 1 is a schematic flow chart of a screw air compressor fault prediction method based on multi-source data fusion provided by an embodiment of the application.
Fig. 2 is a schematic flow chart of constructing a training data fusion channel in a screw air compressor fault prediction method based on multi-source data fusion according to an embodiment of the present application.
Detailed Description
The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the accompanying drawings, and the described embodiments should not be construed as limiting the present application, and all other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present application.
In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict. The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server that comprises a list of steps or elements is not necessarily limited to those steps or elements that are expressly listed or inherent to such process, method, article, or apparatus, but may include other steps or modules that may not be expressly listed or inherent to such process, method, article, or apparatus, and unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used herein is for the purpose of describing embodiments of the application only.
The embodiment of the application provides a screw air compressor fault prediction method based on multi-source data fusion, which comprises the following steps of:
Step S100, multi-source data are acquired, wherein the multi-source data are acquired by sensors arranged on the screw air compressor and comprise exhaust seat vibration data, cylinder body vibration data, rotor axial displacement data, oil inlet pressure data, exhaust port pressure data and exhaust port temperature data. Specifically, key positions, such as an exhaust seat, a cylinder body, a rotor and the like, to be monitored on the screw air compressor are identified, proper sensor types, such as a vibration sensor, a pressure sensor, a temperature sensor and the like, are selected according to monitoring requirements, and the sensors are arranged at corresponding positions of the screw air compressor, so that the sensors can accurately capture required data. The sensor is connected with a data acquisition system (such as a data acquisition card and the like), so that data can be transmitted in real time, the frequency and the sampling period of data acquisition are set according to the operation characteristics and the monitoring requirements of the screw air compressor, the data acquisition system is started, and multi-source data on the screw air compressor, namely data sets from different sensors, are acquired in real time. Specifically, the vibration sensor arranged on the exhaust seat is used for collecting the vibration data of the exhaust seat, and the vibration data of the exhaust seat reflects the vibration condition of the exhaust seat in the running process; acquiring cylinder body vibration data by a vibration sensor arranged on the cylinder body, wherein the cylinder body vibration data reflects the vibration condition of the cylinder body in the running process; acquiring rotor axial displacement data by a rotor axial displacement sensor, wherein the rotor axial displacement data reflects the axial position change of the rotor in the running process; collecting oil inlet pressure data by an oil inlet pressure sensor, wherein the oil inlet pressure data reflects the pressure condition of an oil inlet of a screw air compressor; collecting exhaust port pressure data by an exhaust port pressure sensor, wherein the exhaust port pressure data reflects the pressure condition of an exhaust port of the screw air compressor; and the exhaust port temperature sensor is used for collecting exhaust port temperature data, and the exhaust port temperature data reflects the temperature condition of the exhaust port of the screw air compressor.
Step S200, preprocessing the multi-source data, and extracting monitoring data characteristics of the multi-source data. Specifically, for missing data, an interpolation method (such as linear interpolation, polynomial interpolation, etc.), a filling method (such as mean filling, median filling, etc.), or a deletion method is adopted for processing; according to the data distribution and the actual situation, identifying an abnormal value, and determining whether to delete the abnormal value, replace the abnormal value with other values (such as a mean value, a median value and the like) or keep the abnormal value; if duplicate data records exist in the dataset, they are removed to avoid affecting subsequent analysis. Each feature in the multi-source data is transformed using a normalization or normalization method such as min-max normalization, Z-score normalization, etc., to eliminate dimensional differences and scale differences between different features. And processing the vibration data by using noise reduction algorithms such as wavelet transformation, kalman filtering and the like so as to reduce noise interference and improve the data quality. According to the monitoring requirements and the data characteristics, determining the types of the characteristics (such as time domain characteristics, frequency domain characteristics and the like) to be extracted, processing the multi-source data by applying a corresponding characteristic extraction method (such as Fourier transform, a statistical method, a machine learning algorithm and the like), and extracting the characteristics related to the running state and faults of the screw air compressor. Further filtering and optimization may be performed on the extracted features to select the most representative feature subset.
In a possible implementation manner, the multi-source data is preprocessed, the monitoring data features of the multi-source data are extracted, step S200 further includes step S210, normalization processing is performed on the multi-source data, and multi-source data synchronization alignment is performed according to the monitoring time, so as to obtain preprocessed multi-source data. Specifically, the range and units of each data source are analyzed, normalization methods and parameters are determined, and normalization algorithms (e.g., min-max normalization) are applied to convert the data into a uniform range of values, such as [0,1] or [ -1,1]. And checking time stamps of data acquisition of different sensors, aligning multi-source data according to the time stamps, enabling the data at the same time point to correspond, and filling or correcting missing or abnormal time stamp data by adopting interpolation or other methods. Step S220, mapping the associated twin network branches and the monitoring sensors, and inputting the monitoring data of the corresponding monitoring sensors into the twin network branches according to the mapping association relation to extract the data characteristics. Specifically, the correspondence between a twin network branch, which is a network structure specifically designed for each sensor data, for extracting the characteristics of the specific data, and a specific monitoring sensor is determined. A mapping table or logic is established for inputting the data of a particular sensor into the correct twin network branch. And respectively inputting the synchronously aligned multi-source data into corresponding twin network branches according to the mapping relation, and extracting the characteristics of the respective sensor data by utilizing the trained twin network branches. Step S230, monitoring data characteristics of each sensor are obtained respectively, and the monitoring data characteristics of the multi-source data are combined to obtain the monitoring data characteristics of the multi-source data. Specifically, the processed monitoring data features are extracted from each twin network branch, and the monitoring data features from different sensors are combined to serve as the monitoring data features of the multi-source data. The implementation mode utilizes the twin network to extract the characteristics, and achieves the technical effects of effectively extracting and representing the data characteristics.
In a possible implementation manner, the mapping associates twin network branches and monitoring sensors, and step S220 further includes step S221, and a feature extraction network model is constructed. Specifically, a feature extraction network structure, such as a Convolutional Neural Network (CNN), a cyclic neural network (RNN), etc., is selected or designed according to the data characteristics, and the weights and bias items in the network are initialized by adopting methods of random initialization, xavier initialization, etc. Step S222, performing network replication according to the configuration number based on the feature extraction network model, constructing a plurality of network branches, wherein each network branch has the same network structure and shares weights, and constructing a twin network model, specifically, replicating the constructed feature extraction network model according to the required number to form a plurality of network branches, where the network structures of the plurality of network branches are identical, and weights (parameters) are shared, that is, when the weights of one network branch are updated, the weights of the other network branches are updated correspondingly, and constructing the twin network model through the plurality of network branches sharing the weights. And S223, constructing a connecting channel according to the sensors arranged on the screw air compressor, and establishing a mapping association relation between the monitoring sensor and each network branch of the twin network model. Specifically, by constructing a connection channel, a mapping association relation between the monitoring sensor and each network branch of the twin network model is established, and each sensor data stream corresponds to one network branch, namely, the characteristic extraction process of each data source is consistent. According to the implementation mode, all network branches in the twin network have the same network structure and shared weight, so that consistency can be kept when feature extraction is performed on different sensor data, and the technical effect of improving the accuracy of subsequent feature comparison and fusion is achieved.
Step S300, based on the source of the multi-source data, carrying out feature correlation evaluation by combining the monitored data features, and screening target features. Specifically, the source and the characteristics of the multi-source data are analyzed, the physical meaning represented by each sensor data and the influence of the physical meaning on the running state of the screw air compressor are determined, the correlation coefficient (such as the pearson correlation coefficient, the spearman rank correlation coefficient and the like) among different characteristics is calculated so as to quantify the correlation among the characteristics (whether certain correlation or dependency exists among different characteristics), and a characteristic correlation matrix or a thermodynamic diagram can be drawn so as to intuitively display the correlation relation among the characteristics. According to the fault prediction requirement of the screw air compressor, the characteristics are combined with the factors such as the correlation, redundancy, interpretability and the like of the target prediction task, and the standard and the criterion (used for determining which characteristics should be reserved or removed) of the characteristics are determined. And screening out the characteristics highly correlated with the target prediction task as target characteristics by combining the characteristic correlation evaluation result and the characteristic selection standard, wherein the target characteristics are the characteristic set which is determined after screening and is required for subsequent modeling and analysis, and the characteristics are highly correlated with the target prediction task, so that the running state and fault information of the screw air compressor can be accurately reflected. Screening may be performed, for example, using a threshold-based approach (e.g., setting a minimum threshold for correlation coefficients) or a machine learning model-based approach (e.g., evaluating feature importance using a random forest, gradient hoist, etc. model).
In one possible implementation manner, based on the source of the multi-source data, feature correlation evaluation is performed in combination with the monitored data features, and the target features are screened, and step S300 further includes step S310, according to the source of the multi-source data, a monitored data feature pair is established, where the monitored data feature pair is composed of two monitored data features with different data sources and the same acquisition time. Specifically, the sources of all multi-source data are identified and analyzed, time synchronization operation is carried out on data from different sources and with the same acquisition time, so that the time stamps or the acquisition time periods of the data are consistent, and the two monitoring data features with different sources and the same acquisition time are paired according to the sources and the acquisition time of the data to form a monitoring data feature pair. In step S320, a correlation coefficient of the monitored data feature pair is calculated. Specifically, the correlation coefficient calculation formula is as follows: Wherein, the method comprises the steps of, wherein, Is to monitor the correlation coefficient of the data features to x and y,It is the number of pairs of data features that are monitored,AndIs to monitor the value of the data feature pair at the ith sample point,AndThe average of x and y is given respectively. And calculating the correlation coefficient between each monitoring data characteristic pair by using a correlation coefficient calculation formula. And step S330, screening and combining the multiple co-linear data according to the correlation coefficient to obtain the target feature. Specifically, one or more correlation coefficient thresholds are set, and for pairs of monitored data features whose correlation coefficients exceed the set thresholds, it is considered that there is multiple collinearity, i.e., a change in one feature can be predicted or interpreted from a change in another feature, and for pairs of monitored data features for which there is multiple collinearity, one is selected as a representation (merge), or one is deleted (deletion), to avoid information redundancy or misleading in subsequent analysis or modeling. After screening and merging, the remaining monitoring data features are target features, and the correlation among the target features is low, so that the diversity and complexity of the data can be more comprehensively reflected. The realization method avoids information redundancy by screening and combining the collinearity characteristics, and achieves the technical effect of improving the accuracy and generalization capability of the time sequence prediction model.
Step S400, performing feature fusion according to the target features, determining a multi-source feature fusion rule, and constructing a training data fusion channel, wherein the training data fusion channel is used for performing feature fusion on multi-source data to generate a training data set. Specifically, the type and nature of the target feature, such as time domain features, frequency domain features, statistical features, etc., are analyzed, and feature fusion strategies, such as statistical-based fusion (e.g., mean, median, variance, etc.), model-based fusion (e.g., learning nonlinear relationships between features using a machine learning model), or transform-based fusion (e.g., principal component analysis, wavelet transform, etc.), are selected based on the feature type. According to the monitoring requirement and the target prediction task of the screw air compressor, specific rules of multi-source feature fusion are formulated, wherein the specific rules comprise determining weights or importance of features of different sources in the fusion process, feature combination modes and the like. And writing or configuring codes or tools for data fusion according to the determined feature fusion strategy and rule, and creating one or more data fusion channels for fusing target features in the multi-source data. The method comprises the steps of inputting cleaned, standardized and screened target feature data into a training data fusion channel, fusing target features in multi-source data according to a specified fusion rule and strategy through the training data fusion channel, outputting fused feature data, and combining the fused feature data with corresponding labels (such as the running state and fault type of a screw air compressor) to form a training data set.
As shown in fig. 2, in one possible implementation manner, feature fusion is performed according to the target feature, a multi-source feature fusion rule is determined, and a training data fusion channel is constructed, and step S400 further includes step S410, configuring a data feature class according to the monitored data features of the exhaust seat vibration data, the cylinder body vibration data, the rotor axial displacement data, the oil inlet pressure data, the exhaust port pressure data, and the exhaust port temperature data, where the data feature class includes a data direction and a data fluctuation amplitude. Specifically, key features extracted from exhaust seat vibration data, cylinder body vibration data, rotor axial displacement data, oil inlet pressure data, exhaust port pressure data and exhaust port temperature data are identified, and data direction classification and data fluctuation amplitude statistics are carried out, wherein the data direction refers to the trend or direction of data change, such as the up-down direction of vibration, the rise and fall of temperature and the like; the data fluctuation amplitude represents the range or intensity of the data change, such as vibration amplitude, temperature difference, etc. And step S420, classifying the target features based on the data feature classes, and constructing a multi-feature class set. Specifically, data features having similarity or correlation are classified into one type, such as all vibration-related features are classified into one type, all temperature-related features are classified into one type, and the like, forming a plurality of such feature class sets, each set containing one type of data feature. Step S430, based on the correlation coefficient, performing time screening on the multi-feature class set according to a preset screening threshold value to determine a fusion feature group, wherein the fusion feature group comprises two or more multi-feature classes. Specifically, all feature data are aligned in time, that is, all feature data are based on the same time point, correlation coefficients among different feature class sets are calculated to measure the correlation among the feature class sets, feature class sets with correlation higher than a preset screening threshold (used for determining which feature class sets should be combined into the lowest correlation standard of a fusion feature set) are screened out, and the screened feature class sets with high correlation are combined into the fusion feature set. Step S440, taking the fusion relation of a plurality of different fusion feature groups as the multi-source feature fusion rule. Specifically, fusion relations among multiple feature classes in different fusion feature groups, such as which features are mutually dependent or mutually influenced in the operation of the screw air compressor, are analyzed, and based on the fusion relations, a multi-source feature fusion rule is determined, namely, how to effectively fuse the multiple feature classes together so as to obtain more accurate state information of the screw air compressor. and S450, constructing a multi-channel fusion module according to the multi-source feature fusion rule, integrating a screw air compressor state acquisition channel, and constructing the training data fusion channel. Specifically, a multi-channel fusion module is designed according to a multi-source feature fusion rule, the multi-channel fusion module is used for simultaneously processing a plurality of feature class sets, fusing the multi-feature classes together according to the fusion rule, integrating the multi-channel fusion module into a screw air compressor state acquisition system and obtaining training data fusion channels by acquiring and fusing data of the plurality of feature class sets in real time. According to the implementation mode, the fusion feature group is determined based on the correlation coefficient, so that the feature subjected to fusion is ensured to have strong correlation, and the technical effect of improving the accuracy of feature fusion is achieved.
And S500, constructing a time sequence training data set based on the training data set, and training a time sequence prediction model, wherein the time sequence prediction model is used for predicting the fault occurrence time information of the screw air compressor. Specifically, feature data related to the fault occurrence time of the screw air compressor is extracted from a training data set, the feature data are ordered according to a time sequence according to the characteristics of a time sequence, the time sequence of the data is ensured, the time step of the data (namely the interval of each time point) and the size of a sliding window (namely the data range of each input model) are determined according to the requirements of a prediction model and the characteristics of the data, the ordered feature data are divided according to the time step and the sliding window, a time sequence training data set is constructed, and each time sequence training data set comprises feature data of a plurality of continuous time points and is used for predicting the fault occurrence time information of the next time point by the training model. A time series prediction model, such as a Recurrent Neural Network (RNN), a long and short time memory network (LSTM), a gated loop unit (GRU), etc., is determined and used to capture time dependence and dynamic changes in the sequence data, and parameters and structures of the model, such as the number of hidden layers, the number of neurons, activation functions, etc., are configured. Training the time sequence prediction model by using the constructed time sequence training data sets, taking each time sequence training data set as the input of the time sequence prediction model, taking the fault occurrence time information of the corresponding time point as the output or label of the time sequence prediction model, and optimizing the parameters of the time sequence prediction model through a back propagation algorithm and a gradient descent algorithm so that the time sequence prediction model can accurately predict the fault occurrence time information of the next time point.
In one possible implementation manner, a time sequence training data set is constructed based on the training data set, and step S500 further includes step S510, clustering the training data sets according to the fusion feature set to obtain training data sets of each fusion feature set, where the training data sets include monitoring time identifiers. Specifically, clustering is performed on the training data set by using a clustering algorithm (such as K-means, hierarchical clustering and the like), so that data in each cluster has higher similarity on the fusion feature group, each cluster represents a group of training data with similar fusion features after the clustering is completed, and meanwhile, the data also comprises a monitoring time identifier. And step S520, screening fault samples according to the state information of the screw air compressor in the training data set to obtain a target training data set. Specifically, identifying samples related to screw air compressor faults from a training data set according to fault labels or state identifiers to form a target training data set, wherein screw air compressor state information refers to various state information generated in the running process of the screw air compressor, and the state information comprises a normal state, a fault state and the like; the failure sample refers to sample data containing screw air compressor failure information. And step S530, performing time sequence arrangement on the fusion characteristic groups based on the monitoring time identification of the target training data set, and constructing the time sequence training data set, wherein the time sequence training data set comprises a plurality of groups of training data with time sequence relation, and the plurality of groups of training data have the same fusion characteristic groups and screw air compressor fault information. Specifically, training data sets of the fused feature sets are ordered according to monitoring time identifiers (such as time stamps) in the target training data sets to form time sequence training data sets, wherein the time sequence training data sets comprise time sequence information evolving from a normal state to a fault state. Step S540, judging whether the time sequence training data set reaches a preset number, and when the time sequence training data set does not reach the preset number, performing variation derivative or simulation test based on the time sequence training data set, and expanding the time sequence training data set. Specifically, whether the currently constructed time sequence training data set meets the preset number requirement is checked, the preset number is a number threshold value of the time sequence training data set according to the model training requirement, if the preset number is not met, the time sequence training data set is expanded by using a variation derivative (new data is generated based on variation of existing data) or simulation test (different fault states of the screw air compressor are simulated and corresponding data are generated), and the expanded time sequence training data set is used for a subsequent training process. According to the implementation mode, the time sequence training data set with the time sequence relation and the fault information is effectively constructed through fault sample screening, so that the accuracy of the time sequence training data set is improved, and the technical effect of improving the prediction accuracy of the time sequence prediction model is achieved.
In a possible implementation manner, the training time sequence prediction model, step S500 further includes step S550, determining a time sequence module framework, where the time sequence module framework is a hidden markov model. Specifically, in screw air compressor fault prediction, the fault state cannot be directly observed, but can be indirectly inferred through some observable variables (such as operating parameters, temperature, pressure and the like), and the hidden markov model can deal with such problems with hidden states, so the hidden markov model is used as a time sequence module framework. Step S560, training the time sequence module frame by using the time sequence training data set, using the fault information of the screw air compressor as a target state, determining the probability of transferring to the target state at each time point, and training and converging by using a forward-backward algorithm to obtain model parameters. Specifically, the time sequence training data set is preprocessed according to the requirement of the hidden Markov model, for example, is converted into a feature vector sequence, and initial values are given to initial state probability, state transition probability and output probability distribution of the hidden Markov model. And calculating the probability of a given model parameter and an observation sequence in a certain state at a certain moment by using a forward algorithm, calculating the given model parameter, the observation sequence and the probability of the state at the certain moment by using a backward algorithm, iteratively updating the model parameter by using a Baum-Welch algorithm so as to maximize the probability of the observation sequence, and considering that the model is converged when the change of the model parameter is smaller than a certain threshold value. Step S570, obtaining the time series prediction model based on the model parameters. Specifically, the model parameters obtained through training are saved to form a time sequence prediction model. The implementation mode uses the hidden Markov model as a framework of the time sequence prediction model, the hidden Markov model has strong sequence modeling capability, long-term dependency relationship in time sequence data can be captured, and the technical effect of improving the accuracy of fault prediction is achieved.
And step S600, deploying the trained time sequence prediction model into a monitoring system, and performing fault time sequence prediction according to the multisource monitoring data acquired in real time. Specifically, the monitoring system is ensured to have hardware and software environments required by the running time sequence prediction model, including enough computing resources, proper operating systems, necessary dependency libraries and the like, a module or a component for loading and managing the model is configured in the monitoring system, a stored time sequence prediction model file is uploaded or deployed to a designated position of the monitoring system, and codes are written or configured, so that the monitoring system can automatically load and initialize the time sequence prediction model when the monitoring system is started. And a data acquisition module of the monitoring system is configured to acquire multi-source monitoring data related to the running state of the screw air compressor from each sensor in real time, and necessary preprocessing is carried out on the acquired data, such as data cleaning, format conversion, standardization and the like, so as to meet the input requirements of a time sequence prediction model. The preprocessed real-time monitoring data are organized according to a format required by a time sequence prediction model to form an input data sequence, the input data sequence is transmitted to the loaded time sequence prediction model to perform fault time sequence prediction, and the time sequence prediction model predicts the possibility or time information of faults of the screw air compressor at a certain time point in the future according to the characteristic information in the input data sequence. Setting a corresponding alarm mechanism according to the prediction result, triggering an alarm if the impending failure is predicted, and informing related personnel in a mode of sound, light, mail, short message and the like so as to take preventive measures or perform failure processing in time. According to the embodiment of the application, the operation state of the equipment is comprehensively monitored by adopting a plurality of sensor data which are integrally distributed on the screw air compressor, the target characteristics are screened out, the characteristic fusion is carried out on the target characteristics, the time sequence prediction model is trained, and the like, so that the technical effect of accurately predicting the occurrence of faults is achieved.
In one possible implementation manner, the step S600 further includes a step S610 of performing fault timing prediction according to the multi-source monitoring data collected in real time, and obtaining the target feature according to the multi-source monitoring data collected in real time; step S620, matching is carried out by utilizing the target feature and the multi-source feature fusion rule, and fusion features are obtained through a training data fusion channel. Specifically, these two steps are similar to the processing procedure of steps S100-S400, and will not be described here again. Step S630, predicting the state probability of the fusion feature by using the time sequence prediction model, so as to obtain a fault probability and a corresponding time node. Specifically, the fusion features are input into a trained time sequence prediction model, and the time sequence prediction model outputs the probability of the fault state and the corresponding time node according to the input features, namely the probability of the fault occurring in a certain time node in the future. Step S640, obtaining a trusted coefficient of the time node, and determining an adaptive early warning strategy based on the trusted coefficient and the fault probability. Specifically, the reliability coefficient of the time node is calculated according to the distance between the time node and the current time, the historical prediction accuracy and other factors, the reliability coefficient is an index for measuring the reliability of the prediction result of the time node, the early warning level, mode, time and the like are determined according to the reliability coefficient and the fault probability, and a self-adaptive early warning strategy is formed, namely, the strategy of automatically adjusting the early warning mode, time and the like according to the prediction result. For example, the farther from the current time, the lower the reliability of its predictions, the lower the required failure probability threshold. Step S650, generating a fault timing early warning result according to the adaptive early warning strategy, the fault probability and the corresponding time node. Specifically, according to the self-adaptive early warning strategy, a detailed fault time sequence early warning result is generated by combining the fault probability and the corresponding time node, and the early warning result is output to related personnel in the form of short messages, mails, console messages and the like. According to the implementation mode, through the self-adaptive early warning strategy, the early warning mode is automatically adjusted according to the prediction result, so that the technical effect of ensuring that measures can be timely taken before faults occur and reducing the influence of the faults on production operation is achieved.
In one possible specific application implementation mode, a high-precision vibration sensor is selected, the measuring range is +/-50 g, the frequency response range is 10Hz-1kHz, the vibration sensor is arranged on the surface of the exhaust seat and is tightly contacted with the exhaust seat by using screws for fixation; the high-precision vibration sensor is also selected, the measuring range and the frequency response range are the same as those of the exhaust seat vibration sensor, the exhaust seat vibration sensor is arranged on the surface of the cylinder body, and the exhaust seat vibration sensor is fixed by using a special fixture; the axial displacement of the rotor adopts an indirect measurement mode, a non-contact photoelectric sensor is selected and arranged near a motor shaft and used for collecting rotating speed information, a pressure sensor is arranged at an exhaust port, and the axial displacement of the rotor is estimated indirectly by monitoring the rotating speed of the screw air compressor and the pressure change of the exhaust port and combining the mechanical characteristics of the screw air compressor; the oil inlet pressure sensor is arranged on an oil inlet pipeline, and is connected by using a flange, the measuring range is 0-10bar, and the precision is +/-0.5%; the range of the exhaust port pressure sensor is the same as that of the oil inlet pressure sensor, and the exhaust port pressure sensor is arranged on an exhaust port pipeline and is connected by using a flange; the exhaust port temperature sensor is arranged near the exhaust port, and is connected by screw threads, the measuring range is 0-200 ℃, and the precision is +/-1 ℃. The high-performance data acquisition card is adopted, the input of a plurality of sensors is supported, the sampling frequency is at least 100Hz, the data transmission is in a wired mode, the sensors are connected to the data acquisition card through special cables, so that the stability and the instantaneity of the data transmission are ensured, and the data acquisition software has the functions of real-time display, storage and derivation. Based on the Python programming environment, a time series prediction model is built using a deep learning framework TensorFlow.
The data of each sensor, including vibration data, rotation speed data, pressure data and temperature data, are collected in real time through the data acquisition card, and the data acquisition frequency is set to be 1 time per second, so that the real-time performance and the continuity of the data are ensured. And processing vibration data by adopting a median filtering algorithm, removing high-frequency noise, removing abnormal values exceeding a threshold value by using moving average filtering processing temperature and pressure data, and converting the data to the same scale by using a Z-score standardization method to carry out standardization processing on the data. And extracting the characteristics of vibration peak value, main frequency, frequency band energy, pressure change rate, temperature change rate and the like, and indirectly estimating the approximate value of the axial displacement of the rotor by using the rotating speed and pressure data. And analyzing the correlation among the features by using the pearson correlation coefficient, selecting one of the feature pairs with extremely high correlation (such as the correlation coefficient of two features is close to 1) as a representative, removing the other redundant feature, and taking the screened feature as a target feature.
The rules for feature fusion are as follows: the characteristics of category type or numerical type and low dimensionality are directly spliced together to form new fusion characteristics; the feature dimension is high and has correlation, the Principal Component Analysis (PCA) is used for dimension reduction fusion, and the principal component is selected as the fused feature. According to the fusion rule, fusing the target features to generate new fusion features, and integrating the fused features with time stamps, device IDs and the like to form a complete training data set.
The data from the past 10 minutes was taken as a fixed time window size and the data within each window was taken as a training sample. The long-short-term memory network is trained by using a training data set, wherein the learning rate is set to 0.005, the initial iteration number is set to 150, the batch processing size is set to 64, the number of neurons of an initial hidden layer is set to 3 times of the number of input features by using a long-short-term memory network time sequence prediction algorithm.
The method comprises the steps of deploying a trained long-period memory network model into a monitoring system, collecting multi-source monitoring data of a screw air compressor in real time by the monitoring system, preprocessing the data collected in real time, extracting features from the preprocessed data, fusing the features, and inputting the fused features into the long-period memory network model for fault time sequence prediction.
The above embodiments do not limit the scope of the present application. It will be apparent to those skilled in the art that various modifications, combinations, and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of the present application. In some cases, the acts or steps recited in the present application may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims (4)

1. The screw air compressor fault prediction method based on the multi-source data fusion is characterized by comprising the following steps of:
Acquiring multi-source data, wherein the multi-source data is acquired by a sensor arranged on a screw air compressor and comprises exhaust seat vibration data, cylinder body vibration data, rotor axial displacement data, oil inlet pressure data, exhaust port pressure data and exhaust port temperature data;
preprocessing the multi-source data, and extracting monitoring data characteristics of the multi-source data;
Based on the source of the multi-source data, carrying out feature correlation evaluation by combining the monitored data features, and screening target features;
performing feature fusion according to the target features, determining a multi-source feature fusion rule, and constructing a training data fusion channel, wherein the training data fusion channel is used for performing feature fusion on multi-source data to generate a training data set;
Based on the training data set, constructing a time sequence training data set and training a time sequence prediction model, wherein the time sequence prediction model is used for predicting fault occurrence time information of the screw air compressor;
Deploying the trained time sequence prediction model into a monitoring system, and performing fault time sequence prediction according to the multisource monitoring data acquired in real time;
Wherein, based on the source of the multisource data, combining the monitored data features to perform feature correlation evaluation, screening target features includes:
according to the source of the multi-source data, a monitoring data characteristic pair is established, wherein the monitoring data characteristic pair consists of two monitoring data characteristics with different data sources and the same acquisition time;
calculating the correlation coefficient of the monitoring data feature pair;
screening and combining multiple collinearity data according to the correlation coefficient to obtain the target feature;
the method comprises the steps of carrying out feature fusion according to the target features, determining a multi-source feature fusion rule, and constructing a training data fusion channel, and comprises the following steps:
According to the monitoring data characteristics of the exhaust seat vibration data, the cylinder body vibration data, the rotor axial displacement data, the oil inlet pressure data, the exhaust port pressure data and the exhaust port temperature data, configuring data characteristic classes, wherein the data characteristic classes comprise data directions and data fluctuation amplitudes;
classifying the target features based on the data feature classes to construct a multi-feature class set;
based on the correlation coefficient, carrying out simultaneous screening on the multi-feature class set according to a preset screening threshold value to determine a fusion feature group, wherein the fusion feature group comprises two or more multi-feature classes;
taking the fusion relation of a plurality of different fusion feature groups as the multi-source feature fusion rule;
Constructing a multi-channel fusion module according to the multi-source feature fusion rule, integrating a screw air compressor state acquisition channel, and constructing the training data fusion channel;
Wherein constructing a time-series training data set based on the training data set comprises:
Clustering the training data sets according to the fusion feature groups to obtain training data sets of the fusion feature groups, wherein the training data sets comprise monitoring time identifiers;
Performing fault sample screening according to the state information of the screw air compressor in the training data set to obtain a target training data set;
the time sequence training data sets are constructed by performing time sequence arrangement on the fusion characteristic sets based on the monitoring time identification of the target training data set, wherein the time sequence training data sets comprise a plurality of groups of training data with time sequence relation, and the plurality of groups of training data have the same fusion characteristic set and screw air compressor fault information;
Judging whether the time sequence training data set reaches a preset number, and when the time sequence training data set does not reach the preset number, performing variation derivatization or simulation test based on the time sequence training data set, and expanding the time sequence training data set;
wherein the training time sequence prediction model comprises:
determining a time sequence module framework, wherein the time sequence module framework is a hidden Markov model;
training the time sequence module frame by using the time sequence training data set, taking the fault information of the screw air compressor as a target state, determining the probability of each time point transferring to the target state, and training and converging by using a forward-backward algorithm to obtain model parameters;
And obtaining the time sequence prediction model based on the model parameters.
2. The screw air compressor fault prediction method based on multi-source data fusion as claimed in claim 1, wherein preprocessing the multi-source data to extract monitoring data characteristics of the multi-source data comprises:
carrying out normalization processing on the multi-source data, and carrying out multi-source data synchronous alignment according to the monitoring time to obtain preprocessed multi-source data;
mapping the associated twin network branches and the monitoring sensors, and inputting monitoring data corresponding to the monitoring sensors into the twin network branches according to the mapping association relation to extract data characteristics;
And respectively obtaining the monitoring data characteristics of each sensor, and combining the monitoring data characteristics of the multi-source data to obtain the monitoring data characteristics of the multi-source data.
3. The method for predicting faults of a screw air compressor based on multi-source data fusion as claimed in claim 2, wherein the mapping association twin network branch and monitoring sensor comprises:
constructing a feature extraction network model;
performing network replication according to the configuration quantity based on the feature extraction network model to construct a plurality of network branches, wherein the network branches have the same network structure and share weights to construct a twin network model;
and constructing a connecting channel according to the sensors arranged on the screw air compressor, and establishing a mapping association relation between the monitoring sensor and each network branch of the twin network model.
4. The method for predicting faults of the screw air compressor based on multi-source data fusion as claimed in claim 1, wherein the fault time sequence prediction is performed according to the multi-source monitoring data collected in real time, and the method comprises the following steps:
Acquiring target characteristics according to the multisource monitoring data acquired in real time;
matching the target features with a multisource feature fusion rule, and obtaining fusion features through a training data fusion channel;
Carrying out state probability prediction on the fusion characteristics through the time sequence prediction model to obtain fault probability and corresponding time nodes;
Acquiring a trusted coefficient of a time node, and determining a self-adaptive early warning strategy based on the trusted coefficient and the fault probability;
Generating a fault time sequence early warning result according to the self-adaptive early warning strategy, the fault probability and the corresponding time node.
CN202410805946.9A 2024-06-21 2024-06-21 Screw air compressor fault prediction method based on multi-source data fusion Active CN118378195B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410805946.9A CN118378195B (en) 2024-06-21 2024-06-21 Screw air compressor fault prediction method based on multi-source data fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410805946.9A CN118378195B (en) 2024-06-21 2024-06-21 Screw air compressor fault prediction method based on multi-source data fusion

Publications (2)

Publication Number Publication Date
CN118378195A CN118378195A (en) 2024-07-23
CN118378195B true CN118378195B (en) 2024-09-17

Family

ID=91902110

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410805946.9A Active CN118378195B (en) 2024-06-21 2024-06-21 Screw air compressor fault prediction method based on multi-source data fusion

Country Status (1)

Country Link
CN (1) CN118378195B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118536048B (en) * 2024-07-24 2024-11-15 安徽斯派迪电气技术有限公司 Switchgear insulation status monitoring and management system
CN118981678A (en) * 2024-07-31 2024-11-19 北京向衡科技发展有限公司 Optoelectronic device fault detection platform based on multi-source data fusion
CN119147103A (en) * 2024-11-18 2024-12-17 浙江华科同安监控技术有限公司 Generator rotor temperature monitoring method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114399066A (en) * 2022-01-15 2022-04-26 中国矿业大学(北京) Mechanical equipment predictability maintenance system and maintenance method based on weak supervision learning
CN116561698A (en) * 2023-03-31 2023-08-08 北京航空航天大学 Multi-source data fusion human-computer interaction task online acceleration method and system

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109086804B (en) * 2018-07-12 2022-04-22 中石化石油机械股份有限公司 An early fault prediction method for hydraulic equipment based on the fusion of multi-source condition monitoring information and reliability features
CN111947928B (en) * 2020-08-10 2021-11-05 山东大学 Multi-source information fusion bearing fault prediction system and method
CN112465063B (en) * 2020-12-11 2023-05-23 中国矿业大学 Coal gangue identification method in top coal caving process based on multi-sensing information fusion
CN115014445B (en) * 2022-08-10 2022-11-15 中国农业大学 A multi-dimensional panoramic perception monitoring method, system and device for a smart fishery
CN116187508A (en) * 2022-12-07 2023-05-30 沈阳嘉越电力科技有限公司 Wind turbine generator system fault prediction method integrating knowledge graph and deep learning
CN117972547B (en) * 2024-03-26 2024-06-14 华电电力科学研究院有限公司 Fault early warning method, device and medium for key components of wind turbine generator

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114399066A (en) * 2022-01-15 2022-04-26 中国矿业大学(北京) Mechanical equipment predictability maintenance system and maintenance method based on weak supervision learning
CN116561698A (en) * 2023-03-31 2023-08-08 北京航空航天大学 Multi-source data fusion human-computer interaction task online acceleration method and system

Also Published As

Publication number Publication date
CN118378195A (en) 2024-07-23

Similar Documents

Publication Publication Date Title
CN118378195B (en) Screw air compressor fault prediction method based on multi-source data fusion
Li et al. Data alignments in machinery remaining useful life prediction using deep adversarial neural networks
CN111504676B (en) Equipment fault diagnosis method, device and system based on multi-source monitoring data fusion
CN115034248A (en) Automatic diagnostic method, system and storage medium for equipment
Li et al. Multi-sensor data-driven remaining useful life prediction of semi-observable systems
CN113469060A (en) Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method
CN118568471B (en) Intelligent power distribution station operation fault prediction method and system
US20240069539A1 (en) Sensor-agnostic mechanical machine fault identification
Huang et al. Improved trajectory similarity-based approach for turbofan engine prognostics
JP2004531815A (en) Diagnostic system and method for predictive condition monitoring
Mi et al. A synthetic feature processing method for remaining useful life prediction of rolling bearings
Mathew et al. Regression kernel for prognostics with support vector machines
Son et al. Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge
US20220004163A1 (en) Apparatus for predicting equipment damage
Baptista et al. Classification prognostics approaches in aviation
CN116108346B (en) A lifelong learning method for incremental fault diagnosis of bearings based on generative feature replay
CN110503133A (en) A fault prediction method for centrifugal compressors based on deep learning
Melendez et al. Self-supervised multi-stage estimation of remaining useful life for electric drive units
Wang et al. Degradation pattern identification and remaining useful life prediction for mechanical equipment using SKF-EN
Kozlenko et al. SOFTWARE IMPLEMENTED FAULT DIAGNOSIS OF NATURAL GAS PUMPING UNIT BASED ON FEEDFORWARD NEURAL NETWORK.
Guerroum et al. Machine learning for the predictive maintenance of a Jaw Crusher in the mining industry
Randrianandraina et al. Wind turbine generator bearing anomaly detection and explanation using rrcf approach
Dhiman et al. Enhancing wind turbine reliability through proactive high speed bearing prognosis based on adaptive threshold and gated recurrent unit networks
CN118503884B (en) Equipment state identification method, equipment and medium
CN118609601B (en) A method and system for identifying device operation status based on voiceprint information

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant