CN116106016A - Rolling bearing fault diagnosis method based on joint learning convolutional neural network - Google Patents
Rolling bearing fault diagnosis method based on joint learning convolutional neural network Download PDFInfo
- Publication number
- CN116106016A CN116106016A CN202310261004.4A CN202310261004A CN116106016A CN 116106016 A CN116106016 A CN 116106016A CN 202310261004 A CN202310261004 A CN 202310261004A CN 116106016 A CN116106016 A CN 116106016A
- Authority
- CN
- China
- Prior art keywords
- convolution
- fault
- layer
- convolutional neural
- neural network
- 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.)
- Pending
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 51
- 238000003745 diagnosis Methods 0.000 title claims abstract description 38
- 238000005096 rolling process Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000007246 mechanism Effects 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims description 43
- 230000001133 acceleration Effects 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 28
- 230000004913 activation Effects 0.000 claims description 24
- 238000005070 sampling Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 230000036541 health Effects 0.000 abstract description 4
- 238000013507 mapping Methods 0.000 abstract 1
- 238000002474 experimental method Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
Landscapes
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- General Physics & Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a rolling bearing fault diagnosis method based on a joint learning convolutional neural network, which comprises the steps of firstly mapping an input signal to a high-dimensional characteristic space through a joint characteristic coding network, and fully coding representative characteristics and fault related characteristics of the signal in the high-dimensional characteristic space through convolution; and then, the signal denoising task and the fault diagnosis task are finished simultaneously based on the attention mechanism coding network, the fault classification network and the decoder network, so that the health state of a mechanical system can be accurately predicted, and the fault diagnosis performance is improved.
Description
Technical Field
The invention belongs to the technical field of mechanical fault diagnosis, and particularly relates to a rolling bearing fault diagnosis method based on a joint learning convolutional neural network.
Background
Rolling bearings are widely used as core components in rotating machinery systems in large machinery such as wind turbines, high speed trains, aerospace equipment. The failure of the rolling bearing is light, which causes the shutdown of mechanical equipment and the stagnation of industrial production, and the heavy failure causes serious safety problems and irrecoverable casualties. Therefore, in order to ensure the normal operation of the mechanical system, the real-time monitoring of the health state of the rolling bearing is of great significance.
Currently, deep learning technology is widely accepted in the field of fault diagnosis due to its excellent feature learning ability and automatic decision making ability, and particularly convolutional neural networks, which are favored by researchers because of their advantages in parameter sharing and nonlinear feature learning ability. Although deep learning algorithms have achieved significant success in fault diagnosis, the use of the powerful feature learning capabilities of convolutional neural networks performs well in smooth working environments, there is still a problem that requires researchers to face, i.e., most mechanical systems typically operate in harsh environments, and the monitored signals are very vulnerable to contamination by environmental noise. When the signal has very strong noise, the conventional common signal diagnosis method is difficult to obtain an ideal diagnosis result. In addition, the noise conditions are constantly changing, and the existing methods are only suitable for training and testing in known noise environments and cannot be suitable for diagnosis in unknown noise conditions.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a rolling bearing fault diagnosis method based on a joint learning convolutional neural network, which can improve the fault diagnosis performance of the rolling bearing under the conditions of strong noise environment and variable load and simultaneously realize the denoising task and the fault diagnosis task of signals.
In order to achieve the aim of the invention, the rolling bearing fault diagnosis method based on the joint learning convolutional neural network is characterized by comprising the following steps of:
(1) Acquiring an acceleration vibration signal;
collecting acceleration vibration signals of the rolling bearing in the mth running state under the kth fault category, and recording the acceleration vibration signals as
K represents the total number of fault class numbers, n=1, 2, …, N is the total number of sampling points, m=1, 2, …, M represents the total number of running state numbers of the rolling bearing;
(2) Standardized processing of acceleration vibration signals;
wherein ,
is->
Normalized acceleration vibration signal, +.>
Is->
Average value of all sampling point data in (a), +.>
Is->
Standard deviation of all sampling point data in the database;
all acceleration vibration signals after standardized processing are formed into a training sample set:
(3) Constructing a joint learning convolutional neural network model;
the joint learning convolutional neural network model includes: a joint feature encoding network, a dual-attention mechanism encoding network, a fault classification network, and a decoder network;
the combined characteristic coding network consists of four serially connected convolution layers C1-C4, each attention mechanism coding network consists of four serially connected convolution+attention modules CA 1-CA 4, the output of the convolution layer C1 is used as the input of the convolution layer C2 and the convolution+attention module CA1, the output of the convolution+attention module CA1 and the output of the convolution layer C2 are added to be used as the input of the convolution+attention module CA2, and the like; finally, the output of the convolution+attention module CA4 of one attention mechanism coding network is used as the input of a fault classification network, so that the fault category is determined; taking the output of the convolution+attention module CA4 of the other attention mechanism coding network as the input of the decoder network, thereby completing denoising of the input signal;
(4) Training a joint learning convolutional neural network model;
(4.1) setting the maximum iteration number as the EPOCH, and initializing the current iteration number epoch=1; giving the expected model training error as tau;
(4.2) taking the sample set X as a training set, and setting training samples in the training sample set
From->
Extracting l sample data->
As training data of a single batch, l < N;
then, in the epoch-th iteration, the data is trained
The method comprises the steps of inputting an input signal to a convolution layer C1 of a joint feature coding network, extracting local features of the input signal through the convolution layer C1, inputting the extracted local features to a convolution+attention module CA1 and the convolution layer C2, enhancing weights of interested channels in the local features through the convolution+attention module CA1 and carrying out weighted average on the weights of the interested channels and feature information of other channels, and rolling the local featuresThe lamination C2 performs further feature extraction on the local features extracted by the convolution layer C1, then sums the output of the convolution+attention module CA1 and the output of the convolution layer C2 to serve as the input of the convolution+attention module CA2, the output of the convolution layer C2 serves as the input of the convolution layer C3, and so on;
the output of the convolution+attention module CA4 is used as the input of a convolution layer C5 in a fault classification network, the local features extracted by the convolution layer C5 are transmitted to a GAP calculation module, so that the probability that each sample data in an input signal corresponds to each fault is calculated, and then the fault category corresponding to the maximum probability is used as a prediction result;
the local characteristics of the output of the convolution+attention module CA4 in the other attention mechanism coding network sequentially pass through deconvolution layers D1-D5, and the local characteristics are subjected to signal reduction through five deconvolution layers, so that denoised signals are obtained;
calculating a loss function value
wherein ,
represents the τ sample data in the training data +.>
True probability value of the corresponding fault class, +.>
Represents the τ sample data in the training data +.>
Predictive probability value for the corresponding fault class, +.>
Represents the τ sample data in the training data +.>
Denoising the data value;
judging whether the current iteration times epoch=EPOH or loss is smaller than tau, and if yes, stopping iterative training to obtain a trained joint learning convolutional neural network model; otherwise, updating the joint learning convolutional neural network model by the loss value loss through a back propagation algorithm, and then performing the next training;
(5) Fault classification and denoising of acceleration vibration signals;
and acquiring acceleration vibration signals of the rolling bearing under a certain fault category, and intercepting input data with the length of l to be input into a joint learning convolutional neural network model after the acceleration vibration signals are subjected to standardized processing, so that the corresponding fault category is output.
The invention aims at realizing the following steps:
according to the rolling bearing fault diagnosis method based on the joint learning convolutional neural network, an input signal is mapped to a high-dimensional feature space through a joint feature coding network, and in the high-dimensional feature space, representative features and fault related features of the signal are fully coded through convolution; and then, the signal denoising task and the fault diagnosis task are finished simultaneously based on the attention mechanism coding network, the fault classification network and the decoder network, so that the health state of a mechanical system can be accurately predicted, and the fault diagnosis performance is improved.
Meanwhile, the rolling bearing fault diagnosis method based on the joint learning convolutional neural network has the following beneficial effects:
(1) The invention can effectively restore the signal polluted by noise into a purer signal by means of strong learning ability of the joint learning convolutional neural network, and also can carry out Gaussian denoising and fault diagnosis under the unknown noise condition;
(2) According to the invention, the vibration signal denoising task and the fault diagnosis task are integrated into an end-to-end network frame for the first time, the frame enables the fault diagnosis task to obtain good noise robustness by sharing network parameters and characteristics, and the denoising task can obtain fault information specific to the fault diagnosis task, so that a better denoising effect is obtained for the specific fault signal;
drawings
FIG. 1 is a flow chart of a rolling bearing fault diagnosis method based on a joint learning convolutional neural network;
FIG. 2 is a schematic diagram of a network framework for a joint learning convolutional neural network;
Detailed Description
The following description of the embodiments of the invention is presented in conjunction with the accompanying drawings to provide a better understanding of the invention to those skilled in the art. It is to be expressly noted that in the description below, detailed descriptions of known functions and designs are omitted here as perhaps obscuring the present invention.
Examples
FIG. 1 is a flow chart of a rolling bearing fault diagnosis method based on a joint learning convolutional neural network.
In this embodiment, as shown in fig. 1, the rolling bearing fault diagnosis method based on the joint learning convolutional neural network of the present invention includes the following steps:
s1, acquiring an acceleration vibration signal;
collecting acceleration vibration signals of the rolling bearing in the mth running state under the kth fault category, and recording the acceleration vibration signals as
K represents the total number of fault class numbers, n=1, 2, …, N is the total number of sampling points, m=1, 2, …, M represents the total number of running state numbers of the rolling bearing; in this embodiment, the rolling bearing operates in different harsh environments, and the acquired acceleration vibration signal contains environmental noise.
S2, standardized processing of acceleration vibration signals;
wherein ,
is->
Normalized acceleration vibration signal, +.>
Is->
Average value of all sampling point data in (a), +.>
Is->
Standard deviation of all sampling point data in the database;
all acceleration vibration signals after standardized processing are formed into a training sample set:
s3, constructing a joint learning convolutional neural network model;
as shown in fig. 2, the joint learning convolutional neural network model includes: a joint feature encoding network, a dual-attention mechanism encoding network, a fault classification network, and a decoder network;
the combined characteristic coding network consists of four serially connected convolution layers C1-C4, and each convolution layer adopts an activation function leakage ReLU; the convolution kernel size of the convolution layer C1 is set to 16×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the convolution layer C2 is set to 9×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the convolution layer C3 is set to 6×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the convolution layer C4 is set to 3×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5.
As shown in fig. 2, the dual-attention mechanism coding network comprises an upper branch and a lower branch, the structures of the upper branch and the lower branch are identical, the upper branch is used for fault classification, and the lower branch is used for signal denoising, and each of the upper branch and the lower branch consists of four convolution+attention modules CA 1-CA 4 connected in series; :
the parameter settings of the convolution layers in the convolution+attention modules CA1 to CA4 are identical, specifically: the convolution kernel size of the convolution layer is set to 1×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5.
The decoder network comprises 5 deconvolution layers D1-D5 which are connected in series, and each deconvolution layer adopts an activation function leakage ReLU; the convolution kernel size of the deconvolution layer D1 is set to 3×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D2 is set to 6×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D3 is set to 9×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D4 is set to 12×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D5 is set to 1×1, the step size Stride is set to 1, and the parameter r of the activation function leak ReLU is set to 0.5.
In this embodiment, the input signal is input through the convolution layer C1, the output signal of the convolution layer C1 is used as the input of the convolution layer C2 and the convolution+attention module CA1, the output of the convolution+attention module CA1 is added to the output of the convolution layer C2 to be used as the input of the convolution+attention module CA2, and so on; finally, the output of the convolution+attention module CA4 of one attention mechanism coding network is used as the input of a fault classification network, so that the fault category is determined; taking the output of the convolution+attention module CA4 of the other attention mechanism coding network as the input of the decoder network, thereby completing denoising of the input signal;
s4, training a joint learning convolutional neural network model;
s4.1, setting the maximum iteration number as the EPOCH, and initializing the current iteration number epoch=1; giving the expected model training error as tau;
s4.2, taking the sample set X as a training set, and setting training samples in the training sample set
From->
Extracting l sample data->
As training data of a single batch, l < N; />
Then, in the epoch-th iteration, the data is trained
The method comprises the steps of inputting an input signal to a convolution layer C1 of a joint feature coding network, extracting local features of the input signal through the convolution layer C1, inputting the extracted local features to a convolution+attention module CA1 and a convolution layer C2, enhancing weights of interesting channels in the local features and feature information of other channels through the convolution+attention module CA1, carrying out weighted average on the local features extracted by the convolution layer C1 through the convolution layer C2, then summing output of the convolution+attention module CA1 and output of the convolution layer C2 to serve as input of the convolution+attention module CA2, outputting of the convolution layer C2 to serve as input of the convolution layer C3, and the like;
the output of the convolution+attention module CA4 is used as the input of a convolution layer C5 in a fault classification network, the local features extracted by the convolution layer C5 are transmitted to a GAP calculation module, so that the probability that each sample data in an input signal corresponds to each fault is calculated, and then the fault category corresponding to the maximum probability is used as a prediction result;
the local characteristics of the output of the convolution+attention module CA4 in the other attention mechanism coding network sequentially pass through deconvolution layers D1-D5, and the local characteristics are subjected to signal reduction through five deconvolution layers, so that denoised signals are obtained;
calculating a loss function value
wherein ,
represents the τ sample data in the training data +.>
True probability value of the corresponding fault class, +.>
Represents the τ sample data in the training data +.>
Predictive probability value for the corresponding fault class, +.>
Represents the τ sample data in the training data +.>
Denoising the data value;
judging whether the current iteration times epoch=EPOH or loss is smaller than tau, and if yes, stopping iterative training to obtain a trained joint learning convolutional neural network model; otherwise, updating the joint learning convolutional neural network model by the loss value loss through a back propagation algorithm, and then performing the next training;
s5, fault classification and denoising of the acceleration vibration signals;
and acquiring acceleration vibration signals of the rolling bearing under a certain fault category, and intercepting input data with the length of l to be input into a joint learning convolutional neural network model after the acceleration vibration signals are subjected to standardized processing, so that the corresponding fault category is output.
In order to better illustrate the technical effects of the invention, the invention is experimentally verified by using a sub-experiment included in a specific embodiment. In the experimental verification, a rolling bearing test bed is adopted to simulate the working process of the rolling bearing. The rolling bearing fault diagnosis test bed adopted in the embodiment comprises a driving motor, a belt transmission system, a vertical loading device, a transverse loading device, two fan motors and a control system. The vertical and side load loading devices are designed to simulate the axial and side loads carried by a rolling bearing. The two fan motors may generate wind in the opposite direction to the running direction of the rolling bearing. By means of two accelerometers it is ensured that vibrations in both the horizontal and vertical direction of the rolling bearing can be detected, and the sampling frequency of the signal is set to 5120Hz.
In this example, 10 typical fault conditions and 1 health condition were selected, and the specific description is given in table 1.
TABLE 1 11 bearing states used in the examples
In each fault state, four operating speeds were designed: 60km/h, 90km/h, 120km/h and 150km/h, four different vertical loads: 56kN, 146kN, 236kN and 272kN. In addition, in order to better simulate the complex working condition environment of the high-speed train, gaussian white noise with different signal-to-noise ratios (SNR) is added into the original signal. Experiments of three groups of noise signals with different SNR (-6 dB, 0dB and 6 dB) are set in the experiment verification, and the strong, medium and weak noise working conditions of the rolling bearing are simulated respectively. When less noise is added, its effect on the vibration signal is small. However, when a large amount of noise is added, the original waveform of the vibration signal is completely destroyed, resulting in illegibility. Since the resulting vibration signal is a very long time series, we use a sliding segmentation method to obtain more training samples. The step size of the sliding partition is set to 256, and the length of each sample is set to 2048, so that each sample contains one complete periodic signal. After sliding segmentation, a total of 128874 training samples and 41258 test samples were obtained.
In the experimental verification, the joint learning convolutional neural network model provided by the invention is realized in the environments of a deep learning framework Keras and Python 3.6, and is trained and tested on a server with a GTX 2080 of 8G video memory. During the training process, the size of each batch was set to 256 and adam optimizer was employed to optimize the network parameters with a learning rate set to 0.0001.
First, the experiment verifies that the joint learning architecture is at SNR dB The performance in the case of = -6dB and comparison analysis was performed with the results of training with F-Task-CNN and D-Task-CNN alone, as shown in table 2.
TABLE 2 evaluation results of the Performance of the three methods
| Performance evaluation index | F-Task-CNN | D-Task-CNN | JL-CNN |
| ACC | 0.715 0.002 | N/A | 0.838 0.001 |
| SNR | N/A | 2.348 0.003 | 2.789 0.005 |
| MSE | N/A | 0.582 0.001 | 0.526 0.001 |
According to the experimental results, the joint learning framework can improve diagnosis and denoising performances. The joint learning neural network improves the diagnostic accuracy by 12.3% compared to F-Task-CNN, which indicates that with the help of SD-Task, the network can obtain more valuable information, and the joint learning convolutional neural network performs better than D-Task-CNN for SD-Task measured by SNR and MSE. In summary, the joint learning framework is more suitable for fault diagnosis of the rolling bearing.
Then, the present experiment further compares the joint learning convolutional neural network with 3 known fault diagnosis methods and 4 known denoising methods under the noise conditions of-6 dB, -3dB and 0dB, respectively, and the results are shown in table 3. It can be seen that: the joint learning convolutional neural Network JL-Network has the best performance in both error diagnosis and denoising. For the error diagnosis task, compared with WDCNN, the accuracy of the joint learning convolutional neural network is respectively improved by 10%,6.1% and 2.6% under three types of noise, which means that the invention has excellent error diagnosis performance under the condition of strong noise. For the denoising task, the performance of the joint learning convolutional neural network is superior to that of Wavelet and EMD of two traditional denoising methods and DAE and CAE of other two deep learning denoising methods.
TABLE 3 experimental results of different methods under three noise conditions
The complexity of the joint learning model was analyzed at the end of the experiment and compared with other models, and the training time for 100epochs and the test time required to predict one batch size (256 samples) were recorded, respectively, and the results are shown in table 4. It can be seen that: since the joint learning convolutional neural network needs to process two tasks simultaneously, it has two different branches and a complex network structure. Their training time is nearly identical compared to other deep learning models. However, due to the complexity of the joint learning neural network, the joint learning neural network needs longer testing time, but is only a little more time than other models, and can meet the requirements of real situations.
TABLE 4 time results for different methods
While the foregoing describes illustrative embodiments of the present invention to facilitate an understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, but is to be construed as protected by the accompanying claims insofar as various changes are within the spirit and scope of the present invention as defined and defined by the appended claims.
Claims (4)
1. The rolling bearing fault diagnosis method based on the joint learning convolutional neural network is characterized by comprising the following steps of:
(1) Acquiring an acceleration vibration signal;
collecting acceleration vibration signals of the rolling bearing in the mth running state under the kth fault category, and recording the acceleration vibration signals as
K represents the total number of fault class numbers, n=1, 2,..n, N is the total number of sampling points, m=1, 2,..m, M represents the total number of running state numbers of the rolling bearing;
(2) Standardized processing of acceleration vibration signals;
wherein ,
is->
Normalized acceleration vibration signal, +.>
Is->
Average value of all sampling point data in (a), +.>
Is->
Standard deviation of all sampling point data in the database;
all acceleration vibration signals after standardized processing are formed into a training sample set:
(3) Constructing a joint learning convolutional neural network model;
the joint learning convolutional neural network model includes: a joint feature encoding network, a dual-attention mechanism encoding network, a fault classification network, and a decoder network;
the combined characteristic coding network consists of four serially connected convolution layers C1-C4, each attention mechanism coding network consists of four serially connected convolution+attention modules CA 1-CA 4, the output of the convolution layer C1 is used as the input of the convolution layer C2 and the convolution+attention module CA1, the output of the convolution+attention module CA1 and the output of the convolution layer C2 are added to be used as the input of the convolution+attention module CA2, and the like; finally, the output of the convolution+attention module CA4 of one attention mechanism coding network is used as the input of a fault classification network, so that the fault category is determined; taking the output of the convolution+attention module CA4 of the other attention mechanism coding network as the input of the decoder network, thereby completing denoising of the input signal;
(4) Training a joint learning convolutional neural network model;
(4.1) setting the maximum iteration number as the EPOCH, and initializing the current iteration number epoch=1; giving the expected model training error as tau;
(4.2) taking the sample set X as a training set, and setting training samples in the training sample set
From->
Extracting l sample data->
As training data of a single batch, l < N;
then, in the epoch-th iteration, the data is trained
The local characteristics of the input signal are extracted through the convolution layer C1, and the extracted local characteristics are inputThe method comprises the steps of inputting the local characteristics into a convolution+attention module CA1 and a convolution layer C2, enhancing the weight of an interested channel in the local characteristics through the convolution+attention module CA1 and carrying out weighted average on the local characteristics extracted by the convolution layer C1 and the characteristic information of other channels, carrying out further characteristic extraction on the local characteristics extracted by the convolution layer C2, then summing the output of the convolution+attention module CA1 and the output of the convolution layer C2 to be used as the input of the convolution+attention module CA2, and the output of the convolution layer C2 is used as the input of the convolution layer C3, and so on;
the output of the convolution+attention module CA4 is used as the input of a convolution layer C5 in a fault classification network, the local features extracted by the convolution layer C5 are transmitted to a GAP calculation module, so that the probability that each sample data in an input signal corresponds to each fault is calculated, and then the fault category corresponding to the maximum probability is used as a prediction result;
the local characteristics of the output of the convolution+attention module CA4 in the other attention mechanism coding network sequentially pass through deconvolution layers D1-D5, and the local characteristics are subjected to signal reduction through five deconvolution layers, so that denoised signals are obtained;
calculating a loss function value
wherein ,
represents the τ sample data in the training data +.>
The true probability value for the corresponding fault class,
represents the τ sample data in the training data +.>
Predictive probability value for the corresponding fault class, +.>
Represents the τ sample data in the training data +.>
Denoising the data value;
judging whether the current iteration times epoch=EPOH or loss is smaller than tau, and if yes, stopping iterative training to obtain a trained joint learning convolutional neural network model; otherwise, updating the joint learning convolutional neural network model by the loss value loss through a back propagation algorithm, and then performing the next training;
(5) Fault classification and denoising of acceleration vibration signals;
and acquiring acceleration vibration signals of the rolling bearing under a certain fault category, and intercepting input data with the length of l to be input into a joint learning convolutional neural network model after the acceleration vibration signals are subjected to standardized processing, so that the corresponding fault category is output.
2. The rolling bearing fault diagnosis method based on the joint learning convolutional neural network according to claim 1, wherein the joint feature coding network comprises 4 layers of convolutional layers, and each layer of convolutional layer adopts an activation function leakage ReLU; the convolution kernel size of the convolution layer C1 is set to 16×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the convolution layer C2 is set to 9×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the convolution layer C3 is set to 6×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the convolution layer C4 is set to 3×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5.
3. The rolling bearing fault diagnosis method based on the joint learning convolutional neural network according to claim 1, wherein the dual-attention mechanism coding network comprises an upper branch and a lower branch, the structures of the upper branch and the lower branch are identical, the upper branch is used for fault classification, and the lower branch is used for signal denoising, and the upper branch and the lower branch are respectively composed of four serially connected convolutional+attention modules CA 1-CA 4;
the parameter settings of the convolution layers in the convolution+attention modules CA1 to CA4 are identical, specifically: the convolution kernel size of the convolution layer is set to 1×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5.
4. The rolling bearing fault diagnosis method based on the joint learning convolutional neural network according to claim 1, wherein the decoder network comprises 5 deconvolution layers D1-D5 connected in series, and each deconvolution layer adopts an activation function leakage ReLU; the convolution kernel size of the deconvolution layer D1 is set to 3×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D2 is set to 6×1, the step size is set to 2, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D3 is set to 9×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D4 is set to 12×1, the step size is set to 4, and the parameter r of the activation function leak ReLU is set to 0.5; the convolution kernel size of the deconvolution layer D5 is set to 1×1, the step size Stride is set to 1, and the parameter r of the activation function leak ReLU is set to 0.5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310261004.4A CN116106016A (en) | 2023-03-17 | 2023-03-17 | Rolling bearing fault diagnosis method based on joint learning convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310261004.4A CN116106016A (en) | 2023-03-17 | 2023-03-17 | Rolling bearing fault diagnosis method based on joint learning convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116106016A true CN116106016A (en) | 2023-05-12 |
Family
ID=86267428
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310261004.4A Pending CN116106016A (en) | 2023-03-17 | 2023-03-17 | Rolling bearing fault diagnosis method based on joint learning convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116106016A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118394030A (en) * | 2024-03-29 | 2024-07-26 | 重庆赛力斯凤凰智创科技有限公司 | Vehicle fault diagnosis method, device, computer equipment and storage medium |
| CN118549132A (en) * | 2024-06-14 | 2024-08-27 | 兰州理工大学 | Rolling bearing fault diagnosis method based on high-frequency multi-scale cascade network |
-
2023
- 2023-03-17 CN CN202310261004.4A patent/CN116106016A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118394030A (en) * | 2024-03-29 | 2024-07-26 | 重庆赛力斯凤凰智创科技有限公司 | Vehicle fault diagnosis method, device, computer equipment and storage medium |
| CN118549132A (en) * | 2024-06-14 | 2024-08-27 | 兰州理工大学 | Rolling bearing fault diagnosis method based on high-frequency multi-scale cascade network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110361176B (en) | Intelligent fault diagnosis method based on multitask feature sharing neural network | |
| CN112906644B (en) | Mechanical fault intelligent diagnosis method based on deep migration learning | |
| CN108334948B (en) | Mechanical bearing fault diagnosis technology based on wide residual error network learning model | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN110516305B (en) | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model | |
| CN107657250B (en) | Bearing fault detection and positioning method and detection and positioning model implementation system and method | |
| CN109299705A (en) | Rotary machinery fault diagnosis method based on one-dimensional depth residual error convolutional neural networks | |
| CN116106016A (en) | Rolling bearing fault diagnosis method based on joint learning convolutional neural network | |
| CN110672343A (en) | Rotary machine fault diagnosis method based on multi-attention convolutional neural network | |
| CN110991295B (en) | Self-adaptive fault diagnosis method based on one-dimensional convolutional neural network | |
| CN113505655A (en) | Bearing fault intelligent diagnosis method for digital twin system | |
| CN116358871B (en) | Rolling bearing weak signal composite fault diagnosis method based on graph rolling network | |
| CN111680788A (en) | Equipment fault diagnosis method based on deep learning | |
| CN113409213B (en) | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump | |
| CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
| CN111766513B (en) | Capsule network-based variable-working-condition multi-fault diagnosis method for three-phase induction motor | |
| CN116256174A (en) | Noise robust bearing fault diagnosis method based on interpretable multilayer wavelet drive | |
| CN114048787B (en) | Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model | |
| CN114861349B (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| CN114548153B (en) | Planetary gear box fault diagnosis method based on residual error-capsule network | |
| CN113312719B (en) | Rotary machine fault diagnosis method based on class unbalance weight cross entropy | |
| CN113111752B (en) | Rolling mill fault diagnosis method for sample imbalance enhanced extended depth confidence network | |
| Hu et al. | A Masked One-Dimensional Convolutional Autoencoder for Bearing Fault Diagnosis Based on Digital Twin Enabled Industrial Internet of Things | |
| CN113887586A (en) | FFCNN-SVM transfer learning fault diagnosis method based on feature fusion under small sample | |
| CN117315328B (en) | Gear fault diagnosis method for improving convolutional neural network under attention mechanism |
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 |