CN116007937A - Intelligent fault diagnosis method and device for mechanical equipment transmission part - Google Patents
Intelligent fault diagnosis method and device for mechanical equipment transmission part Download PDFInfo
- Publication number
- CN116007937A CN116007937A CN202211575101.2A CN202211575101A CN116007937A CN 116007937 A CN116007937 A CN 116007937A CN 202211575101 A CN202211575101 A CN 202211575101A CN 116007937 A CN116007937 A CN 116007937A
- Authority
- CN
- China
- Prior art keywords
- self
- layer
- network
- fault diagnosis
- model
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000003745 diagnosis Methods 0.000 title claims abstract description 80
- 230000005540 biological transmission Effects 0.000 title claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 57
- 230000007246 mechanism Effects 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims description 34
- 230000006870 function Effects 0.000 claims description 27
- 230000008569 process Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 20
- 239000013598 vector Substances 0.000 claims description 18
- 230000000873 masking effect Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 12
- 230000036541 health Effects 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000003862 health status Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 238000009499 grossing Methods 0.000 claims description 2
- 238000013519 translation Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 abstract description 5
- 230000008859 change Effects 0.000 abstract description 3
- 238000012423 maintenance Methods 0.000 abstract description 3
- 230000015556 catabolic process Effects 0.000 abstract 1
- 238000006731 degradation reaction Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 11
- 238000013135 deep learning Methods 0.000 description 6
- 239000000284 extract Substances 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 150000001875 compounds Chemical class 0.000 description 5
- 238000005096 rolling process Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000012633 leachable Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000009347 mechanical transmission Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to the technical field of intelligent operation and maintenance of mechanical equipment, in particular to an intelligent fault diagnosis method and device for a transmission part of mechanical equipment. According to the invention, vibration signals under different working conditions are collected as training samples, the problem of performance degradation of a model caused by working condition change is avoided, and a self-supervision pre-training network is established, so that an easily-obtained label-free sample training network is fully utilized, more effective characteristics can be extracted from the network, and dependence on label samples is reduced; in addition, through the encoder model and the decoder model based on the self-attention mechanism, more comprehensive global features are extracted, redundant features are restrained, effective features are enhanced, pre-enhancement of input data through preprocessing is not needed, and diagnosis efficiency is improved.
Description
Technical Field
The invention relates to the technical field of intelligent operation and maintenance of mechanical equipment, in particular to an intelligent fault diagnosis method and device for a transmission part of mechanical equipment.
Background
The mechanical equipment is the most main carrier in industrial production, and plays an important role in ensuring the quality of products, improving the production efficiency and creating economic benefits. The transmission part of the mechanical equipment is a core part of the mechanical equipment and plays key roles of bearing load, transmitting power and the like. However, mechanical transmission components often operate in harsh operating environments and require prolonged uninterrupted service, with serious consequences upon failure. The device maintenance time can be reasonably planned by timely monitoring and diagnosing the transmission parts of the mechanical device, so that the economic benefit is increased, and catastrophic results are avoided. Therefore, the research of the intelligent diagnosis method for the mechanical equipment transmission part is significant for improving the stability and reliability of the mechanical equipment transmission part.
In recent years, with the development of industrial big data and artificial intelligence technology, deep learning is widely applied to intelligent fault diagnosis methods of mechanical equipment transmission parts. By constructing a deep neural network, a deep confidence network, a cyclic neural network, a convolutional neural network and other multi-layer deep structures, the intelligent fault diagnosis model can approximately map complex functional relations between signals and corresponding fault types, and therefore accurate diagnosis results are obtained. Compared with the traditional method, the intelligent fault diagnosis method based on deep learning has the advantages of no need of data preprocessing, automatic feature extraction, capability of processing big data and the like, so that the intelligent fault diagnosis method based on deep learning is more and more widely studied.
Although the intelligent diagnosis method based on deep learning has the advantages, the intelligent diagnosis method based on deep learning has some defects which prevent the intelligent diagnosis method from being applied to engineering, on one hand, the intelligent diagnosis method based on deep learning needs to use a large number of label samples in the process of training a network model, but is unrealistic in actual engineering, because expert knowledge and engineering experience are needed for judging the type of faults, a large amount of manpower and material resources are needed for labeling the samples, and the network model needs to be trained from the beginning when facing a new diagnosis task or a new data set; on the other hand, most of the existing methods are insufficient in capability of extracting high-level features, the features are pre-enhanced by means of pre-processing, which reduces the efficiency of the whole diagnosis process and the diagnosis performance is reduced when it processes the raw vibration data. Second, existing deep learning models typically extract features from the local but cannot determine whether they are redundant features or fault type related valid features. Finally, the existing method generally sets training data and test data under the same working condition, and when the working condition is changed, the performance of the model is inevitably reduced, so that the practicability of the method is reduced. These shortcomings will lead to the fact that the existing intelligent fault diagnosis method based on self-supervision learning is not comprehensive in learned characteristics, and the generalization capability and the practicability of the diagnosis model are poor.
Disclosure of Invention
Therefore, the invention aims to solve the technical problems of insufficient comprehensive learned characteristics, poor generalization performance and high dependence on label samples in the prior art.
In order to solve the technical problems, the invention provides an intelligent fault diagnosis method for a transmission part of mechanical equipment, which comprises the following steps:
collecting vibration signals of a mechanical equipment transmission part under different working conditions, intercepting the vibration signals for a plurality of times according to preset data point lengths to obtain a plurality of sample data, dividing the plurality of sample data according to preset proportions, calibrating one part of the sample data according to fault types to be used as a labeled data set, and the other part of the sample data is used as a non-labeled data set;
establishing a self-supervision pre-training network by constructing a random mask module, an encoder model and a decoder model based on a self-attention mechanism, masking and reconstructing samples in the unlabeled dataset by using the self-supervision pre-training network, wherein the reconstruction comprises an encoding process and a decoding process, calculating reconstruction loss between signals before and after reconstruction, optimizing the self-supervision pre-training network by taking minimized reconstruction loss as an objective function, and updating network parameters of the self-supervision pre-training network;
establishing a fine tuning network by constructing the encoder model and the classifier model, transferring parameters corresponding to the encoder model in the optimized self-supervision pre-training network to the encoder model of the fine tuning network, classifying samples in the labeled data set by using the fine tuning network, calculating classification loss, optimizing the fine tuning network by taking the minimized classification loss as an objective function, and updating network parameters of the fine tuning network to obtain an intelligent fault diagnosis model of the mechanical equipment transmission part;
and inputting the vibration signal to be tested into the intelligent fault diagnosis model of the mechanical equipment transmission part to obtain the fault type of the mechanical equipment transmission part.
Preferably, said masking samples in said unlabeled dataset with said self-supervised pretraining network includes:
for input samples [ X ] 1 ；X 2 ；...；X n ]Performing random masking operation to obtain a mask output matrix X m ＝[X 1 ；X 2 ；...；X n ][c 1 ；c 2 ；...；c n ] T Wherein n is the preset data point length, x 1 To x n Representing n row vectors of the input matrix c 1 To c n Representing randomly generated mask vectors, half of which have values of 1 and the rest of which are 0, T representing the transpose;
removing the row vector with 0 in the mask output matrix to obtain a final mask output matrix X M ＝X m -X 0 。
Preferably, the reconstructed encoding process includes:
adding one row of category token vectors for aggregating the classification information to the final mask output matrix, and superposing position codes for indicating the position information to obtain a first coding characteristic
wherein ,
representing a category token vector, ">
Representing a merge matrix->
Representing a position code;
the first coding feature is subjected to layer normalization processing to obtain a second coding feature
Wherein, gamma is a scaling parameter, beta is a translation parameter, and a standardized value +.>
Average of first coding feature in last dimension
Variance->
N dim Dimension of the last dimension of the first code feature;
processing the second coding feature through multi-stage stacked multi-head self-attention layers and multi-layer perceptron layers to obtain a target coding feature
The processing procedure of the d-level multi-head self-attention layer and the d-level multi-layer perceptron layer is as follows:
calculating the self-attention value of the d-1 level multi-layer sensor layer output
The output of the d-1 level multi-layer sensor layer is added with>
Residual connection to obtain the output of the d-level multi-head self-attention layer>
Wherein depth is the number of levels of the multi-headed self-focusing layer and multi-layered sensor layer stack, when d=1, the +.>
Self-attention value for the second coding feature, is->
For the first encoding feature;
outputting the d-level multi-head self-attention layer
And first linear layer weight +.>
Product of (c) and first linear layer deviation b 1 And, through GeLU excitationAfter the living function processing, the second linear layer weight is +.>
Multiplied by the second linear layer deviation b 2 Adding, finally, with the output of the d-level multi-head self-attention layer->
Residual connection to obtain the output +.>
Preferably, the decoding process of the reconstruction includes:
restoring the target coding feature to the vector y removed at the random masking block d ＝y+X 0 And the third linear layer is used for reducing the dimension of the first linear layer to obtain a first decoding characteristic y d ′＝y d *w d +b d, wherein ,
is a third linear layer weight matrix, b d Is a third linear layer bias;
and processing the first decoding characteristic through the multi-stage stacked multi-head self-attention layer and the multi-layer perceptron layer, and recovering dimension through a fourth linear layer to obtain a reconstruction characteristic.
Preferably, the reconstruction loss is:
wherein ,nB Representing the size of the batch and,
representing the reconstruction feature, x i Representing the mask output matrix.
Preferably, the optimization method of the objective function is a back propagation and gradient descent method.
Preferably, the classifier model is a Softmax classifier, probability values of each fault belonging to different categories are output through the Softmax classifier, and a calculation formula of the corresponding fault probability values is as follows:
wherein ,xi ' represents the ith input feature, y i Output probability value, n, representing corresponding class s Indicating the total number of fault categories.
Preferably, the classification loss is:
wherein ,nb Indicating batch size, n s ' total number of categories representing health status, 1 {. Cndot. } represents a function at y j Return 1 when k, otherwise return 0, y j A real label representing a sample is provided,
representing the predicted probability of all health states.
The invention also provides an intelligent fault diagnosis device for the mechanical equipment transmission part, which comprises the following components:
the data set construction module is used for collecting vibration signals of the mechanical equipment transmission part under different working conditions, intercepting the vibration signals for a plurality of times according to preset data point lengths to obtain a plurality of sample data, dividing the plurality of sample data according to preset proportions, calibrating one part of the sample data according to fault types to be used as a labeled data set, and the other part of the sample data is used as a non-labeled data set;
the self-supervision pre-training network construction module is used for constructing a self-supervision pre-training network by constructing a random mask module, a self-attention mechanism-based encoder model and a self-attention mechanism-based decoder model, masking and reconstructing samples in the unlabeled dataset by utilizing the self-supervision pre-training network, wherein the reconstruction comprises an encoding process and a decoding process, calculating reconstruction loss between signals before and after reconstruction, optimizing the self-supervision pre-training network by taking the minimized reconstruction loss as an objective function, and updating network parameters of the self-supervision pre-training network;
the prediction model construction module is used for building a fine tuning network by constructing the encoder model and the classifier model, transferring parameters corresponding to the encoder model in the optimized self-supervision pre-training network to the encoder model of the fine tuning network, classifying samples in the labeled data set by using the fine tuning network, calculating classification loss, optimizing the fine tuning network by taking the minimized classification loss as an objective function, and updating network parameters of the fine tuning network to obtain an intelligent fault diagnosis model of the mechanical equipment transmission part;
and the fault diagnosis module is used for inputting the vibration signal to be tested into the intelligent fault diagnosis model of the mechanical equipment transmission part to obtain the fault type of the mechanical equipment transmission part.
The invention also provides a computer readable storage medium, wherein the computer readable storage medium is stored with a computer program, and the computer program realizes the steps of the intelligent fault diagnosis method for the mechanical equipment transmission part when being executed by a processor.
Compared with the prior art, the technical scheme of the invention has the following advantages:
according to the intelligent fault diagnosis method for the mechanical equipment transmission part, vibration signals under different working conditions are collected to serve as training samples, the problem that the performance of a model is reduced due to the change of the working conditions is avoided, a self-supervision pre-training network is built, an easily-obtained label-free sample training network is fully utilized, the network can extract more effective characteristics, and dependence on label samples is reduced; in addition, through the encoder model and the decoder model based on the self-attention mechanism, more comprehensive global features are extracted, redundant features are restrained, effective features are enhanced, pre-enhancement of input data through preprocessing is not needed, and diagnosis efficiency is improved. The intelligent fault diagnosis method for the mechanical equipment transmission part can complete the diagnosis task of the intelligent fault diagnosis model under the new working condition, has good generalization performance, is suitable for identifying the health state of the mechanical equipment transmission part on site in real time, provides a reliable and convenient tool for the intelligent diagnosis method based on deep self-supervision learning, and has important field significance and wide application prospect.
Drawings
In order that the invention may be more readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings, in which:
FIG. 1 is a flow chart of an intelligent fault diagnosis method for a mechanical equipment transmission part;
FIG. 2 is a schematic diagram of a self-supervising pre-training network of the present invention;
FIG. 3 is a schematic diagram of a fine-tuning network of the present invention;
FIG. 4 is a schematic diagram of an encoder model in a self-supervising pre-training network and a fine tuning network according to the present invention;
fig. 5 is a block diagram of a mechanical device transmission component intelligent fault diagnosis device according to an embodiment of the present invention.
Detailed Description
The invention provides an intelligent fault diagnosis method and device for a mechanical equipment transmission part and a computer storage medium, the learned characteristics are more comprehensive, and the generalization capability and practicability of a diagnosis model are improved.
In order to better understand the aspects of the present invention, the present invention will be described in further detail with reference to the accompanying drawings and detailed description. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, fig. 1 is a flowchart illustrating an implementation method of intelligent fault diagnosis for a transmission component of a mechanical device according to the present invention; the specific operation steps are as follows:
s101, collecting vibration signals of a mechanical equipment transmission part under different working conditions, intercepting the vibration signals for a plurality of times according to preset data point lengths to obtain a plurality of sample data, dividing the plurality of sample data according to preset proportions, calibrating one part of the sample data according to fault types to obtain a labeled data set, and using the other part of the sample data as a non-labeled data set;
the method comprises the steps of collecting vibration signals of a mechanical equipment transmission part under various working conditions by using a sensor, intercepting the vibration signals according to a certain data point length to obtain a large number of sample sets X, and selecting samples under one working condition as test data sets
The samples under other working conditions are training data sets, samples with known fault types in the training data sets are calibrated according to fault types, category labels Y are set, and the training data sets are divided into label-free data sets +.>
There is tag dataset->
wherein nT ，n U ，n L Representing the number of samples in the test dataset, the unlabeled dataset and the labeled dataset, respectively. Wherein the data sets are all derived from a laboratory bench.
As shown in fig. 2:
s102, establishing a self-supervision pre-training network by constructing a random mask module, an encoder model and a decoder model based on a self-attention mechanism, masking and reconstructing samples in the unlabeled dataset by using the self-supervision pre-training network, wherein the reconstruction comprises an encoding process and a decoding process, calculating reconstruction loss between signals before and after reconstruction, optimizing the self-supervision pre-training network by taking the minimized reconstruction loss as an objective function, and updating network parameters of the self-supervision pre-training network;
the random masking module masks and rejects the input label-free signal, the encoder model extracts the representation from the reserved original signal, the decoder model reconstructs the signal according to the extracted representation, and the encoder model and the decoder model are both based on a self-attention mechanism, so that the problems of incomplete characteristics and lack of pertinence extracted by the model can be effectively avoided.
The encoder model includes a multi-headed self-attention layer, a layer normalization layer, a multi-layered perceptron layer, and a residual connection.
The decoder model includes a linear layer, a multi-headed self-attention layer, a layer normalization layer, a multi-layered perceptron layer, and a residual connection.
As shown in fig. 3:
s103, establishing a fine tuning network by constructing the encoder model and the classifier model, transferring parameters corresponding to the encoder model in the optimized self-supervision pre-training network to the encoder model of the fine tuning network, classifying samples in the labeled data set by using the fine tuning network, calculating classification loss, optimizing the fine tuning network by taking minimized classification loss as an objective function, and updating network parameters of the fine tuning network to obtain an intelligent fault diagnosis model of the mechanical equipment transmission part;
the building of the fine tuning network comprises an encoder model and a classifier model, wherein the encoder model is consistent with the encoder model in the pre-training network, the repetition is omitted, the corresponding parameters of the encoder model in the pre-training network are migrated, the characteristic extraction capacity is inherited, and then a small amount of tagged data is utilized to train the network to adapt to the fault diagnosis task.
The optimization method of the objective function is a back propagation and gradient descent method.
The classifier model is a Softmax classifier, probability values of each fault belonging to different categories are output through the Softmax classifier, and a corresponding calculation formula of the fault probability values is as follows:
wherein ,xi ' represents the ith input feature, y i Output probability value, n, representing corresponding class s Indicating the total number of fault categories.
The classification loss is:
wherein ,nb Indicating batch size, n s ' total number of categories representing health status, 1 {. Cndot. } represents a function at y j Return 1 when k, otherwise return 0, y j A real label representing a sample is provided,
representing the predicted probability of all health states.
S104, inputting the vibration signal to be tested into the intelligent fault diagnosis model of the mechanical equipment transmission part to obtain the fault type of the mechanical equipment transmission part.
According to the intelligent fault diagnosis method for the mechanical equipment transmission part, vibration signals under different working conditions are collected to serve as training samples, the problem that the performance of a model is reduced due to the change of the working conditions is avoided, a self-supervision pre-training network is built, an easily-obtained label-free sample training network is fully utilized, the network can extract more effective characteristics, and dependence on label samples is reduced; in addition, through the encoder model and the decoder model based on the self-attention mechanism, more comprehensive global features are extracted, redundant features are restrained, effective features are enhanced, pre-enhancement of input data through preprocessing is not needed, and diagnosis efficiency is improved. The intelligent fault diagnosis method for the mechanical equipment transmission part can complete the diagnosis task of the intelligent fault diagnosis model under the new working condition, has good generalization performance, is suitable for identifying the health state of the mechanical equipment transmission part on site in real time, provides a reliable and convenient tool for the intelligent diagnosis method based on deep self-supervision learning, and has important field significance and wide application prospect.
Based on the above embodiments, the present embodiment further describes in detail step S102:
the masking samples in the unlabeled dataset with the self-supervised pretraining network includes:
for input samples [ X ] 1 ；X 2 ；...；X n ]Performing random masking operation to obtain a mask output matrix X m ＝[X 1 ；X 2 ；...；X n ][c 1 ；c 2 ；...；c n ] T Wherein n is the preset data point length, x 1 To x n Representing n row vectors of the input matrix c 1 To c n Representing randomly generated mask vectors, half of which have values of 1 and the rest of which are 0, T representing the transpose;
converting the row vector with 0 in the mask output matrix into a trainable parameter and temporarily removing to obtain a final mask output matrix X M ＝X m -X 0 。
Mask matrix for final output
The input is to the encoder model for encoding characterization.
The reconstructing of the samples in the unlabeled dataset using the self-supervised pretraining network specifically includes:
as shown in fig. 4, the reconstructed encoding process includes:
adding one row of category token vectors for aggregating the classification information to the final mask output matrix, and superposing position codes for indicating the position information to obtain a first coding characteristic
wherein ,
representing a category token vector, ">
Representing a merge matrix->
Representing a position code;
and carrying out layer standardization processing on the first coding feature to obtain a second coding feature X, wherein the mathematical expression is as follows:
wherein μ is the average of the input in the last dimension, σ 2 As a function of the variance of the values,
for normalized values, ε is a smoothing factor, the output is infinity when the variance is prevented from being 0, and the final batch normalized output is introduced with a scaling parameter γ and a shifting parameter β that are updated with a back-propagation algorithm during network training to further improve the numerical output stability, N dim Dimension of the last dimension of the first code feature;
processing the second coding feature through multi-stage stacked multi-head self-attention layers and multi-layer perceptron layers to obtain a target coding feature
Wherein, the d-level multi-head self-attention layer and the d-level multi-head self-attention layerThe processing procedure of the d-level multilayer sensor layer is as follows:
calculating the self-attention value of the d-1 level multi-layer sensor layer output
The output of the d-1 level multi-layer sensor layer is added with>
Residual connection to prevent gradient extinction and obtain output of d-level multi-head self-attention layer
d=1, 2, 3.., depth, where depth is the number of levels of the multi-headed self-attention layer and multi-layered sensor layer stack, when d=1, the _on_is>
Self-attention value for the second coding feature, is->
For the first encoding feature;
outputting the d-level multi-head self-attention layer
And first linear layer weight +.>
Product of (c) and first linear layer deviation b 1 After GeLU activation function processing, and the second linear layer weight +.>
Multiplied by the second linear layer deviation b 2 Adding, finally, with the output of the d-level multi-head self-attention layer->
Residual connection is carried out to obtain the output of the d-level multi-layer perceptron layer so as to realize the multiplexingHybrid nonlinear mapping
d＝1,2,3,...,depth。
The mathematical expression for calculating the multi-headed self-attention value is:
Q＝X·W q ,K＝X·W k ,V＝X·W v
head i ＝Attention(Q·W i Q ,K·W i K ,V·W i V )
y MSA ′＝Concat(head 1 ,...head i )W O
wherein ,
representing a query matrix for subsequent matching with the key matrix,/->
Representing a key matrix for being matched by the query matrix, < >>
Representing a matrix of values representing the input +.>
Information extracted from the Chinese herb, jersey>
and />
Respectively, a respective leachable transformation matrix, < >>
For scaling factorThe sub-matching process calculates the product of the query matrix and the key matrix, divides the scaling factor and passes through a Softmax function, the calculation result is the correlation of the query matrix and the key matrix, and the calculated result is then used as the weight of the value matrix to be multiplied by the value matrix; acquisition from different feature subspaces through multiple sets of different transformation matrices
′
Information, combining multiple groups of self-attention calculation results to obtain a multi-head self-attention value y MSA 。
The decoding process of the reconstruction includes:
restoring the target coding feature to the vector y removed at the random masking block d ＝y+X 0 ，
And the third linear layer is used for reducing the dimension to obtain a first decoding characteristic y d ′＝y d *w d +b d ，/>
wherein ,
is a third linear layer weight matrix, b d Is a third linear layer bias;
and processing the first decoding characteristic through the multi-stage stacked multi-head self-attention layer and the multi-layer perceptron layer, and recovering dimension through a fourth linear layer to obtain a reconstruction characteristic.
The reconstruction loss is as follows:
wherein ,nB Representing the size of the batch and,
representing the reconstruction feature, x i Representing the mask output matrix.
Based on the above examples, the present example verifies the effectiveness of the present invention in one specific experiment:
the data used in the experiment come from a bearing fault simulation experiment table of Suzhou university, and the bearing fault data set has seven health states, including 0.2mm fault of the inner ring of the bearing, 0.2mm fault of the rolling body, 0.2mm fault of the outer ring and compound fault types (including compound fault of the inner ring and the rolling body, compound fault of the inner ring and the outer ring, compound fault of the outer ring and the rolling body, compound fault of the inner ring, the outer ring and the rolling body). The data for each health state were collected at 800rpm and four different loads (0 KN,0.8KN,1.6KN,2.5 KN). Each state of health for each condition contained 200 samples, each sample containing 1024 data points. The data under the load of 2.5KN is selected as a test data set, the data under the rest loads are training test sets, the number of partial sample labels under the load of 1.6KN in the training test sets is given, the number is 5, 10, 15 and 20 of each health state, and the rest are label-free samples. The details of the samples used in the experiments are shown in table 1.
Table 1 number of samples for each health status
To verify the effectiveness of the proposed invention, the experimental case performs diagnostic tasks with four different few label samples, the number of label samples being 5, 10, 15 and 20 for each health state, respectively, 2.5%,5%,7.5% and 10% of the total number of samples. In order to verify the effectiveness of the intelligent fault diagnosis method and system for the mechanical equipment transmission part provided by the invention, five other advanced intelligent diagnosis methods based on self-supervision learning are used as comparison, and the intelligent fault diagnosis method comprises a self-attention encoder (SAE), a convolution self-encoder (CAE), a context self-encoder (CE), a Denoising Convolutional Neural Network (DCNN) and a denoising self-encoder (DAE). The parameters during training of each network model are set as follows: the initial learning rate was 0.001, the batch size was 128, and the number of iterations was 50.
Compared with other self-supervision learning intelligent diagnosis methods, the classification accuracy of the method has the best effect on diagnosis tasks under the condition of four different few-label samples, and the superiority of the method is shown, and the comparison result is shown in table 2.
Table 2 accuracy of each method at each diagnostic task
Aiming at the problems that the existing intelligent fault diagnosis method seriously depends on a label sample and the characteristics extracted by a diagnosis network are not comprehensive and lack of pertinence, the invention takes a mechanical equipment transmission part as a research object, extracts supervision information from a label-free signal by a self-supervision learning method, enhances the performance of effective characterization extracted by the network, builds a network model by using a self-attention mechanism, extracts comprehensive and pertinence characteristics, and improves the generalization performance of the model.
Referring to fig. 5, fig. 5 is a block diagram of a mechanical device transmission component intelligent fault diagnosis apparatus according to an embodiment of the present invention; the specific apparatus may include:
the data set construction module 100 is configured to collect vibration signals of a mechanical device transmission component under different working conditions, intercept the vibration signals for multiple times according to a preset data point length to obtain multiple sample data, divide the multiple sample data according to a preset proportion, calibrate one part of the sample data according to a fault type to be used as a labeled data set, and use the other part of the sample data as an unlabeled data set;
a self-supervision pre-training network construction module 200, configured to construct a self-supervision pre-training network by constructing a random mask module, a self-attention mechanism-based encoder model, and a decoder model, mask and reconstruct samples in the unlabeled dataset using the self-supervision pre-training network, wherein the reconstruction includes an encoding process and a decoding process, calculate reconstruction loss between signals before and after reconstruction, and optimize the self-supervision pre-training network with a minimized reconstruction loss as an objective function, and update network parameters of the self-supervision pre-training network;
the prediction model construction module 300 is configured to build a fine tuning network by constructing the encoder model and the classifier model, migrate parameters corresponding to the encoder model in the optimized self-supervision pre-training network to the encoder model of the fine tuning network, classify the samples in the labeled dataset by using the fine tuning network, calculate classification loss, optimize the fine tuning network by taking minimized classification loss as an objective function, and update network parameters of the fine tuning network to obtain an intelligent fault diagnosis model of the mechanical equipment transmission part;
the fault diagnosis module 400 is configured to input a vibration signal to be tested into the intelligent fault diagnosis model of the mechanical device transmission component, so as to obtain a fault class of the mechanical device transmission component.
The intelligent fault diagnosis device for a mechanical device driving part of the present embodiment is used to implement the foregoing intelligent fault diagnosis method for a mechanical device driving part, so that the specific implementation of the intelligent fault diagnosis device for a mechanical device driving part can be found in the foregoing example portions of the intelligent fault diagnosis method for a mechanical device driving part, for example, the data set building module 100, the self-monitoring pre-training network building module 200, the prediction model building module 300, and the fault diagnosis module 400 are respectively used to implement steps S101, S102, S103, and S104 in the foregoing intelligent fault diagnosis method for a mechanical device driving part, so that the specific implementation thereof can refer to the description of the corresponding examples of each portion and will not be repeated herein.
The specific embodiment of the invention also provides a computer readable storage medium, wherein the computer readable storage medium is stored with a computer program, and the computer program realizes the steps of the intelligent fault diagnosis method for the mechanical equipment transmission part when being executed by a processor.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations and modifications of the present invention will be apparent to those of ordinary skill in the art in light of the foregoing description. It is not necessary here nor is it exhaustive of all embodiments. While still being apparent from variations or modifications that may be made by those skilled in the art are within the scope of the invention.
Claims (10)
1. An intelligent fault diagnosis method for a transmission part of a mechanical device is characterized by comprising the following steps:
collecting vibration signals of a mechanical equipment transmission part under different working conditions, intercepting the vibration signals for a plurality of times according to preset data point lengths to obtain a plurality of sample data, dividing the plurality of sample data according to preset proportions, calibrating one part of the sample data according to fault types to be used as a labeled data set, and the other part of the sample data is used as a non-labeled data set;
establishing a self-supervision pre-training network by constructing a random mask module, an encoder model and a decoder model based on a self-attention mechanism, masking and reconstructing samples in the unlabeled dataset by using the self-supervision pre-training network, wherein the reconstruction comprises an encoding process and a decoding process, calculating reconstruction loss between signals before and after reconstruction, optimizing the self-supervision pre-training network by taking minimized reconstruction loss as an objective function, and updating network parameters of the self-supervision pre-training network;
establishing a fine tuning network by constructing the encoder model and the classifier model, transferring parameters corresponding to the encoder model in the optimized self-supervision pre-training network to the encoder model of the fine tuning network, classifying samples in the labeled data set by using the fine tuning network, calculating classification loss, optimizing the fine tuning network by taking the minimized classification loss as an objective function, and updating network parameters of the fine tuning network to obtain an intelligent fault diagnosis model of the mechanical equipment transmission part;
and inputting the vibration signal to be tested into the intelligent fault diagnosis model of the mechanical equipment transmission part to obtain the fault type of the mechanical equipment transmission part.
2. The method of claim 1, wherein masking samples in the unlabeled dataset with the self-supervised pre-training network comprises:
for input samples [ X ] 1 ；X 2 ；...；X n ]Performing random masking operation to obtain a mask output matrix X m ＝[X 1 ；X 2 ；...；X n ][c 1 ；c 2 ；...；c n ] T Wherein n is the preset data point length, x 1 To x n Representing n row vectors of the input matrix c 1 To c n Representing randomly generated mask vectors, half of which have values of 1 and the rest of which are 0, T representing the transpose;
removing the row vector with 0 in the mask output matrix to obtain a final mask output matrix X M ＝X m -X 0 。
3. The intelligent fault diagnosis method for a mechanical device transmission component according to claim 2, wherein the reconstructed encoding process comprises:
adding one row of category token vectors for aggregating the classification information to the final mask output matrix, and superposing position codes for indicating the position information to obtain a first coding characteristic
wherein ,
representing a category token vector, ">
Representing a merge matrix->
Representing a position code;
the first coding feature is subjected to layer normalization processing to obtain a second coding feature
Wherein, gamma is a scaling parameter, beta is a translation parameter, and a standardized value +.>
Average of first coding feature in last dimension
Variance->
Epsilon is a smoothing factor, N dim Dimension of the last dimension of the first code feature;
processing the second coding feature through multi-stage stacked multi-head self-attention layers and multi-layer perceptron layers to obtain a target coding feature
The processing procedure of the d-level multi-head self-attention layer and the d-level multi-layer perceptron layer is as follows:
calculating the self-attention value of the d-1 level multi-layer sensor layer output
The output of the d-1 level multi-layer sensor layer is added with>
Residual connection to obtain the output of the d-level multi-head self-attention layer>
Wherein depth is the number of levels of the multi-headed self-focusing layer and multi-layered sensor layer stack, when d=1, the +.>
Self-attention value for the second coding feature, is->
For the first encoding feature;
outputting the d-level multi-head self-attention layer
And first linear layer weight +.>
Product of (c) and first linear layer deviation b 1 After GeLU activation function processing, and the second linear layer weight +.>
Multiplied by the second linear layer deviation b 2 Adding, finally, with the output of the d-level multi-head self-attention layer->
Residual connection to obtain the output +.>
4. A mechanical device transmission component intelligent fault diagnosis method according to claim 3, wherein the reconstructed decoding process comprises:
restoring the target coding feature to the vector y removed at the random masking block d ＝y+X 0 And the third linear layer is used for reducing the dimension of the first linear layer to obtain a first decoding characteristic y d ′＝y d *w d +b d, wherein ,
is a third linear layer weight matrix, b d Is a third linear layer bias;
and processing the first decoding characteristic through the multi-stage stacked multi-head self-attention layer and the multi-layer perceptron layer, and recovering dimension through a fourth linear layer to obtain a reconstruction characteristic.
5. The intelligent fault diagnosis method for a mechanical equipment transmission component according to claim 4, wherein the reconstruction loss is:
wherein ,nB Representing the size of the batch and,
representing the reconstruction feature, x i Representing the mask output matrix.
6. The intelligent fault diagnosis method for a mechanical equipment transmission component according to claim 1, wherein the optimization method of the objective function is a back propagation and gradient descent method.
7. The intelligent fault diagnosis method for the mechanical equipment transmission component according to claim 1, wherein the classifier model is a Softmax classifier, probability values of each fault belonging to different categories are output through the Softmax classifier, and a calculation formula of the corresponding fault probability values is as follows:
wherein ,xi ' represents the ith input feature, y i Output probability value, n, representing corresponding class s Indicating the total number of fault categories.
8. The intelligent fault diagnosis method for a mechanical equipment transmission component according to claim 1, wherein the classification loss is:
wherein ,nb Indicating batch size, n s ' total number of categories representing health status, 1 {. Cndot. } represents a function at y j Return 1 when k, otherwise return 0, y j A real label representing a sample is provided,
representing the predicted probability of all health states.
9. An intelligent fault diagnosis device for a transmission part of a mechanical device, comprising:
the data set construction module is used for collecting vibration signals of the mechanical equipment transmission part under different working conditions, intercepting the vibration signals for a plurality of times according to preset data point lengths to obtain a plurality of sample data, dividing the plurality of sample data according to preset proportions, calibrating one part of the sample data according to fault types to be used as a labeled data set, and the other part of the sample data is used as a non-labeled data set;
the self-supervision pre-training network construction module is used for constructing a self-supervision pre-training network by constructing a random mask module, a self-attention mechanism-based encoder model and a self-attention mechanism-based decoder model, masking and reconstructing samples in the unlabeled dataset by utilizing the self-supervision pre-training network, wherein the reconstruction comprises an encoding process and a decoding process, calculating reconstruction loss between signals before and after reconstruction, optimizing the self-supervision pre-training network by taking the minimized reconstruction loss as an objective function, and updating network parameters of the self-supervision pre-training network;
the prediction model construction module is used for building a fine tuning network by constructing the encoder model and the classifier model, transferring parameters corresponding to the encoder model in the optimized self-supervision pre-training network to the encoder model of the fine tuning network, classifying samples in the labeled data set by using the fine tuning network, calculating classification loss, optimizing the fine tuning network by taking the minimized classification loss as an objective function, and updating network parameters of the fine tuning network to obtain an intelligent fault diagnosis model of the mechanical equipment transmission part;
and the fault diagnosis module is used for inputting the vibration signal to be tested into the intelligent fault diagnosis model of the mechanical equipment transmission part to obtain the fault type of the mechanical equipment transmission part.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of a method for intelligent fault diagnosis of a transmission component of a mechanical device according to any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211575101.2A CN116007937B (en) | 2022-12-08 | 2022-12-08 | Intelligent fault diagnosis method and device for mechanical equipment transmission part |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211575101.2A CN116007937B (en) | 2022-12-08 | 2022-12-08 | Intelligent fault diagnosis method and device for mechanical equipment transmission part |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116007937A true CN116007937A (en) | 2023-04-25 |
| CN116007937B CN116007937B (en) | 2024-09-06 |
Family
ID=86020785
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211575101.2A Active CN116007937B (en) | 2022-12-08 | 2022-12-08 | Intelligent fault diagnosis method and device for mechanical equipment transmission part |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116007937B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117874593A (en) * | 2023-12-11 | 2024-04-12 | 华北电力大学 | Fan bearing fault diagnosis model training method and device, electronic equipment and medium |
| CN117992862A (en) * | 2024-04-07 | 2024-05-07 | 山东大学 | Two-stage rotating mechanical equipment reliability assessment method and system based on large model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111562109A (en) * | 2020-05-19 | 2020-08-21 | 重庆邮电大学 | Deep learning state identification and diagnosis method for mechanical equipment |
| CN111916067A (en) * | 2020-07-27 | 2020-11-10 | 腾讯科技（深圳）有限公司 | Training method and device of voice recognition model, electronic equipment and storage medium |
| CN112990578A (en) * | 2021-03-16 | 2021-06-18 | 北京航空航天大学 | Industrial low-quality high-noise data modeling and product quality prediction method |
| WO2022110730A1 (en) * | 2020-11-27 | 2022-06-02 | 平安科技（深圳）有限公司 | Label-based optimization model training method, apparatus, device, and storage medium |
| CN114993677A (en) * | 2022-05-11 | 2022-09-02 | 山东大学 | Rolling bearing fault diagnosis method and system based on unbalanced small sample data |
| US20220327035A1 (en) * | 2020-06-03 | 2022-10-13 | Soochow University | Intra-class adaptation fault diagnosis method for bearing under variable working conditions |
| CN115410032A (en) * | 2022-07-26 | 2022-11-29 | 北京工业大学 | OCTA image classification structure training method based on self-supervision learning |
-
2022
- 2022-12-08 CN CN202211575101.2A patent/CN116007937B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111562109A (en) * | 2020-05-19 | 2020-08-21 | 重庆邮电大学 | Deep learning state identification and diagnosis method for mechanical equipment |
| US20220327035A1 (en) * | 2020-06-03 | 2022-10-13 | Soochow University | Intra-class adaptation fault diagnosis method for bearing under variable working conditions |
| CN111916067A (en) * | 2020-07-27 | 2020-11-10 | 腾讯科技（深圳）有限公司 | Training method and device of voice recognition model, electronic equipment and storage medium |
| WO2022110730A1 (en) * | 2020-11-27 | 2022-06-02 | 平安科技（深圳）有限公司 | Label-based optimization model training method, apparatus, device, and storage medium |
| CN112990578A (en) * | 2021-03-16 | 2021-06-18 | 北京航空航天大学 | Industrial low-quality high-noise data modeling and product quality prediction method |
| CN114993677A (en) * | 2022-05-11 | 2022-09-02 | 山东大学 | Rolling bearing fault diagnosis method and system based on unbalanced small sample data |
| CN115410032A (en) * | 2022-07-26 | 2022-11-29 | 北京工业大学 | OCTA image classification structure training method based on self-supervision learning |
Non-Patent Citations (2)
| Title |
|---|
| 岳增营 等: "基于语言模型的预训练技术研究综述", 中文信息学报, vol. 35, no. 9, 30 September 2021 (2021-09-30), pages 15 - 29 * |
| 戴俊 等: "基于生成对抗网络和自动编码器的机械系统异常检测", 仪器仪表学报, vol. 40, no. 9, 30 September 2019 (2019-09-30), pages 16 - 26 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117874593A (en) * | 2023-12-11 | 2024-04-12 | 华北电力大学 | Fan bearing fault diagnosis model training method and device, electronic equipment and medium |
| CN117992862A (en) * | 2024-04-07 | 2024-05-07 | 山东大学 | Two-stage rotating mechanical equipment reliability assessment method and system based on large model |
| CN117992862B (en) * | 2024-04-07 | 2024-07-12 | 山东大学 | Two-stage rotating mechanical equipment reliability assessment method and system based on large model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116007937B (en) | 2024-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
| CN111914873B (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN116007937B (en) | Intelligent fault diagnosis method and device for mechanical equipment transmission part | |
| CN115293280A (en) | Power equipment system anomaly detection method based on space-time feature segmentation reconstruction | |
| CN115017826B (en) | Method for predicting residual service life of equipment | |
| Nasser et al. | A hybrid of convolutional neural network and long short-term memory network approach to predictive maintenance | |
| CN110851654A (en) | Industrial equipment fault detection and classification method based on tensor data dimension reduction | |
| CN114757432A (en) | Future execution activity and time prediction method and system based on flow log and multi-task learning | |
| CN117034143B (en) | Distributed system fault diagnosis method and device based on machine learning | |
| CN114266301A (en) | Intelligent power equipment fault prediction method based on graph convolution neural network | |
| CN112947385A (en) | Aircraft fault diagnosis method and system based on improved Transformer model | |
| Xu et al. | Global attention mechanism based deep learning for remaining useful life prediction of aero-engine | |
| CN114169091A (en) | Method for establishing prediction model of residual life of engineering mechanical part and prediction method | |
| Pan et al. | A meta network pruning framework for remaining useful life prediction of rocket engine bearings with temporal distribution discrepancy | |
| CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
| CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
| Ma et al. | A collaborative central domain adaptation approach with multi-order graph embedding for bearing fault diagnosis under few-shot samples | |
| CN116227586A (en) | Meta learning fault diagnosis method and device based on depth residual error shrinkage prototype network | |
| CN117667495B (en) | Association rule and deep learning integrated application system fault prediction method | |
| CN117390407B (en) | Fault identification method, system, medium and equipment of substation equipment | |
| Liu et al. | A novel health prognosis method for system based on improved degenerated Hidden Markov model | |
| CN116383757B (en) | Bearing fault diagnosis method based on multi-scale feature fusion and migration learning | |
| Bond et al. | A hybrid learning approach to prognostics and health management applied to military ground vehicles using time-series and maintenance event data | |
| CN112560252A (en) | Prediction method for residual life of aircraft engine | |
| CN116628444A (en) | Water quality early warning method based on improved meta-learning |
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 |