CN116204781A - Rotary machine fault migration diagnosis method and system - Google Patents
Rotary machine fault migration diagnosis method and system Download PDFInfo
- Publication number
- CN116204781A CN116204781A CN202211397639.9A CN202211397639A CN116204781A CN 116204781 A CN116204781 A CN 116204781A CN 202211397639 A CN202211397639 A CN 202211397639A CN 116204781 A CN116204781 A CN 116204781A
- Authority
- CN
- China
- Prior art keywords
- domain
- migration
- data
- sample
- data set
- 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
- 238000000034 method Methods 0.000 title claims abstract description 106
- 238000013508 migration Methods 0.000 title claims abstract description 85
- 230000005012 migration Effects 0.000 title claims abstract description 85
- 238000003745 diagnosis Methods 0.000 title claims abstract description 41
- 238000012544 monitoring process Methods 0.000 claims abstract description 60
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 38
- 238000003062 neural network model Methods 0.000 claims abstract description 34
- 238000007781 pre-processing Methods 0.000 claims abstract description 23
- 238000005457 optimization Methods 0.000 claims abstract description 18
- 238000009795 derivation Methods 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 62
- 238000012549 training Methods 0.000 claims description 41
- 238000005516 engineering process Methods 0.000 claims description 25
- 230000003044 adaptive effect Effects 0.000 claims description 20
- 230000009466 transformation Effects 0.000 claims description 20
- 238000013528 artificial neural network Methods 0.000 claims description 18
- IOCMOSUFEBBJOB-UHFFFAOYSA-K diethyl-bis[10-(triethylazaniumyl)decyl]azanium triiodide Chemical compound [I-].[I-].[I-].C(C)[N+](CCCCCCCCCC[N+](CC)(CC)CC)(CCCCCCCCCC[N+](CC)(CC)CC)CC IOCMOSUFEBBJOB-UHFFFAOYSA-K 0.000 claims description 18
- 230000036541 health Effects 0.000 claims description 15
- 230000006978 adaptation Effects 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 12
- 230000002159 abnormal effect Effects 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 8
- 238000002405 diagnostic procedure Methods 0.000 claims description 5
- 230000001172 regenerating effect Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 abstract description 16
- 230000004913 activation Effects 0.000 description 14
- 238000010606 normalization Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 238000009826 distribution Methods 0.000 description 6
- 238000011176 pooling Methods 0.000 description 6
- 238000013526 transfer learning Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design 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
- 238000012545 processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241000512668 Eunectes Species 0.000 description 1
- 238000005299 abrasion Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention provides a fault migration diagnosis method and system for rotary machinery, comprising the following steps: respectively acquiring a marked monitoring data set and a non-marked monitoring data set of the rotating component under different running conditions; the same data preprocessing is carried out on the source domain data set and the target domain data set; selecting a deep migration learning algorithm and constructing a deep neural network model; inputting source domain data with labels to obtain output prediction labels; inputting source domain data and target domain data, and calculating loss between features of the two domain data; calculating the deviation between the predicted result and the real result; gradient derivation and optimization are carried out on the loss; repeating the steps to obtain a trained neural network model; and (3) preprocessing the new monitoring data through the data in the step (2) to obtain a prediction tag of the data sample. The invention realizes the integration of the self-adaptive methods in different fields on the algorithm level by constructing a self-adaptive integration framework.
Description
Technical Field
The invention relates to the technical field of mechanical state monitoring and fault diagnosis, in particular to a rotary machine fault migration diagnosis method and system, and especially relates to a rotary machine fault migration diagnosis method and system based on dynamic field self-adaptive integration.
Background
Rotary machines are widely used as one of the core devices in manufacturing. The rotary machine fault diagnosis research is developed, the safe and reliable operation of the rotary machine is ensured, and the rotary machine fault diagnosis method has great practical significance for improving the production benefit of enterprises and ensuring the safety of national economy. The traditional rotary machine fault diagnosis method is often based on a time-frequency domain analysis technology to extract a mode of typical fault characteristics, and most of the methods are complex in operation and have high professional requirements on detection personnel. In the current large background of mass monitoring data, the rapid, real-time and efficient diagnosis and analysis requirements are difficult to adapt. Intelligent diagnosis methods based on artificial intelligence techniques such as deep learning have been widely used since the advent of the artificial intelligence era. These intelligent diagnostic methods have continually made new research progress in the field of fault diagnosis by virtue of powerful feature extraction and fitting capabilities.
While these deep smart diagnostic methods have many advantages, there are two reasons that limit their application in complex real-world scenarios. First, these deep learning models require training and testing data to follow the same data distribution, but in practical applications, the collected monitoring data often encompasses different operating conditions and even different mechanical devices. In addition, the monitoring signal data does not have a fault tag, and the cost of manual labeling is quite high, so that the fault data with the tag is seriously deficient. With the continuous progress of the related technology, the transfer learning hopefully relieves the requirement of data acquisition, and provides a possible solution to the above challenges. The purpose of the migration learning is to find a method to combine the source domain containing rich information data and the target domain lacking information, and to use the knowledge learned by the source domain to apply to the related tasks in the target domain.
Patent document with publication number CN113076834B discloses a rotary machine fault information processing method, processing system, processing terminal, medium, constructing a neural network model comprising a depth feature extractor, a domain classifier and a state predictor, automatically extracting migration fault features from laboratory simulation data and rotary part monitoring data in actual engineering equipment by using the depth feature extractor through the neural network model; the difference between two data distributions is shortened by using a domain classifier, a state predictor is used, domain adaptation constraint is introduced, a fault diagnosis model based on a depth domain self-adaptive countermeasure network is formed, and intelligent fault diagnosis of the rotary machine is realized by using the model. However, the patent document still has the defect that monitoring signal data does not have a fault label, subject to the same data distribution.
Disclosure of Invention
In view of the defects in the prior art, the invention aims to provide a fault migration diagnosis method for rotary machinery.
The invention provides a fault migration diagnosis method for rotary machinery, which comprises the following steps:
step 1: respectively acquiring a marked monitoring data set and a non-marked monitoring data set of a rotating component under different running conditions to form a source domain data set and a target domain data set;
step 2: the same data preprocessing is carried out on the source domain data set and the target domain data set, so that two sample sets which can be identified by an integrated network algorithm are obtained;
step 3: selecting different deep migration learning algorithms, and constructing a deep neural network model according to an integrated migration framework;
step 4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function;
step 5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected deep migration learning algorithm;
step 6: calculating the deviation between the predicted result and the real result of different migration methods, and calculating the corresponding optimization self-adaptive factors of different deep migration learning methods in the integrated framework according to the deviation;
step 7: gradient derivation and optimization are carried out on the loss;
step 8: repeating the iteration steps 4-7 until reaching the convergence condition, stopping training to obtain a trained neural network model;
step 9: and (3) preprocessing the new monitoring data in the step (2), and inputting the data into a trained deep neural network model to obtain a predictive label of the data sample.
Preferably, in the step 1, the source domain data set is expressed as
The target domain dataset is denoted +.>
wherein ,
Monitoring data of the ith sample in the source domain dataset and corresponding health condition markers,/respectively>
For the ith sample in the target domain dataset, n is the minimum number of training samples for the batch.
Preferably, in the step 6, the deviation between the predicted result and the actual result of the different migration methods is calculated by using the a-distance.
Preferably, in the step 7, the foregoing loss is gradient derived and optimized by using a back propagation algorithm.
Preferably, the step 2 specifically includes the following steps:
step 2.1: abnormal values of the original vibration signal of the rotating part are removed, and the abnormal values are removed by using the Laida rule, and the formula is as follows:
wherein ,
mean value of signal segment, x i The value of the ith sample point in the signal sample;
Is the standard deviation, n is the total number of sampling points in the signal segment;
step 2.2: slicing the time sequence vibration signals with the abnormal values eliminated, wherein each segment comprises 4096 vibration signal sampling points, and obtaining a wavelet time-frequency pattern book through continuous wavelet transformation as the input of a network model, wherein the continuous wavelet transformation formula is as follows:
where Φ (t) is the wavelet mother function, τ is the time shift coefficient, a is the scale coefficient, and a+.0.
Preferably, in the step 3, the deep neural network model is a DDAE based neural network DDAENN;
the DDAE-based neural network DDAENN comprises a feature extractor G f Domain discriminator G d Tag predictor G l ；
The DDAE-based neural network DDAENN adopts a domain countermeasure technology and an MMD-based domain adaptation technology.
Preferably, in the step 4, the classification accuracy of the deep neural network on the source domain labeled data is trained, and the cross entropy loss is used, and the loss function is as follows:
wherein ,Wy and by Is made by a tag predictor G l Matrix vector pair obtained after linear transformation, f represents the matrix vector pair obtained by the feature extractor G f The extracted features, F (·) represents the softmax function, I { · } represents the indicator function, k is the source domain ith sample
A corresponding real sample tag;
in the step 5, for domain countermeasure technology, the loss function is expressed as:
wherein φ represents a sigmod function, W d and bd Is a pair of linear transformation matrix vectors for a domain classifier, d i A domain label representing the ith training sample, f s and ft Respectively represented by the feature extractor G f The extracted source domain and target domain features, m is the total number of samples in one training batch;
for MMD-based domain adaptation techniques, its loss function is expressed as:
wherein ,ns and nt Representing the number of batch training samples from the source domain and the target domain respectively,
and
Domain invariant feature representing two domains of depth feature extractor output +_>
Is a regenerative core Hilbert space with k characteristic cores.
Preferably, the step 6 specifically includes the following steps:
step 6.1: the A-distance after different migration technologies are calculated, and the calculation formula is as follows:
wherein ,xj Represents the j-th sample, G c (x j ) Is a classifier G c For sample x j Is provided with an output of (a),
representing an indication function, n' being two different domain data sets D s and Dt Is the sample size of err i (G c ) Data representing two different fields in the i-th field adaptive method are in classifier G c Errors in the above;
Two fields D after the use of the ith field adaptation method s and Dt A-distance between;
step 6.2: for the used i-th field self-adaptive method, respectively calculating the corresponding optimized self-adaptive factor alpha i The calculation formula is as follows:
wherein ,
a-distance parameter which is the i-th field adaptive method +.>
Sum of all A-distance parameters +.>
Is used to represent the weight of the migration technique, +.>
The weight occupied by the j-th domain self-adaptive method;
step 6.3: the obtained optimized adaptive factor alpha i Loss function of field adaptive method selected in step 5
In combination, the total loss function of the migration method is obtained, expressed as:
preferably, the step 9 specifically includes the following steps:
step 9.1: acquiring on-line monitoring data of an actual rotary machine through a sensor, preprocessing the monitoring data in the step 2, and then regularizing the monitoring data into a proper wavelet time-frequency diagram sample;
step 9.2: and (3) sequentially inputting the obtained time-frequency pattern book into the trained neural network model obtained in the step (8) to obtain a prediction label of a corresponding sample, and evaluating the running state of the current equipment according to the prediction result of the label.
The invention also provides a rotary machine fault migration diagnosis system, which comprises the following steps:
module M1: respectively acquiring a marked monitoring data set and a label-free monitoring data set of a rotating component under different running conditions to form a source domain data set and a target domain data set, wherein the monitoring data of an ith sample in the source domain data set and a corresponding health condition mark thereof are respectively the ith sample in the target domain data set, and n is the number of training samples in the minimum batch;
module M2: the same data preprocessing is carried out on the source domain data set and the target domain data set, so that two sample sets which can be identified by an integrated network algorithm are obtained;
module M3: selecting different deep migration learning algorithms, and constructing a deep neural network model according to an integrated migration framework;
module M4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function;
module M5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected deep migration learning algorithm;
module M6: calculating the deviation between the predicted result and the real result of different migration methods by using the A-distance, and calculating the corresponding optimization self-adaptive factors of different deep migration learning methods in the integrated framework according to the deviation;
module M7: carrying out gradient derivation and optimization on the loss by using a back propagation algorithm;
module M8: repeating the triggering modules M4-M7 to execute iteration until reaching convergence conditions, and stopping training to obtain a trained neural network model;
module M9: and (3) preprocessing the new monitoring data by the data of the module 2, and inputting the new monitoring data into a trained deep neural network model to obtain a prediction label of the data sample.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention integrates the ideas related to transfer learning and integrated learning, and realizes the integration of the self-adaptive methods in different fields on the algorithm level by constructing a self-adaptive integrated framework;
2. the invention realizes the integration of the self-adaptive methods in different fields at the algorithm level by introducing the optimal adaptation factor alpha, and designs a feasible neural network (DDAE Neural network, DDAENN) model based on DDAE and a fault diagnosis flow based on DDAENN based on the proposed integrated learning framework;
3. the DDAENN of the present invention includes three particular modules: the feature extractor, the domain classifier and the label self-adaptive predictor adopt two domain self-adaptive technologies, namely a domain countermeasure technology and a distribution self-adaptive technology based on the maximum mean value difference, and realize the integrated diagnosis of the two domain self-adaptive technologies.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:
FIG. 1 is a flow chart of steps of a rotary machine fault migration diagnostic method of the present invention;
FIG. 2 is a block diagram of a DDAE framework of the present invention;
FIG. 3 is a schematic flow chart of the present invention;
FIG. 4 is a schematic diagram of a comparison of raw vibration signals under different conditions;
FIG. 5 is a T-SNE profile of an output feature;
FIG. 6 is a schematic diagram of a confusion matrix of model predictors.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present invention.
Example 1:
as shown in fig. 1 to 6, the present embodiment provides a fault migration diagnosis method for a rotary machine, including the steps of:
step 1: respectively acquiring a marked monitoring data set and a non-marked monitoring data set of a rotating component under different running conditions to form a source domain data set and a target domain data set; the source domain dataset is represented as
The target domain dataset is denoted +.>
wherein ,
Monitoring data of the ith sample in the source domain dataset and corresponding health condition markers,/respectively>
For the ith sample in the target domain dataset, n is the minimum number of training samples for the batch.
Step 2: the same data preprocessing is carried out on the source domain data set and the target domain data set, so that two sample sets which can be identified by an integrated network algorithm are obtained; the step 2 specifically comprises the following steps:
step 2.1: abnormal values of the original vibration signal of the rotating part are removed, and the abnormal values are removed by using the Laida rule, and the formula is as follows:
wherein ,
mean value of signal segment, x i The value of the ith sample point in the signal sample;
is the standard deviation, n is the total number of sampling points in the signal segment;
step 2.2: slicing the time sequence vibration signals with the abnormal values eliminated, wherein each segment comprises 4096 vibration signal sampling points, and obtaining a wavelet time-frequency pattern book through continuous wavelet transformation as the input of a network model, wherein the continuous wavelet transformation formula is as follows:
where Φ (t) is the wavelet mother function, τ is the time shift coefficient, a is the scale coefficient, and a+.0.
Step 3: selecting different deep migration learning algorithms, and constructing a deep neural network model according to an integrated migration framework; the deep neural network model is a neural network DDAENN based on DDAE;
the DDAE-based neural network DDAENN comprises a feature extractor G f Domain discriminator G d Tag predictor G l ；
The DDAE-based neural network DDAENN adopts a domain countermeasure technology and an MMD-based domain adaptation technology.
Step 4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function; training the classification precision of the deep neural network on the source domain labeled data, and using cross entropy loss, wherein a loss function is as follows:
wherein ,Wy and by Is made by a tag predictor G l Matrix vector pair obtained after linear transformation, f represents the matrix vector pair obtained by the feature extractor G f The extracted features, F (·) represents the softmax function, I { · } represents the indicator function, k is the source domain ith sample
Corresponding real sample tags.
Step 5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected deep migration learning algorithm; for domain countermeasure techniques, its loss function is expressed as:
wherein φ represents a sigmod function, W d and bd Is a pair of linear transformation matrix vectors for a domain classifier, d i A domain label representing the ith training sample, f s and ft Respectively represented by the feature extractor G f Extracted fromSource domain and target domain features, m is the total number of samples in a training batch;
for MMD-based domain adaptation techniques, its loss function is expressed as:
wherein ,ns and nt Representing the number of batch training samples from the source domain and the target domain respectively,
and
Domain invariant feature representing two domains of depth feature extractor output +_>
Is a regenerative core Hilbert space with k characteristic cores.
Step 6: calculating the deviation between the predicted result and the real result of different migration methods, and calculating the corresponding optimization self-adaptive factors of different deep migration learning methods in the integrated framework according to the deviation; calculating the deviation between the predicted result and the real result of different migration methods by using the A-distance; the step 6 specifically comprises the following steps:
step 6.1: the A-distance after different migration technologies are calculated, and the calculation formula is as follows:
wherein ,xj Represents the j-th sample, G c (x j ) Is a classifier G c For sample x j Is provided with an output of (a),
representing an indication function, n' being two different domain data sets D s and Dt Is the sample size of err i (G c ) Data representing two different fields in the i-th field adaptive method are in classifier G c Errors in the above;
Two fields D after the use of the ith field adaptation method s and Dt A-distance between;
step 6.2: for the used i-th field self-adaptive method, respectively calculating the corresponding optimized self-adaptive factor alpha i The calculation formula is as follows:
wherein ,
a-distance parameter which is the i-th field adaptive method +.>
Sum of all A-distance parameters +.>
Is used to represent the weight of the migration technique, +.>
The weight occupied by the j-th domain self-adaptive method;
step 6.3: the obtained optimized adaptive factor alpha i And select in step 5Loss function of fixed domain adaptive method
In combination, the total loss function of the migration method is obtained, expressed as:
here, the
I.e.the +.appearing in step 5>
and
Step 7: gradient derivation and optimization are carried out on the loss; the loss is gradient derived and optimized using a back propagation algorithm.
Step 8: repeating the iteration steps 4-7 until reaching the convergence condition, stopping training to obtain a trained neural network model;
step 9: after preprocessing the new monitoring data in the step 2, inputting the new monitoring data into a trained deep neural network model to obtain a prediction label of the data sample; the step 9 specifically comprises the following steps:
step 9.1: acquiring on-line monitoring data of an actual rotary machine through a sensor, preprocessing the monitoring data in the step 2, and then regularizing the monitoring data into a proper wavelet time-frequency diagram sample;
step 9.2: and (3) sequentially inputting the obtained time-frequency pattern book into the trained neural network model obtained in the step (8) to obtain a prediction label of a corresponding sample, and evaluating the running state of the current equipment according to the prediction result of the label.
Example 2:
the embodiment provides a rotary machine fault migration diagnosis system, which comprises the following modules:
module M1: respectively acquiring a marked monitoring data set and a label-free monitoring data set of a rotating component under different running conditions to form a source domain data set and a target domain data set, wherein the monitoring data of an ith sample in the source domain data set and a corresponding health condition mark thereof are respectively the ith sample in the target domain data set, and n is the number of training samples in the minimum batch;
module M2: the same data preprocessing is carried out on the source domain data set and the target domain data set, so that two sample sets which can be identified by an integrated network algorithm are obtained;
module M3: selecting different deep migration learning algorithms, and constructing a deep neural network model according to an integrated migration framework;
module M4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function;
module M5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected deep migration learning algorithm;
module M6: calculating the deviation between the predicted result and the real result of different migration methods by using the A-distance, and calculating the corresponding optimization self-adaptive factors of different deep migration learning methods in the integrated framework according to the deviation;
module M7: carrying out gradient derivation and optimization on the loss by using a back propagation algorithm;
module M8: repeating the iteration of the modules 4-7 until reaching the convergence condition, stopping training to obtain a trained neural network model;
module M9: and (3) preprocessing the new monitoring data by the data of the module 2, and inputting the new monitoring data into a trained deep neural network model to obtain a prediction label of the data sample.
Example 3:
the present embodiment will be understood by those skilled in the art as more specific descriptions of embodiment 1 and embodiment 2.
The embodiment provides a rotary machine fault migration diagnosis method based on dynamic field self-adaptive integration, belongs to the technical field of machine state monitoring and fault diagnosis, and particularly relates to a deep migration integration diagnosis method for rotary machine part faults.
Aiming at the difficulties and challenges of the prior art, the embodiment provides a rotating machinery fault migration diagnosis method based on dynamic field self-adaptive integration.
The technical scheme of the embodiment is as follows: a fault migration diagnosis method of rotary machinery based on dynamic field self-adaptive integration comprises the following steps:
step 1: acquiring marked monitoring data sets of rotating parts under different operating conditions respectively
And no tag monitoring dataset->
Composing a source domain and a target domain dataset, wherein +.>
Monitoring data of the ith sample in the source domain dataset and corresponding health condition markers,/respectively>
The method comprises the steps that (1) the ith sample in a target domain data set is obtained, and n is the number of training samples in the minimum batch;
step 2: the two acquired data sets are subjected to the same data preprocessing to obtain two sample sets which can be identified by an integrated network algorithm;
step 3: selecting different depth migration algorithms, and constructing a depth neural network model according to an integrated migration framework;
step 4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function;
step 5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected transfer learning algorithm;
step 6: calculating the deviation between the predicted result and the real result of different migration methods by using the A-distance, and calculating the corresponding optimization self-adaptive factor of the different migration methods in the integrated framework according to the deviation;
step 7: carrying out gradient derivation and optimization on the loss by using a back propagation algorithm;
step 8: and repeating the steps 4, 5, 6 and 7 until reaching the convergence condition, and stopping training. Obtaining a trained neural network model;
step 9: and (3) preprocessing the new monitoring data in the step (2) and inputting the data into a trained neural network model to obtain a predictive label of the data sample.
Because rotating machine monitoring data often lacks sufficient fault data and relevant state markers, it is difficult to train an intelligent diagnostic model with better generalization and robustness. The embodiment provides a new migration learning integration framework (Dynamic Domain Adaptive Ensemble, DDAE), realizes the integration of the self-adaptive methods (Domain adaptive method, DAM) in different fields at an algorithm level by introducing an optimal adaptation factor alpha, and designs a feasible neural network (DDAE Neural network, DDAENN) model based on the DDAE and a fault diagnosis flow based on the DDAENN based on the proposed integration learning framework. DDAENN includes three particular modules: feature extractor, domain classifier and label adaptive predictor. Two domain self-adaptive technologies, namely a domain countermeasure technology and a distribution self-adaptive technology based on the maximum mean value difference are adopted, so that integrated diagnosis of the two domain self-adaptive technologies is realized. The verification of the case experiment shows that the method has higher experimental precision, and the experimental effect meets the expectations.
Example 4:
the present embodiment will be understood by those skilled in the art as more specific descriptions of embodiment 1 and embodiment 2.
The embodiment provides a rotary machine fault migration diagnosis method based on dynamic field self-adaptive integration, which comprises the following steps:
step 1: as shown in fig. 3, the marked monitoring data sets of the rotating component under different running conditions are acquired respectively
And no tag monitoring dataset->
Composing a source domain and a target domain dataset, wherein +.>
Monitoring data of the ith sample in the source domain dataset and corresponding health condition markers,/respectively>
The method comprises the steps that (1) the ith sample in a target domain data set is obtained, and n is the number of training samples in the minimum batch;
step 2: as shown in fig. 3, after the two acquired data sets are subjected to the same data preprocessing, two sample sets which can be identified by the integrated network algorithm are obtained;
further, the method comprises the following steps:
step 2-1: and removing abnormal values of the original vibration signals of the rotating component. Here, the rada rule is used to remove outliers as follows:
wherein ,
representing the average value of signal segments，x i Is the value of the i-th sample point in the signal sample,/->
Is the standard deviation, n is the total number of sampling points in the signal segment;
step 2-2: the time series vibration signal after eliminating the abnormal value is sliced, each segment contains 4096 vibration signal sampling points, then a wavelet time frequency pattern book is obtained through continuous wavelet transformation as the input of a network model, and the continuous wavelet transformation formula is as follows:
where Φ (t) is the wavelet mother function, τ is the time shift coefficient, a is the scale coefficient, and a+.0.
Step 3: selecting different depth migration algorithms, and constructing a depth neural network model according to the integrated migration framework shown in fig. 2; as shown in FIG. 3, we construct a DDAE-based neural network DDAENN consisting of three modules, namely a feature extractor G f Domain discriminator G d And tag predictor G l The network adopts two common domain self-adaption methods, namely a domain countermeasure technology and a domain self-adaption technology based on MMD, the model uses a CNN backbone network to learn useful characteristic representation and performs domain self-adaption training under a DDAE framework;
detailed parameter configuration of the network structure will be given in specific embodiments;
step 4: inputting labeled source domain data, obtaining an output prediction label through forward propagation in a constructed network, and comparing the prediction label with a real label to obtain a loss function, wherein the step is mainly used for training the classification precision of the source domain labeled data by the network, and using cross entropy loss, the loss function is as follows:
wherein ,Wy and by Is made by a tag predictor G l Matrix vector pair obtained after linear transformation, f represents the matrix vector pair obtained by the feature extractor G f The extracted features, F (·) represents the softmax function, I { · } represents the indicator function, k is the source domain ith sample
A corresponding real sample tag;
step 5: inputting source domain data and target domain data, and calculating loss between data features from two fields by using a selected transfer learning algorithm;
for domain countermeasure techniques, its loss function can be expressed as:
wherein φ represents a sigmod function, W d and bd Is a pair of linear transformation matrix vectors for a domain classifier, d i A domain label representing the ith training sample, f s and ft Respectively represented by the feature extractor G f The extracted source domain and target domain features, m is the total number of samples in one training batch;
for MMD-based domain adaptation techniques, its loss function can be expressed as:
wherein ns and nt Representing the number of batch training samples from the source domain and the target domain respectively,
and
Domain invariant feature representing two domains of depth feature extractor output, but +.>
Is a regenerated core Hilbert space with k characteristic cores, and the characteristic core k can improve the test precision of the method by using different kernel functions;
step 6: calculating the deviation between the predicted result and the real result of different migration methods by using the A-distance, and calculating the corresponding optimization self-adaptive factor of the different migration methods in the integrated framework according to the deviation;
further, the method comprises the following steps:
step 6-1: the A-distance after different migration technologies are calculated, and the calculation formula is as follows:
wherein ,xj Represents the j-th sample, G c (x j ) Is a classifier G c For sample x j Is provided with an output of (a),
representing an indication function, n' being two different domain data sets D s and Dt Is the sample size of err i (G c ) Data representing two different fields in the i-th field adaptive method are in classifier G c Error in->
Namely two fields D after the i-th field self-adaptive method is used s and Dt A-distance between;
step 6.2: for different migration technologies, respectively calculating the corresponding optimization adaptive factor alpha i The calculation formula is as follows:
wherein ,
a-distance parameter which is the i-th field adaptive method +.>
Sum of all A-distance parameters +.>
Is used to represent the weight of the migration technique, +.>
The weight occupied by the j-th domain self-adaptive method;
step 6.3: the obtained optimized adaptive factor alpha i Loss function of domain adaptation method in step 5
In combination, the total loss function of the migration method is obtained, expressed as:
step 7: carrying out gradient derivation and optimization on the loss by using a back propagation algorithm; the total loss function can be expressed as:
where λ is the regularization parameter, L y Is a mark in a training batchTraining the classification loss function of the sample. The process of back propagation derivation can be expressed as:
wherein Θc Representing network parameters to be optimized in the model;
step 8: and (5) repeating the steps 4, 5, 6 and 7 until the convergence condition is reached, and stopping training after the total Loss function Loss is reduced. Obtaining a trained neural network model;
step 9: the trained network model can be directly used in online diagnosis of the rotating machinery. Further, the method comprises the following steps:
step 9-1: acquiring on-line monitoring data of an actual rotary machine through a sensor, preprocessing the monitoring data in the step 2, and then regularizing the monitoring data into a proper wavelet time-frequency diagram sample;
step 9-2: and (3) sequentially inputting the obtained time-frequency pattern book into the trained neural network model obtained in the step (8) to obtain a prediction label of the corresponding sample. And the running state of the current equipment can be estimated according to the prediction result of the label.
Example 5:
the person skilled in the art will understand this embodiment as a more specific description of embodiment 1, embodiment 2, embodiment 4.
In this embodiment, for the problems existing in the fault diagnosis of the rotating equipment, a method for fault migration diagnosis of the rotating machinery based on adaptive integration in the dynamic field is provided, and the intelligent fault intelligent diagnosis of the gearbox under different operation conditions and loads is finally realized by carrying out migration training on labeled experimental data under a single working condition load to unlabeled gearbox data under variable working condition multiple loads.
The data set employed in this embodiment is gearbox fault simulation data derived from laboratory simulations. The gearbox test bed mainly comprises an alternating current motor, a flywheel, a speed change gearbox and a load brake for applying load. The maximum rotation speed of the test stand was 3600rpm, and the maximum output torque was 50N. The vibration signal is collected by a three-channel acceleration sensor. Four health states, including a health mode and three failure modes, were simulated in total on the test bench. Faults include pitting faults, wear faults, and fracture faults. All faults were manufactured manually. A total of 6 operating conditions were tested, including variable speed and variable load. Wherein, the working conditions (OC) 1 to 4 are constant speed loads, and OC 5 to 6 are variable speed loads. The sampling frequency is 10000Hz. The data signal acquisition in each OC lasted 30 seconds. For variable speed OCs, data was collected for a 30 second gradual rise in speed from 1200rpm to 2400 rpm. Signal data was collected for approximately 15s during the speed ramp-up phase. Stationary vibration signals for four health states are shown in fig. 4.
The method for diagnosing the fault of the target domain comprises the following steps:
step 1: and (3) carrying out data preprocessing on the original vibration signal, firstly cutting each piece of data by using a sliding time window with a fixed size, carrying out outlier removal operation, and then obtaining a wavelet time-frequency diagram sample by using a formula (2). And (3) selecting four kinds of health state data under the working condition 1 as a source domain, and setting up a source domain and a target domain data set by using the other 5 kinds of working condition data as target domains.
Step 2: according to FIG. 3, a DDAE-based neural network DDAENN is constructed, which, as described above, consists of three modules, namely a feature extractor G f Domain discriminator G d And tag predictor G l The specific configurations of the network structures are shown in table 1, table 2 and table 3:
TABLE 1 feature extractor G f
| Layer number | Name of the name | Operation of | |
| 1 | Input device | Input sample | 64×64×3 |
| 2 | |
Convolution | / |
| 3 | |
Batch standardization | / |
| 4 | |
Pooling | 4 |
| 5 | |
ReLU function activation | 0.2 |
| 6 | |
Convolution | / |
| 7 | |
Batch standardization | / |
| 8 | |
Pooling | 4 |
| 9 | |
Relu function activation | 0.2 |
| 10 | Convolutional layer 3 | Convolution | / |
| 11 | Batch normalization layer 3 | Batch standardization | / |
| 12 | Pooling layer 3 | Pooling | 16 |
| 13 | Activation layer 3 | Relu function activation | 0.2 |
| 14 | Full tie layer 3 | Tensor flattening | / |
Table 2 tag predictor G l
| Layer number | Name of the name | Operation of | |
| 1 | |
|
100 |
| 2 | |
Batch standardization | / |
| 3 | |
Relu function activation | 0.2 |
| 4 | |
Dropout | 0.3 |
| 5 | |
|
100 |
| 6 | |
Batch standardization | / |
| 7 | |
Relu function activation | 0.2 |
| 8 | Full tie layer 3 | Classification | Total category number |
Table 3 domain discriminator G d
| Layer number | Name of the name | Operation of | |
| 1 | |
|
100 |
| 2 | |
Batch standardization | / |
| 3 | |
Relu function activation | 0.2 |
| 4 | |
|
100 |
| 5 | |
Batch standardization | / |
| 6 | |
Relu function activation | 0.2 |
| 7 | Full tie layer 3 | |
2 |
Step 3: and training the network model on the training set of the source domain and the target domain according to the training method. During training, the source domain training set consists of a wavelet time-frequency pattern book and corresponding labels thereof, the target domain training set only comprises wavelet time-frequency pattern samples and does not comprise labels, and finally a trained network model is obtained. The examples were implemented in the Win10 system, in the Anaconda-based Python3.6 and Pytorch environments.
Step 4: the test signal is input into a trained network model after data preprocessing, and passes through the feature extractor G once through forward propagation f And tag predictor G l And obtaining an output equipment running state prediction result.
Fig. 5 shows the T-SNE distribution results of the original input data (fig. 5 (a)) and the output features of the training model (fig. 5 (b)), respectively. The results show that in the raw data, only one sample (pitting failure) can be isolated, while the other health samples are mixed together and are indistinguishable. The output data characteristics of the samples processed by the trained DDAENN model can be well in one-to-one correspondence and form four different clusters, and each cluster is composed of functions of labels with the same health state in a source domain and a target domain. The result shows that the proposed DDAENN-based fault diagnosis method can successfully extract common similar features from source domain and target domain samples, and map data features with the same type of health state to each other.
Fig. 6 shows a confusion matrix of predicted results, the left graph is a specific predicted case of 1200 test samples, and the right graph is a normalized result. The result shows that the method has the prediction accuracy of 99.75% for the monitoring data of the unlabeled gearbox, and only 3 abrasion fault samples in the total 1200 monitoring data samples are mispredicted, so that the method can successfully realize the fault diagnosis of the variable working condition load of the gearbox under the condition of lacking fault labels, and the model trained by the method has higher diagnosis accuracy and stronger generalization performance.
The invention realizes the integration of the self-adaptive methods in different fields on the algorithm level by constructing a self-adaptive integration framework.
Those skilled in the art will appreciate that the invention provides a system and its individual devices, modules, units, etc. that can be implemented entirely by logic programming of method steps, in addition to being implemented as pure computer readable program code, in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Therefore, the system and various devices, modules and units thereof provided by the invention can be regarded as a hardware component, and the devices, modules and units for realizing various functions included in the system can also be regarded as structures in the hardware component; means, modules, and units for implementing the various functions may also be considered as either software modules for implementing the methods or structures within hardware components.
The foregoing describes specific embodiments of the present invention. It is to be understood that the invention is not limited to the particular embodiments described above, and that various changes or modifications may be made by those skilled in the art within the scope of the appended claims without affecting the spirit of the invention. The embodiments of the present application and features in the embodiments may be combined with each other arbitrarily without conflict.
Claims (10)
1. A rotary machine fault migration diagnostic method, comprising the steps of:
step 1: respectively acquiring a marked monitoring data set and a non-marked monitoring data set of a rotating component under different running conditions to form a source domain data set and a target domain data set;
step 2: the same data preprocessing is carried out on the source domain data set and the target domain data set, so that two sample sets which can be identified by an integrated network algorithm are obtained;
step 3: selecting different deep migration learning algorithms, and constructing a deep neural network model according to an integrated migration framework;
step 4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function;
step 5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected deep migration learning algorithm;
step 6: calculating the deviation between the predicted result and the real result of different migration methods, and calculating the corresponding optimization self-adaptive factors of different deep migration learning methods in the integrated framework according to the deviation;
step 7: gradient derivation and optimization are carried out on the loss;
step 8: repeating the iteration steps 4-7 until reaching the convergence condition, stopping training to obtain a trained neural network model;
step 9: and (3) preprocessing the new monitoring data in the step (2), and inputting the data into a trained deep neural network model to obtain a predictive label of the data sample.
2. The rotary machine fault migration diagnostic method of claim 1, wherein in step 1, the source domain data set is represented as
The target domain dataset is denoted +.>
wherein ,
Monitoring data of the ith sample in the source domain dataset and corresponding health condition markers,/respectively>
For the ith sample in the target domain dataset, n is the minimum number of training samples for the batch.
3. The rotary machine fault migration diagnosis method according to claim 1, wherein in the step 6, the deviation between the predicted result and the true result of the different migration methods is calculated using the a-distance.
4. A method of diagnosing a fault migration of a rotary machine according to claim 3, wherein in step 7, the loss is gradient derived and optimized by a back propagation algorithm.
5. The rotary machine fault migration diagnosis method according to claim 1, wherein the step 2 specifically comprises the steps of:
step 2.1: abnormal values of the original vibration signal of the rotating part are removed, and the abnormal values are removed by using the Laida rule, and the formula is as follows:
wherein ,
mean value of signal segment, x i The value of the ith sample point in the signal sample;
is the standard deviation, n is the total number of sampling points in the signal segment;
step 2.2: slicing the time sequence vibration signals with the abnormal values eliminated, wherein each segment comprises 4096 vibration signal sampling points, and obtaining a wavelet time-frequency pattern book through continuous wavelet transformation as the input of a network model, wherein the continuous wavelet transformation formula is as follows:
where Φ (t) is the wavelet mother function, τ is the time shift coefficient, a is the scale coefficient, and a+.0.
6. The rotary machine fault migration diagnosis method according to claim 5, wherein in the step 3, the deep neural network model is DDAENN based on DDAE neural network;
the DDAE-based neural network DDAENN comprises a feature extractor G f Domain discriminator G d Tag predictor G l ；
The DDAE-based neural network DDAENN adopts a domain countermeasure technology and an MMD-based domain adaptation technology.
7. The method according to claim 6, wherein in the step 4, the classification accuracy of the source domain labeled data by the deep neural network is trained, and the cross entropy loss is used, and the loss function is as follows:
wherein ,Wy and by Is made by a tag predictor G l Matrix vector pair obtained after linear transformation, f represents the matrix vector pair obtained by the feature extractor G f The extracted features, F (·) represents the softmax function, I { · } represents the indicator function, k is the source domain ith sample
A corresponding real sample tag;
in the step 5, for domain countermeasure technology, the loss function is expressed as:
wherein φ represents a sigmod function, W d and bd Is a pair of linear transformation matrix vectors for a domain classifier, d i A domain label representing the ith training sample, f s and ft Respectively represented by the feature extractor G f The extracted source domain and target domain features, m is the total number of samples in one training batch;
for MMD-based domain adaptation techniques, its loss function is expressed as:
wherein ,ns and nt Representing the number of batch training samples from the source domain and the target domain respectively,
and
Domain invariant feature representing two domains of depth feature extractor output +_>
Is a regenerative core Hilbert space with k characteristic cores.
8. The fault migration diagnosis method of rotary machine according to claim 1, wherein the step 6 specifically comprises the steps of:
step 6.1: the A-distance after different migration technologies are calculated, and the calculation formula is as follows:
wherein ,xj Represents the j-th sample, G c (x j ) Is a classifier G c For sample x j Is provided with an output of (a),
representing an indication function, n' being two different domain data sets D s and Dt Is the sample size of err i (G c ) Data representing two different fields in the i-th field adaptive method are in classifier G c Errors in the above;
Two fields D after the use of the ith field adaptation method s and Dt A-distance between;
step 6.2: for the used i-th field self-adaptive method, respectively calculating the corresponding optimized self-adaptive factor alpha i The calculation formula is as follows:
wherein ,
a-distance parameter which is the i-th field adaptive method +.>
Sum of all A-distance parameters
Is used to represent the weight of the migration technique, +.>
The weight occupied by the j-th domain self-adaptive method;
step 6.3: the obtained optimized adaptive factor alpha i Loss function of field adaptive method selected in step 5
In combination, the total loss function of the migration method is obtained, expressed as:
9. the fault migration diagnosis method of rotary machine according to claim 1, wherein the step 9 specifically comprises the steps of:
step 9.1: acquiring on-line monitoring data of an actual rotary machine through a sensor, preprocessing the monitoring data in the step 2, and then regularizing the monitoring data into a proper wavelet time-frequency diagram sample;
step 9.2: and (3) sequentially inputting the obtained time-frequency pattern book into the trained neural network model obtained in the step (8) to obtain a prediction label of a corresponding sample, and evaluating the running state of the current equipment according to the prediction result of the label.
10. A rotary machine fault migration diagnostic system, comprising the steps of:
module M1: respectively acquiring a marked monitoring data set and a label-free monitoring data set of a rotating component under different running conditions to form a source domain data set and a target domain data set, wherein the monitoring data of an ith sample in the source domain data set and a corresponding health condition mark thereof are respectively the ith sample in the target domain data set, and n is the number of training samples in the minimum batch;
module M2: the same data preprocessing is carried out on the source domain data set and the target domain data set, so that two sample sets which can be identified by an integrated network algorithm are obtained;
module M3: selecting different deep migration learning algorithms, and constructing a deep neural network model according to an integrated migration framework;
module M4: inputting source domain data with labels, obtaining output prediction labels through forward propagation in a constructed network, and comparing the prediction labels with real labels to obtain a loss function;
module M5: inputting source domain data and target domain data, and calculating loss between features of the two domain data by using a selected deep migration learning algorithm;
module M6: calculating the deviation between the predicted result and the real result of different migration methods by using the A-distance, and calculating the corresponding optimization self-adaptive factors of different deep migration learning methods in the integrated framework according to the deviation;
module M7: carrying out gradient derivation and optimization on the loss by using a back propagation algorithm;
module M8: repeating the triggering modules M4-M7 to execute iteration until reaching convergence conditions, and stopping training to obtain a trained neural network model;
module M9: and (3) preprocessing the new monitoring data by the data of the module 2, and inputting the new monitoring data into a trained deep neural network model to obtain a prediction label of the data sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211397639.9A CN116204781A (en) | 2022-11-09 | 2022-11-09 | Rotary machine fault migration diagnosis method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211397639.9A CN116204781A (en) | 2022-11-09 | 2022-11-09 | Rotary machine fault migration diagnosis method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116204781A true CN116204781A (en) | 2023-06-02 |
Family
ID=86517990
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211397639.9A Pending CN116204781A (en) | 2022-11-09 | 2022-11-09 | Rotary machine fault migration diagnosis method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116204781A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407698A (en) * | 2023-12-14 | 2024-01-16 | 青岛明思为科技有限公司 | Hybrid distance guiding field self-adaptive fault diagnosis method |
-
2022
- 2022-11-09 CN CN202211397639.9A patent/CN116204781A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407698A (en) * | 2023-12-14 | 2024-01-16 | 青岛明思为科技有限公司 | Hybrid distance guiding field self-adaptive fault diagnosis method |
| CN117407698B (en) * | 2023-12-14 | 2024-03-08 | 青岛明思为科技有限公司 | Hybrid distance guiding field self-adaptive fault diagnosis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Intelligent fault diagnosis of rolling bearings under varying operating conditions based on domain-adversarial neural network and attention mechanism | |
| CN109146246B (en) | Fault detection method based on automatic encoder and Bayesian network | |
| CN112629863A (en) | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions | |
| CN113076834B (en) | Rotating machine fault information processing method, processing system, processing terminal, and medium | |
| Li et al. | Partial domain adaptation in remaining useful life prediction with incomplete target data | |
| CN115618733A (en) | Multi-scale hybrid attention mechanism modeling method for predicting remaining service life of aircraft engine | |
| CN112926642B (en) | Multi-source domain self-adaptive intelligent mechanical fault diagnosis method and system | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| Saravanakumar et al. | Hierarchical symbolic analysis and particle swarm optimization based fault diagnosis model for rotating machineries with deep neural networks | |
| CN116007937B (en) | Intelligent fault diagnosis method and device for mechanical equipment transmission part | |
| Chen et al. | Lightweight convolutional transformers enhanced meta-learning for compound fault diagnosis of industrial robot | |
| CN116204781A (en) | Rotary machine fault migration diagnosis method and system | |
| CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
| CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
| Guo et al. | Fault diagnosis of reciprocating compressor valve based on transfer learning convolutional neural network | |
| CN116070126A (en) | Aviation plunger pump oil distribution disc abrasion detection method and system based on countermeasure self-supervision | |
| CN115876467A (en) | Pseudo label transfer type two-stage field self-adaptive rolling bearing fault diagnosis method | |
| CN116911378A (en) | Hydropower unit fault diagnosis method and system based on transfer learning and PSO algorithm | |
| Zhao et al. | Hybrid semi-supervised learning for rotating machinery fault diagnosis based on grouped pseudo labeling and consistency regularization | |
| CN114548295A (en) | Bearing fault classification system and method based on multi-scale domain adaptive network | |
| Liu et al. | A novel health prognosis method for system based on improved degenerated Hidden Markov model | |
| CN117951646A (en) | Data fusion method and system based on edge cloud | |
| CN113778811A (en) | Fault monitoring method and system based on deep convolution migration learning software system | |
| CN117633690A (en) | Rotary machine health state monitoring method and equipment based on data driving | |
| CN117112992A (en) | Fault diagnosis method for polyester esterification stage |
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 |