CN115630299A - Rotary machine fault diagnosis method and system based on joint domain adaptive network - Google Patents
Rotary machine fault diagnosis method and system based on joint domain adaptive network Download PDFInfo
- Publication number
- CN115630299A CN115630299A CN202211238858.2A CN202211238858A CN115630299A CN 115630299 A CN115630299 A CN 115630299A CN 202211238858 A CN202211238858 A CN 202211238858A CN 115630299 A CN115630299 A CN 115630299A
- Authority
- CN
- China
- Prior art keywords
- domain
- data
- joint
- adaptation
- training
- 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
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention provides a rotary machine fault diagnosis method and a rotary machine fault diagnosis system based on a joint domain adaptive network, which comprise the following steps: an initial pre-training model is obtained through training by combining consistency regular loss and pseudo labels, so that the prediction accuracy of target domain data in an initial stage in a domain adaptation process is improved; and the utilization rate of fault information in the label-free data is improved by generating the pseudo label for the label-free data of the target domain.
Description
Technical Field
The invention relates to the technical field of rotary machine fault diagnosis, in particular to a rotary machine fault diagnosis method and system based on a joint domain adaptive network.
Background
The traditional intelligent fault diagnosis method needs to rely on enough mark data to train a diagnosis model, but in a real industrial scene, most of equipment normally operates, and fault data are difficult to obtain. The diagnostic knowledge can be reused among a plurality of related machines based on the idea that the diagnostic knowledge of a laboratory bearing can help to identify the health state of the bearing in an industrial scene, and the diagnostic knowledge learned in labeled source domain data can be migrated to unlabeled target domain data by adopting domain adaptation and semi-supervised learning, so that the problem that the labeled data are not available in some real industrial scenes is well solved. However, the current domain adaptation-based fault diagnosis mainly has the following problems:
domain-invariant features between the source domain and the target domain are learned in some distance metric and counter learning manners, and these schemes are only aligned from the perspective of the feature space, and the influence of label information on alignment is ignored. This would result in only globally aligning the two data distributions, ignoring the inter-class distribution alignment;
a single domain adaptive alignment mode is adopted, the difference in data distribution is large, the effect of the alignment mode is greatly influenced, and multi-mode information contained in data cannot be captured;
the method includes that a model trained by source domain data is directly adopted to predict target domain data, then domain adaptation alignment is carried out, in the early stage of training, due to the fact that the source domain data and the target domain data are large in distribution difference, the prediction effect of the target domain data is poor, domain adaptation depends on the prediction result of the target domain data, and if the prediction effect of the target domain data is poor, the model is difficult to train and cannot be converged.
Disclosure of Invention
The invention aims to provide a rotary machine fault diagnosis method and system based on a combined domain adaptation network, and aims to solve the problem that the existing fault diagnosis effect based on domain adaptation is poor.
In order to solve the above technical problem, the present invention provides a rotating machine fault diagnosis method based on a joint domain adaptation network, including:
an initial pre-training model is obtained through training by combining consistency regular loss and a pseudo label, so that the prediction accuracy of target domain data in an initial stage in a domain adaptation process is improved; and
the utilization rate of fault information in the label-free data is improved by generating the pseudo label for the label-free data of the target domain.
Optionally, in the method for diagnosing a fault of a rotating machine based on a joint domain adaptation network, the method further includes:
the method comprises the steps that the joint distribution of characteristics between source domain data and target domain data is aligned in a plurality of modes and the joint distribution of labels between the source domain data and the target domain data is aligned in a plurality of modes by combining the alignment modes of joint maximum mean difference domain adaptation and condition countermeasure domain adaptation so as to capture multi-modal information contained in the data; and
the two joint distributions are aligned from the whole situation, and the alignment among different categories is carried out, so that the alignment accuracy under different data distribution differences and the source domain target domain similar feature alignment effect are improved.
Optionally, in the method for diagnosing a fault of a rotating machine based on a joint domain adaptation network, the method further includes:
and introducing a semi-supervised pre-training model, and introducing target domain data with higher reliability through threshold screening in the semi-supervised training process to replace the model trained by the source domain data directly to predict the target domain data in a pre-adaptation stage so as to improve the accuracy of prediction.
Optionally, in the method for diagnosing a fault of a rotating machine based on a joint domain adaptation network, the method further includes:
constructing two modules for carrying out domain adaptation of a source domain and a target domain;
performing domain adaptation on the feature extraction and classification layers by combining maximum mean difference domain adaptation by using the combined distribution difference of the input features and the output labels;
performing domain confrontation training between the features and the predictive labels to reduce domain drift;
based on the uncertainty of the domain classifier prediction, reweighing the samples in the domain classifier through the entropy calculation weight omega to reduce the influence of the target example with uncertain prediction; and
the maximum distinction between the categories and the domain adaptation in the multi-mode are performed by two modules.
Optionally, in the method for diagnosing a fault of a rotating machine based on a joint domain adaptive network, the training to obtain an initial pre-training model by combining the consistency canonical loss and the pseudo label includes:
aiming at the labeled data, weak enhancement is carried out;
aiming at the non-tag data, respectively carrying out weak enhancement and strong enhancement, wherein the proportion of noise added by the weak enhancement is different from that added by the strong enhancement;
predicting a pseudo label for weakly enhanced unlabeled data, and reserving the pseudo label when the model generates a prediction above a threshold value;
predicting classification probability of the strongly enhanced label-free data, and measuring the consistency of prediction of the strong data and the weak data through cross entropy loss; and
and combining the supervision loss of the labeled data and the consistency regular loss of the unlabeled data to obtain a pre-training model, so that the error of target domain data prediction in the domain adaptation stage is reduced.
Optionally, in the rotating machine fault diagnosis method based on the joint domain adaptive network, improving the alignment effect of the similar features of the source domain and the target domain includes:
respectively carrying out weak enhancement on the source domain data with the mark and the target domain data without the mark, and sending the data into a pre-training model obtained by the weak enhancement in the pre-training stage;
performing multi-layer linear transformation on the final characteristics and the labels to express the joint distribution of the characteristics and the labels;
adopting a joint maximum mean difference domain adaptation mode and a conditional domain versus reactance domain adaptation mode to align the joint distribution of the features and the labels; and
and training to obtain the domain invariant features of the source domain and the target domain by combining the loss of the label classification combined maximum mean difference domain adaptation mode and the loss of the conditional domain versus resistance domain adaptation mode so as to improve the accuracy of the migration learning fault diagnosis.
Optionally, in the method for diagnosing a fault of a rotating machine based on a joint domain adaptive network, the pre-training further includes:
pre-training introduces consistency regularization loss;
loss of consistency regularity includes: after the noise is injected into the unlabeled data, the classifier outputs the same probability distribution for the unlabeled data, and forces a unlabeled sample to be classified into the same classification as the enhanced sample of the classifier; and
by forcing a label-free sample to be classified into the same classification as the self enhanced sample, and introducing information without labeled data into the pre-training model, the accuracy of the pre-training model is higher than that of a model which is directly trained by adopting source domain data to predict target domain data, so that the problem that the model cannot be converged at the early stage is avoided.
Optionally, in the method for diagnosing a fault of a rotating machine based on a joint domain adaptation network, the joint distribution of the two domain adaptation modules in combination with the alignment feature and the tag further includes:
combining two different domain adaptation modes to complementarily align the joint distribution of the data of the two domains so as to avoid the situation that the difference of the data distribution of the source domain and the target domain is greater than a data distribution threshold value: aligning the distribution among the features, neglecting category information, aligning the distribution of the two features on the whole, and neglecting the relationship among the categories; and
combining two different domain adaptation modes to complementarily align the joint distribution of the two domain data to avoid the situation of aligning the joint distribution by adopting a single domain adaptation module: based on the difference between the distributions that the information complexity is greater than the complexity threshold, a single domain adaptation module cannot capture multi-modal information therein.
Optionally, in the rotating machine fault diagnosis method based on the joint domain adaptive network, the method further includes:
migrating under different working conditions of the same kind of data set; based on the fact that vibration signals of bearings or gears collected under different working conditions have differences, the rotating machinery runs under variable working conditions, the data volume collected under at least one working condition is smaller than a data volume threshold value, the working condition that the data volume is larger than the data volume threshold value is used as source domain data, and diagnosis knowledge trained from the source domain data is migrated to other working conditions; the difference comprises a load parameter and a rotating speed parameter;
migrating between different data sets; simulating the collected bearing data by using a laboratory; identifying the health state of the bearing in an engineering scene based on the diagnosis knowledge of the laboratory bearing so as to be used for bearing fault diagnosis in the engineering scene; and
migrating under different working conditions of mixed fault types; based on the condition that the bearing and the gear simultaneously have faults and the condition that the data distribution difference collected under different working conditions is larger than the data distribution threshold, a plurality of domain adaptation modes and/or information without labeled data are/is introduced, and the influence of the data distribution difference is reduced by introducing a pre-training model and combining a plurality of domain adaptation modules.
The invention also provides a rotary machine fault diagnosis system based on the joint domain adaptive network, which comprises the following components:
the pre-training model module is configured to obtain an initial pre-training model through training by combining consistency regular loss and pseudo labels so as to improve the prediction accuracy of target domain data in an initial stage in a domain adaptation process; and
and the semi-supervised training module is configured to improve the utilization rate of fault information in the label-free data in a mode of generating a pseudo label for the label-free data of the target domain.
In the rotary machine fault diagnosis method and system based on the joint domain adaptation network, an initial pre-training model is obtained through training by combining the consistency regular loss and the pseudo label, so that the prediction effect of target domain data in the initial stage in the domain adaptation process is not poor, the model is prevented from being difficult to train and incapable of converging, and the problem that the target domain data in the existing domain adaptation fault diagnosis algorithm is not fully utilized is solved.
Furthermore, the invention combines the two domain adaptive alignment modes of the combined maximum mean difference and the condition anti-domain, aligns the combined distribution of the source domain data and the target domain data characteristics and the labels in multiple modes, can capture the multi-mode information contained in the data, aligns the two distributions globally, and considers the alignment among different categories, thereby achieving good alignment effect when the data distribution difference is large, and solving the problem of poor alignment effect of similar characteristics of the source domain and the target domain.
The invention provides a semi-supervised rotating machine fault diagnosis method based on a joint domain adaptive network, and aims to solve the problem that a good diagnosis effect can still be obtained through the method under the condition that no labeled data exists in a real industrial scene. The scheme mainly combines the ideas of semi-supervised learning and domain adaptation, designs a semi-supervised rotating machine fault diagnosis method based on a joint domain adaptation network, does not need prior knowledge, completely learns the domain invariant characteristics through a neural network automatically, does not need a label in a target domain, is a semi-supervised algorithm independent of the label in the target domain, and can solve the problem of insufficient data marking.
The invention provides a fault diagnosis algorithm with good diagnosis effect in the scene of fault marking data insufficiency or completely unmarked fault data. Compared with the traditional fault diagnosis algorithm of single domain adaptive alignment, the invention combines two domain adaptive alignment modes, simultaneously considers the class information of the data, aligns the joint distribution of the two data instead of the edge distribution, greatly reduces the probability of negative migration, and improves the migration effect. Compared with some migration fault diagnosis algorithms which directly adopt a model with well-trained source domain data to predict target domain data and then carry out domain adaptive alignment, the method firstly adopts a pre-training method to introduce information without labeled data into a pre-training model, improves the prediction effect of the target domain data to a certain extent, and avoids the problems that the model is difficult to train and cannot be converged.
Drawings
FIG. 1 is a schematic diagram of a pre-training topology of a fault diagnosis method for a rotating machine based on a joint domain adaptive network according to an embodiment of the present invention;
fig. 2 is a domain adaptation topology diagram of a rotating machine fault diagnosis method based on a joint domain adaptation network according to an embodiment of the present invention.
Detailed Description
The invention is further elucidated with reference to the drawings in conjunction with the detailed description.
It should be noted that the components in the figures may be shown exaggerated for illustrative purposes and are not necessarily to scale. In the figures, identical or functionally identical components are provided with the same reference symbols.
In the present invention, "disposed on" \\8230 "", "disposed over" \8230 "", and "disposed over" \8230 "", do not exclude the presence of an intermediate therebetween, unless otherwise specified. Furthermore, "arranged on or above" \\8230 ", merely indicates a relative positional relationship between two components, and in certain cases, such as after reversing the product direction, may also be converted to" arranged under or below \8230 ", and vice versa.
In the present invention, the embodiments are only intended to illustrate the aspects of the present invention, and should not be construed as limiting.
In the present invention, the terms "a" and "an" do not exclude the presence of a plurality of elements, unless otherwise specified.
It is further noted herein that in embodiments of the present invention, only a portion of the components or assemblies may be shown for clarity and simplicity, but those of ordinary skill in the art will appreciate that, given the teachings of the present invention, required components or assemblies may be added as needed in a particular scenario. Furthermore, features from different embodiments of the invention may be combined with each other, unless otherwise indicated. For example, a feature of the second embodiment may be substituted for a corresponding or functionally equivalent or similar feature of the first embodiment, and the resulting embodiments are likewise within the scope of the disclosure or recitation of the present application.
It is also noted herein that, within the scope of the present invention, the terms "same", "equal", and the like do not mean that the two values are absolutely equal, but allow some reasonable error, that is, the terms also encompass "substantially the same", "substantially equal". By analogy, in the present disclosure, the terms "perpendicular," parallel, "and the like in the directions of the tables also encompass the meanings of" substantially perpendicular, "" substantially parallel.
The numbering of the steps of the methods of the present invention does not limit the order of execution of the steps of the methods. Unless specifically stated, the method steps may be performed in a different order.
The present invention provides a method and a system for diagnosing a fault of a rotating machine based on a joint domain adaptive network, which are described in further detail below with reference to the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description. It is to be noted that the drawings are in a very simplified form and are not to precise scale, which is provided for the purpose of facilitating and clearly illustrating embodiments of the present invention.
The invention aims to provide a rotary machine fault diagnosis method and system based on a combined domain adaptation network, and aims to solve the problem that the existing fault diagnosis effect based on domain adaptation is poor.
In order to achieve the above object, the present invention provides a rotating machine fault diagnosis method and system based on a joint domain adaptive network, including: an initial pre-training model is obtained through training by combining consistency regular loss and pseudo labels, so that the prediction accuracy of target domain data in an initial stage in a domain adaptation process is improved; and generating a pseudo label by giving label-free data to the target domain, and fully utilizing fault information in the label-free data.
Fig. 1-2 provide a first embodiment of the present invention, which respectively show a domain adaptive network-based rotating machine fault diagnosis method pre-training and a domain adaptive topology diagram.
The invention provides a semi-supervised rotating machine fault diagnosis method based on a joint domain adaptive network, and aims to solve the problem that a good diagnosis effect can still be obtained by the method under the condition that no labeled data exists in a real industrial scene. The scheme mainly combines the ideas of semi-supervised learning and domain adaptation, and designs a semi-supervised rotating machine fault diagnosis method based on a joint domain adaptation network; has the following beneficial effects:
the domain invariant features are automatically learned completely through a neural network without priori knowledge;
the target domain does not need a label, and the method is a semi-supervised algorithm independent of the label of the target domain and can solve the problem of insufficient data marking;
by combining consistency regular loss and pseudo labels, an initial pre-training model is obtained through training, so that the prediction effect of target domain data in the initial stage in the domain adaptation process is not poor, the model is prevented from being difficult to train and incapable of converging, and the problem that the target domain data in the existing domain adaptation fault diagnosis algorithm is not fully utilized is solved; and
the two domain adaptive alignment modes of the source domain data and the target domain data are aligned in a plurality of modes by combining the two domain adaptive alignment modes of the combined maximum mean difference and the condition countermeasure domain, and the multi-mode information contained in the data can be captured, so that the two distributions are aligned globally, and the alignment among different categories is considered, so that a good alignment effect can be achieved when the data distribution difference is large, and the problem of poor alignment effect of similar features of the source domain and the target domain is solved.
As shown in fig. 1, the pre-training includes: aiming at the labeled data, only weak enhancement is carried out; aiming at the non-label data, weak enhancement and strong enhancement are respectively carried out, and the difference is that the proportion of the added noise is different; predicting a pseudo label for weakly enhanced unlabeled data, and only when the model generates a prediction above a threshold value, reserving the pseudo label; predicting classification probability of the strongly enhanced label-free data, and measuring the consistency of prediction of the strong data and the weak data through cross entropy loss; and combining the supervision loss of the labeled data and the consistency regular loss of the unlabeled data to obtain a pre-training model, so that the error of target domain data prediction in the domain adaptation stage is reduced.
As shown in fig. 2, the domain adaptation includes: respectively carrying out weak enhancement on the source domain data with the mark and the target domain data without the mark, and sending the data into a pre-training model obtained by the weak enhancement obtained in the pre-training stage; performing multi-layer linear transformation on the final feature f and the label l to express the joint distribution of the feature and the label; aligning the joint distribution of the features and the labels by adopting two field adaptation modes of joint maximum mean difference JMMD and conditional field confrontation CDA; by combining the label classification seven loss, the JMMD loss and the CDA loss training, the domain invariant characteristics of the source domain and the target domain can be obtained, and a good migration learning fault diagnosis effect is achieved.
Compared with the method for predicting the target domain data by directly using the model trained by the source domain data in the pre-adaptation stage, the method introduces the semi-supervised pre-training model, introduces the relatively credible target domain data through threshold screening in the semi-supervised training process, and can improve the accuracy of prediction.
The present invention constructs two modules to achieve domain adaptation of the source domain and the target domain. On one hand, domain adaptation is performed at a feature extraction and classification layer through JMMD by using the joint distribution difference of the input features and the output labels. On the other hand, domain confrontation training is performed between the features and the predictive labels to reduce domain drift. Meanwhile, considering the prediction uncertainty of the domain classifier, the samples in the domain classifier are reweighed through the entropy calculation weight omega, and the influence of the target example with uncertain prediction is reduced. The two modules not only realize the maximum distinction between the categories, but also realize the field adaptation under the multi-mode.
Specifically, pre-training is necessary, which introduces a loss of consistency regularity. The idea of consistency regularization loss is that even after unlabeled data is injected with noise, the classifier should output the same probability distribution for it, i.e. it is forced that an unlabeled sample should be classified as the same class as the enhancement sample itself. By the method, information without labeled data is introduced into the pre-training model, so that the accuracy of the pre-training model is higher than that of a model which is trained by directly adopting source domain data to predict target domain data, and the problems that the model is difficult to train in the early stage and cannot be converged are solved.
Further, two domain adaptation modules are necessary in combination with the joint distribution of alignment features and labels: when the data distribution difference between the source domain and the target domain is large, only the distribution between the features is aligned, the category information is ignored, only the distribution of the two features is aligned on the whole, and the relationship between the categories is ignored, so that the alignment effect is greatly influenced. Similarly, only a single domain adaptation module is adopted to align and jointly distribute, and because the difference information among the distributions is complex, the single domain adaptation module is difficult to capture the multi-mode information contained in the distribution, and a good alignment effect is difficult to obtain. Thus, two different domain adaptation modes are combined to complementarily align the joint distribution of the two domain data.
The invention provides a fault diagnosis algorithm with good diagnosis effect in the scene of fault marking data shortage or no marking fault data. Compared with the traditional fault diagnosis algorithm of single domain adaptive alignment, the scheme of the invention combines two domain adaptive alignment modes, simultaneously considers the class information of the data, aligns the joint distribution of the two data instead of the edge distribution, greatly reduces the probability of negative migration, and improves the migration effect. Compared with some migration fault diagnosis algorithms which directly adopt a model with trained source domain data to predict target domain data and then carry out domain adaptive alignment, the method firstly adopts a pre-training method to introduce information without labeled data into a pre-training model to improve the prediction effect of the target domain data to a certain extent, and avoids the problems that the model is difficult to train and cannot be converged.
In addition, the invention realizes the migration under different working conditions of the same data set, bearing or gear vibration signals collected under different working conditions are different, such as different loads and different rotating speeds, and considering that the rotating machinery is mostly in a variable working condition when running and the data quantity collected under some working conditions is possibly insufficient, if the invention has enough marked data under one working condition, the invention can use the marked data as source domain data, and the method can migrate the diagnosis knowledge from the source domain data science to the working condition with insufficient data quantity, thereby obtaining a good diagnosis effect.
Furthermore, the invention realizes the migration among different data sets, the faults are seldom generated in a real industrial scene, and the collected data is seldom used for training a model with good effect. The invention can adopt the bearing data collected by laboratory simulation, the diagnosis knowledge from the laboratory bearing can help to identify the health state of the bearing in the engineering scene, and the scheme can be used for bearing fault diagnosis in the engineering scene.
Furthermore, the invention realizes the migration of mixed fault types under different working conditions, the condition that the bearing and the gear simultaneously have faults exists in a real industrial scene, the data distribution difference collected under different working conditions is larger, and the difference of the data distribution can be further increased by the mutual influence of the different fault types, and in this case, a single domain adaptation mode is adopted, or the information without marked data is not introduced, so that an effective alignment effect is difficult to obtain. The influence caused by the difference can be greatly reduced by introducing a pre-training model and combining various domain adaptation modules.
In summary, the above embodiments have described in detail different configurations of the method and system for diagnosing a fault of a rotating machine based on a joint domain adaptive network, but it is understood that the present invention includes, but is not limited to, the configurations listed in the above embodiments, and any modifications based on the configurations provided by the above embodiments are within the scope of protection of the present invention. One skilled in the art can take the contents of the above embodiments to take a counter-measure.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The above description is only for the purpose of describing the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention, and any variations and modifications made by those skilled in the art based on the above disclosure are within the scope of the appended claims.
Claims (10)
1. A rotary machine fault diagnosis method based on a joint domain adaptive network is characterized by comprising the following steps:
training to obtain an initial pre-training model based on the consistency regular loss and the pseudo label so as to improve the prediction accuracy of target domain data in the initial stage in the domain adaptation process; and
the utilization rate of fault information in the non-label data is improved by generating the pseudo label for the non-label data of the target domain.
2. The joint domain adaptive network-based rotating machine fault diagnosis method according to claim 1, further comprising:
the method comprises the steps that the joint distribution of characteristics between source domain data and target domain data is aligned in a plurality of modes and the joint distribution of labels between the source domain data and the target domain data is aligned in a plurality of modes by combining the alignment modes of joint maximum mean difference domain adaptation and condition countermeasure domain adaptation so as to capture multi-modal information contained in the data; and
and aligning the two joint distributions from the whole situation and aligning different categories so as to improve the alignment accuracy under different data distribution differences and the alignment effect of similar features of the target domain of the source domain.
3. The joint domain adaptive network-based rotating machine fault diagnosis method according to claim 2, further comprising:
and introducing a semi-supervised pre-training model, and introducing target domain data with higher reliability through threshold screening in the semi-supervised training process to replace the model trained by the source domain data directly to predict the target domain data in a pre-adaptation stage so as to improve the accuracy of prediction.
4. The joint domain adaptive network-based rotating machine fault diagnosis method according to claim 3, further comprising:
constructing two modules for carrying out domain adaptation of a source domain and a target domain;
performing domain adaptation on the feature extraction and classification layers by combining maximum mean difference domain adaptation by using the combined distribution difference of the input features and the output labels;
performing domain confrontation training between the features and the prediction labels to reduce domain drift;
based on the uncertainty of the domain classifier prediction, reweighing the samples in the domain classifier through the entropy calculation weight omega to reduce the influence of the target example with uncertain prediction; and
the maximum distinction between the categories and the domain adaptation in the multi-mode are carried out by two modules.
5. The method of claim 4, wherein the training of the initial pre-training model by combining the consistency regularization loss and the pseudo label comprises:
aiming at the labeled data, weak enhancement is carried out;
aiming at the non-tag data, respectively carrying out weak enhancement and strong enhancement, wherein the proportion of noise added by the weak enhancement is different from that added by the strong enhancement;
predicting a pseudo label for weakly enhanced unlabeled data, and reserving the pseudo label when the model generates a prediction above a threshold value;
predicting classification probability of strongly enhanced non-label data, and measuring the predicted consistency of strong data and weak data through cross entropy loss; and
and combining the supervision loss of the labeled data and the consistency regular loss of the unlabeled data to obtain a pre-training model, so that the error of target domain data prediction in the domain adaptation stage is reduced.
6. The rotating machinery fault diagnosis method based on the joint domain adaptive network as claimed in claim 5, wherein improving the alignment effect of the similar features of the target domain of the source domain comprises:
respectively carrying out weak enhancement on the source domain data with the mark and the target domain data without the mark, and sending the data into a pre-training model obtained by the weak enhancement in the pre-training stage;
performing multi-layer linear transformation on the final features and the labels to represent the joint distribution of the features and the labels;
adopting a joint maximum mean difference domain adaptation mode and a conditional domain versus reactance domain adaptation mode to align the joint distribution of the features and the labels; and
and training to obtain the domain invariant features of the source domain and the target domain by combining the loss of the label classification and the loss of the adaptive mode of the maximum mean difference domain and the loss of the adaptive mode of the conditional domain versus the robust domain so as to improve the accuracy of fault diagnosis of transfer learning.
7. The joint domain adaptive network-based rotating machine fault diagnosis method according to claim 6, wherein the pre-training further comprises:
pre-training introduces consistency regularization loss;
loss of consistency regularization includes: after the noise is injected into the unlabeled data, the classifier outputs the same probability distribution for the unlabeled data, and forces a unlabeled sample to be classified into the same classification as the enhanced sample of the classifier; and
by forcing a label-free sample to be classified into the same classification as the self enhanced sample, and introducing information without labeled data into the pre-training model, the accuracy of the pre-training model is higher than that of a model which is directly trained by adopting source domain data to predict target domain data, so that the problem that the model cannot be converged at the early stage is avoided.
8. The joint domain adaptation network-based rotating machine fault diagnosis method of claim 7, wherein the two domain adaptation modules combining the joint distribution of alignment features and labels further comprises:
combining two different domain adaptation modes to complementarily align the joint distribution of the data of the two domains to avoid the situation that the difference of the data distribution of the source domain and the target domain is greater than the data distribution threshold value: aligning the distribution among the features, neglecting category information, aligning the distribution of the two features on the whole, and neglecting the relationship among the categories; and
combining two different domain adaptation modes to complementarily align the joint distribution of the two domain data to avoid the situation of aligning the joint distribution by adopting a single domain adaptation module: based on the difference between the distributions that the information complexity is greater than the complexity threshold, a single domain adaptation module cannot capture multi-modal information therein.
9. The joint domain adaptive network-based rotating machine fault diagnosis method according to claim 8, further comprising:
migrating under different working conditions of the same kind of data set; based on the fact that vibration signals of bearings or gears collected under different working conditions have differences, the rotating machinery runs under variable working conditions, the data volume collected under at least one working condition is smaller than a data volume threshold value, the working condition that the data volume is larger than the data volume threshold value is used as source domain data, and diagnosis knowledge trained from the source domain data is migrated to other working conditions; the differences include load parameters and speed parameters;
migrating between different data sets; simulating the collected bearing data by using a laboratory; identifying the health state of the bearing in an engineering scene based on the diagnosis knowledge of the laboratory bearing for bearing fault diagnosis in the engineering scene; and
migrating under different working conditions of mixed fault types; based on the condition that the bearing and the gear simultaneously have faults and the condition that the data distribution difference collected under different working conditions is larger than the data distribution threshold, a plurality of domain adaptation modes and/or information without labeled data are/is introduced, and the influence of the data distribution difference is reduced by introducing a pre-training model and combining a plurality of domain adaptation modules.
10. A rotary machine fault diagnosis system based on a joint domain adaptive network, characterized by comprising:
the pre-training model module is configured to train to obtain an initial pre-training model by combining consistency regular loss and pseudo labels so as to improve the prediction accuracy of target domain data in an initial stage in a domain adaptation process; and
and the semi-supervised training module is configured to improve the utilization rate of fault information in the label-free data in a mode of generating a pseudo label for the label-free data of the target domain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211238858.2A CN115630299A (en) | 2022-10-11 | 2022-10-11 | Rotary machine fault diagnosis method and system based on joint domain adaptive network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211238858.2A CN115630299A (en) | 2022-10-11 | 2022-10-11 | Rotary machine fault diagnosis method and system based on joint domain adaptive network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115630299A true CN115630299A (en) | 2023-01-20 |
Family
ID=84905411
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211238858.2A Pending CN115630299A (en) | 2022-10-11 | 2022-10-11 | Rotary machine fault diagnosis method and system based on joint domain adaptive network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115630299A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116229080A (en) * | 2023-05-08 | 2023-06-06 | 中国科学技术大学 | Semi-supervised domain adaptive image semantic segmentation method, system, equipment and storage medium |
-
2022
- 2022-10-11 CN CN202211238858.2A patent/CN115630299A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116229080A (en) * | 2023-05-08 | 2023-06-06 | 中国科学技术大学 | Semi-supervised domain adaptive image semantic segmentation method, system, equipment and storage medium |
| CN116229080B (en) * | 2023-05-08 | 2023-08-29 | 中国科学技术大学 | Semi-supervised domain adaptive image semantic segmentation method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cheng et al. | Wasserstein distance based deep adversarial transfer learning for intelligent fault diagnosis with unlabeled or insufficient labeled data | |
| Jenssen et al. | Intelligent monitoring and inspection of power line components powered by UAVs and deep learning | |
| CN111581116B (en) | Cross-project software defect prediction method based on hierarchical data screening | |
| CN115630299A (en) | Rotary machine fault diagnosis method and system based on joint domain adaptive network | |
| Azari et al. | A systematic literature review on transfer learning for predictive maintenance in industry 4.0 | |
| Tan et al. | MiDAN: A framework for cross-domain intelligent fault diagnosis with imbalanced datasets | |
| CN112001939A (en) | Image foreground segmentation algorithm based on edge knowledge conversion | |
| CN114021285B (en) | Rotary machine fault diagnosis method based on mutual local countermeasure migration learning | |
| Zhou et al. | SWDet: Anchor-based object detector for solid waste detection in aerial images | |
| Wang et al. | A graph neural network-based data cleaning method to prevent intelligent fault diagnosis from data contamination | |
| Yang et al. | Data augmentation considering distribution discrepancy for fault diagnosis of drilling process with limited samples | |
| CN114444374A (en) | Multi-source to multi-target domain self-adaption method based on similarity measurement | |
| Zheng et al. | Fault detection in complex mechatronic systems by a hierarchical graph convolution attention network based on causal paths | |
| CN113884300A (en) | Rolling bearing fault diagnosis method for deep anti-migration learning | |
| Xu et al. | A progressive fault diagnosis method for rolling bearings based on VMD energy entropy and a deep adversarial transfer network | |
| Malanchini et al. | Leveraging machine learning for industrial wireless communications | |
| Yang et al. | Defect prediction by using cluster ensembles | |
| CN116026590A (en) | Self-supervision bearing fault diagnosis method | |
| CN113724233B (en) | Transformer equipment appearance image defect detection method based on fusion data generation and transfer learning technology | |
| Wang et al. | A cross-domain intelligent fault diagnosis method based on deep subdomain adaptation for few-shot fault diagnosis | |
| CN116089812A (en) | Fault diagnosis method based on semi-supervised contrast domain generalization intelligent model | |
| CN115496233A (en) | Markov model-based data middlebox operation fault prediction method | |
| Zheng et al. | Few-shot intelligent fault diagnosis based on an improved meta-relation network | |
| CN114528679A (en) | Multi-mode fault early warning method and system for numerical control system | |
| Bai et al. | Studying the coupling and coordination of regional economic and university development levels based on a deep learning model |
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 |