CN114021285A - Rotary machine fault diagnosis method based on mutual local countermeasure transfer learning - Google Patents
Rotary machine fault diagnosis method based on mutual local countermeasure transfer learning Download PDFInfo
- Publication number
- CN114021285A CN114021285A CN202111362333.5A CN202111362333A CN114021285A CN 114021285 A CN114021285 A CN 114021285A CN 202111362333 A CN202111362333 A CN 202111362333A CN 114021285 A CN114021285 A CN 114021285A
- Authority
- CN
- China
- Prior art keywords
- domain
- data
- mutual
- classifier
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 91
- 238000003745 diagnosis Methods 0.000 title claims abstract description 71
- 238000013526 transfer learning Methods 0.000 title claims abstract description 42
- 230000000694 effects Effects 0.000 claims description 35
- 238000009826 distribution Methods 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 20
- 238000013508 migration Methods 0.000 claims description 19
- 230000005012 migration Effects 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 230000006978 adaptation Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000013461 design Methods 0.000 claims description 5
- 230000036541 health Effects 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 2
- 230000008901 benefit Effects 0.000 description 6
- 239000003795 chemical substances by application Substances 0.000 description 6
- 238000005259 measurement Methods 0.000 description 4
- 230000003042 antagnostic effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000000712 assembly Effects 0.000 description 2
- 238000000429 assembly Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 206010043417 Therapeutic response unexpected Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/04—Constraint-based CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
Abstract
The invention provides a rotary machine fault diagnosis method based on mutual local countermeasure transfer learning, which comprises the following steps: considering that the label information hidden in the target domain data is not fully utilized, a mutual learning network structure is designed to learn reliable pseudo labels, and further the target domain data without labels is fully utilized, so that the model performance is improved. Is more robust than the model than a single network structure. Considering the problem that alignment deviation may occur when the data of the source domain and the target domain are aligned directly from the global perspective by ignoring the category information of the fault, a plurality of local anti-domain classifiers are provided to replace the global anti-domain classifier, and each local anti-domain classifier aligns the source domain and the target domain feature of each category respectively to obtain better similar features.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a rotary machine fault diagnosis method based on mutual local countermeasure transfer learning.
Background
The traditional intelligent fault diagnosis method needs to rely on enough marking 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 idea that diagnostic knowledge can be reused across multiple related machines, such as laboratory bearing diagnostic knowledge may be helpful in identifying the health of a bearing in an industrial setting. Therefore, the diagnosis knowledge learned from the source domain data with sufficient data volume can be migrated to the target domain data with insufficient data volume by adopting the migration learning mode, and the problem of insufficient marker data in a real industrial scene is well solved. Currently, migration learning fault diagnosis is mainly classified into 3 types:
example-based methods: the data similar to the target domain is found in the source domain, the weight of the data is adjusted, so that the new data is matched with the data of the target domain, and then training and learning are carried out to obtain the model suitable for the target domain. The method has the advantages of simple method and easy realization. The disadvantage is that the choice of weights and the measure of similarity are empirically dependent and the data distribution of the source domain and the target domain tend to be different. The disadvantage of this method is that a priori knowledge is required.
The method based on parameter fine tuning comprises the following steps: the first stage constructs a pre-trained diagnostic model using labeled source data. In the second stage, the parameters of the pre-trained diagnostic model are adjusted using a small number of labeled data in the target domain. Due to the pre-trained diagnostic model, the parameter-based approach may save significant computational resources in processing large amounts of data. However, this fine-tuning-based approach can only be applied to supervised transfer learning tasks, since labeled target samples are necessary for fine-tuning the pre-trained model. The drawback of this approach is that a target domain label is required.
Disclosure of Invention
The invention aims to provide a rotary machine fault diagnosis method based on mutual local antagonistic transfer learning, and aims to solve the problems that the existing transfer learning fault diagnosis method is poor in transfer effect and weak in model robustness.
In order to solve the above technical problem, the present invention provides a rotating machine fault diagnosis method based on mutual local countermeasure transfer learning, including:
the mutual learning network structure is utilized to carry out fault diagnosis, and the two networks mutually guide to obtain a more reliable pseudo label, so that target domain data can be fully utilized, and the model effect and the robustness thereof are improved;
and providing a plurality of local contrast domain classifiers in place of the global contrast domain classifier, each local contrast domain classifier aligning the source domain and target domain features of each class separately for improved similarity features.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the diagnosing the fault by using a mutual learning network structure includes:
providing a mutual learning network structure, wherein the mutual learning network structure comprises at least two networks, and the at least two networks are mutual partner networks;
the generalization capability of each network is improved by referring to the learning experience of the partner network while marked source domain fault data is adopted in the training process;
learning experience is shared between the two networks to carry out mutual learning, and fault diagnosis is carried out according to mutual learning results.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes:
the class probability estimation contains the essential rule of the data extracted by the network, and can reflect the intrinsic essential characteristics of the data;
information provided by a single data label is only whether a sample belongs to a certain class and can not represent the relation among different classes, and the class probability estimation output by the network can recover the information to a certain extent;
and improving generalization performance based on the network, interchanging the data distribution characteristics of each network transmission study by adopting a mutual study structure, and mutually guiding and studying until convergence.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes:
the distribution difference of class probability estimation is constrained to be minimum by adopting Wasserstein distance, so that the regularization effect is achieved;
when the class probability estimation of one network is transmitted to the other network, additional class prior constraint is provided, the label classifier is prevented from being over-fitted to the 0-1 distribution of the truth label, and the over-fitting occurrence probability is effectively reduced.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes:
the agent classifier serves as a bridge between the label classifier and the feature extractor, and the agent classifier, the label classifier and the feature extractor are ensured to use the perception information in the target domain;
the agent classifier enables the two label classifiers not to be aligned with each other, so that two independent networks are prevented from convincing the prediction result of the other side, the agent classifier connects the two networks in the training process, and the effect of exchanging the information learned by the two networks is achieved; and
the structure of the proxy classifier enables the following actions to be performed between the label classifier and the feature extractor:
and instructing the other party to obtain the pseudo label, wherein the distribution difference between the predicted value of each step and the predicted value of the proxy classifier of the other party is minimum, so as to avoid the situation that the traditional mutual learning structure instructs each group by using the predicted label of another group, but does not instruct how to obtain the pseudo label, so that the two networks are not forced to obtain very similar feature representations, namely the two networks may obtain dissimilar feature representations and may have a poor mutual influence effect.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes:
based on the fact that one target domain sample is only aligned with the most relevant class data of the target domain sample, the local domain discriminators conduct fine-grained alignment of different classes of fault data distribution, the probability of negative migration is reduced, and therefore the wrong alignment of discrimination structures in different distributions is avoided.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes:
the data preprocessing module is used for preprocessing the vibration signals collected on the rotating mechanical equipment;
the feature extraction module constructs a multilayer CNN convolution network extraction feature, and extracts features of different scales by setting different initialization parameters;
the domain adaptation module designs 3 Loss functions to form a final Loss function, wherein the 3 Loss functions comprise source domain label data classification Softmax multi-classification LossclsSoftmax two-classification domain discriminator LossdomAnd Loss of mutual learning structureml；
Training to obtain the domain invariant characteristics of the source domain and the target domain by minimizing the final Loss function so as to improve the fault diagnosis effect of the transfer learning;
wherein LossclsAnd a loss function with label data in the source domain has supervision characteristics based on the loss function, so that the effect of the whole model is ensured.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes:
Lossdomthe system comprises k Softmax two-classification domain discriminators, wherein k is the number of fault classes, and a plurality of local domain classifiers are adopted to align the source domain and the target domain characteristics of each class respectively based on different fault types so as to avoid the domain countermeasure classifier from aligning the source domain and the target domain characteristics from the global angle to generate deviation;
Lossmlthe pseudo label prediction by the agent classifier of the Wassertein distance measuring network 1 and the pseudo label prediction by the label classifier of the network 2 are adopted to obtain the distribution difference, and the pseudo label prediction by the agent classifier of the Wassertein distance measuring network 2 and the pseudo label prediction by the label classifier of the network 2 are adopted to obtain the mutual guidance and mutual learning effects of the two networks;
Loss＝Losscls+Lossdom+Lossml。
optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes that the same kind of data set is transferred under different working conditions:
on the basis of the first hypothesis, the second hypothesis and the third hypothesis, migrating the diagnosis knowledge learned from the source domain data to the working condition with insufficient data quantity so as to improve the diagnosis effect;
the first assumption includes: bearing or gear vibration signals collected under different working conditions are different;
the second assumption includes: the rotating machinery operates under the condition of variable working conditions;
the third assumption includes: insufficient data is collected under at least one condition, wherein enough mark data is available as source domain data under one condition;
the differences of the vibration signals of the bearings or gears collected under different working conditions comprise different loads and/or different rotating speeds.
Optionally, in the method for diagnosing a fault of a rotating machine based on mutual local countermeasure transfer learning, the method further includes transferring between different data sets:
faults in a real industrial scene occur rarely, collected data are few, and a good-effect model is difficult to train. At the moment, bearing data and bearing diagnosis knowledge simulated by a laboratory can be collected, and the bearing data and the bearing diagnosis knowledge are adopted to identify the health state of the bearing in an engineering scene so as to be used for bearing fault diagnosis in the engineering scene.
The inventor of the present invention has found through research that the feature-based method in the prior art mainly comprises: the method mainly comprises a distance measurement mode and a domain counterstudy mode, wherein the distance measurement mode indicates that the domain invariant feature between the source domain and the target domain is obtained if the distance measurement of the features in the two domains is smaller, the domain counterstudy mode indicates that the domain invariant feature between the two domains is obtained if the learned features cannot be distinguished by a domain discriminator, and the domain invariant feature between the two domains is obtained if the learned features cannot be distinguished by the domain discriminator.
However, the inventor has observed that although the current feature-based method has achieved a good diagnostic effect, in this way, the tag information in the target domain data is not fully utilized, the tag information in the target domain data is ignored, and training only with the tagged data of the source domain has a certain effect on the migration effect. In addition, the existing method does not consider fault category information, and aligns data of a source domain and target domain from a global perspective when aligning features, so that the problem of poor migration effect caused by alignment deviation between different categories exists. Therefore, in the invention, the inventor designs a transfer learning fault diagnosis algorithm by combining the ideas of mutual learning and local countermeasure domain adaptation.
In the rotating machinery fault diagnosis method based on mutual local countermeasure transfer learning, a novel mutual learning network structure is designed for fault diagnosis, compared with the traditional single network structure, the robustness of a model is greatly improved, and each network not only adopts marked source domain fault data in the training process, but also refers to the learning experience of a companion network to further improve the generalization capability. In the whole process, learning experience is continuously shared between two networks, mutual learning and mutual progress are realized, in addition, a plurality of local anti-domain classifiers are designed to replace a global anti-domain classifier, the source domain and the target domain characteristics of each category are respectively aligned, and better similar characteristics can be obtained so as to obtain better similar characteristics. In summary, according to the above scheme, the fault diagnosis method based on mutual local countermeasure transfer learning is realized, and a good diagnosis effect can be obtained through the method under the situation that the data amount is insufficient in a real industrial scene.
The rotary machine fault diagnosis method based on mutual local countermeasure transfer learning provided by the invention also has the following advantages:
1. the domain invariant features are automatically learned completely through a neural network without prior knowledge.
2. 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.
3. By adopting a structure that two networks learn each other, the two classification models can synchronously deduce the label distribution of the target domain to guide each other. The method considers that the label information of the target domain can be utilized in the training process, and the classifier is trained by utilizing the source domain target domain data in a mode of generating the pseudo label, so that the method is more robust than the classifier which is obtained by only adopting the source domain labeled data in the similar scheme, and the problem that the target domain data in the existing migration learning fault diagnosis algorithm is not fully utilized is solved.
4. A plurality of local anti-domain classifiers are designed to replace a global anti-domain classifier, the source domain and target domain characteristics of each category are respectively aligned, better similar characteristics can be obtained, and the problem of poor alignment effect of the similar characteristics of the source domain and the target domain is solved.
5. The fault diagnosis algorithm with good diagnosis effect is provided for the scenes that fault marking data are insufficient or no mark fault data exist completely.
6. Compared with the traditional fault diagnosis algorithm of a single network structure, the scheme can obtain a more robust model due to mutual guidance of two networks.
7. Compared with some migration fault diagnosis algorithms for aligning the characteristics of the source domain and the target domain from the overall angle, the method and the device respectively align the characteristics aiming at each type of the fault, greatly reduce the probability of negative migration and further improve the migration effect.
Drawings
FIG. 1 is a schematic diagram of a fault diagnosis method for a rotating machine based on mutual local countermeasure transfer learning according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a fault diagnosis method for a rotating machine based on mutual local countermeasure transfer learning according to another 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 exaggerated and not necessarily to scale for illustrative purposes. In the figures, identical or functionally identical components are provided with the same reference symbols.
In the present invention, "disposed on …", "disposed over …" and "disposed over …" do not exclude the presence of an intermediate therebetween, unless otherwise specified. Further, "disposed on or above …" merely indicates the relative positional relationship between two components, and may also be converted to "disposed below or below …" and vice versa in certain cases, such as after reversing the product direction.
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 invention, the terms "perpendicular", "parallel" and the like in the directions of the tables also cover 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 for diagnosing faults of a rotating machine based on mutual local antagonistic transfer learning, which is further described in detail 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 merely for the purpose of facilitating and distinctly claiming the embodiments of the present invention.
The invention aims to provide a rotary machine fault diagnosis method based on mutual local antagonistic transfer learning, and aims to solve the problems that the existing transfer learning fault diagnosis method is poor in transfer effect and weak in model robustness.
In order to achieve the above object, the present invention provides a rotating machine fault diagnosis method based on mutual local countermeasure transfer learning, including: the mutual learning network structure is utilized to carry out fault diagnosis, and the two networks mutually guide to obtain a more reliable pseudo label, so that target domain data can be fully utilized, and the model effect and the robustness thereof are improved; and providing a plurality of local contrast domain classifiers in place of the global contrast domain classifier, each local contrast domain classifier aligning the source domain and target domain features of each class separately for improved similarity features.
Aiming at the problems in the prior art, the inventor ingeniously conceives that the scheme can simultaneously overcome a plurality of technical problems and realize unexpected technical effects, and the scheme is specifically shown in figure 1: the invention provides a rotary machine fault diagnosis method based on mutual local countermeasure transfer learning, which comprises the following steps: the mutual learning network structure is utilized to carry out fault diagnosis, and the two networks mutually guide to obtain a more reliable pseudo label, so that target domain data can be fully utilized, and the model effect and the robustness thereof are improved; and providing a plurality of local contrast domain classifiers in place of the global contrast domain classifier, each local contrast domain classifier aligning the source domain and target domain features of each class separately for improved similarity features. For example:
in an embodiment of the present invention, a distance function needs to be predefined to learn the similar features of the source domain and the target domain through a domain adaptation method based on distance measurement, and how to select an appropriate distance function needs to be determined according to a specific task. Similar characteristics of the source domain and the target domain can be automatically learned by adopting an adaptation mode of the anti-domain. The existing methods adopting the adaptation of the anti-domain are characterized in that the source domain and the target domain are regarded as two whole bodies to align the features from the global angle, and the fault category is not considered;
most of the existing methods adopt a feature alignment mode, label information of a target domain is ignored, and the target domain has no label. The invention adopts a structure of mutual learning, and predicts a reliable pseudo label for the target domain data, thereby utilizing the label information of the target domain, and gradually improving the credibility of the pseudo label in the mutual learning and mutual guiding process of two networks, thereby fully utilizing the data of the source domain and the target domain to train the model;
in one embodiment of the invention, the Wasserstein distance is used to minimize the distribution difference between the pseudo label predicted by the label classifier of one network on the target domain data and the pseudo label predicted by the proxy classifier of the other network on the target domain data to guide the two networks to learn each other, and the Wasserstein distance has the advantage over the KL divergence and the JS divergence that the Wasserstein distance can reflect the distance between the two distributions even if the distributions are not overlapped. And the JS divergence is a constant of 2log2 in this case, and the KL divergence is probably meaningless, so that the problem of gradient disappearance in the network optimization process is avoided.
Therefore, the improvement points of the invention mainly lie in that:
firstly: and a network structure which learns mutually is designed for fault diagnosis, and compared with the traditional single network structure, the robustness of the model is greatly improved. In the training process, each network not only adopts the labeled source domain fault data, but also refers to the learning experience of the companion network to further improve the generalization capability. In the whole process, learning experience is continuously shared between the two networks, and mutual learning and common progress are realized;
secondly, the method comprises the following steps: a plurality of local anti-domain classifiers are designed to replace a global anti-domain classifier, and the source domain and target domain features of each category are aligned respectively, so that better similar features can be obtained.
Specifically, the reason why the mutual learning structure works is that: the class probability estimation contains the essential rule of the data mined by the network, and can reflect the intrinsic essential characteristics of the data. The information provided by the single truth label only contains whether the sample belongs to a certain class, but lacks the connection among different classes, and the class probability estimation output by the network can recover the information to a certain extent, so that the interaction of the class probability estimation between the networks can transfer the learned data distribution characteristics, thereby helping the network to improve the generalization performance. Therefore, the invention adopts such a mutual learning structure to exchange the data distribution characteristics of each network transmission learning, mutually guide and mutually learn until convergence.
In one embodiment of the invention, the Wasserstein distance is used to constrain the class probability distribution differences to a minimum to function as a regularization. When the class probability of one network is transmitted to the other network, additional class prior constraint is provided essentially, so that the label classifier is prevented from over-fitting 0-1 distribution of the truth label, and the over-fitting occurrence probability is effectively reduced.
In one embodiment of the invention, the proxy classifier acts as a bridge between the tag classifier and the feature extractor, ensuring that they use the perceptual information in the target domain. Another advantage of proxy classifiers is that they eliminate the need for label classifiers to be aligned with each other. Conventional mutual learning structures guide each group with another group of prediction labels but do not tell them how to obtain these pseudo labels, and therefore do not force them to obtain very similar signatures, i.e. two networks may obtain dissimilar signatures, which may affect each other less effectively. The structure of the proxy classifier tells the other party how to obtain these pseudo labels, and the predicted value of each step must have the smallest difference from the distribution of the predicted values of the other party's proxy classifier.
In one embodiment of the invention, the local domain discriminators perform fine-grained alignment of the distribution of different types of fault data, so that the probability of negative migration is reduced, because for a target domain sample, the local domain discriminator is only aligned with the most relevant type data of the local domain discriminator, and therefore the wrong alignment of the discrimination structure in different distributions is avoided.
In one embodiment of the present invention, the present invention mainly includes 3 modules, as shown in fig. 2, a data preprocessing module, a feature extraction module, and a domain adaptation module. Number ofThe data preprocessing module is used for preprocessing the vibration signals collected on the rotating mechanical equipment; the feature extraction module constructs a multilayer CNN convolution network extraction feature, and extracts features of different scales by setting different initialization parameters; the domain adaptation module designs 3 Loss functions to form a final Loss function, namely source domain label data classification Softmax multi-classification LossclsSoftmax two-class domain discriminator LossdomAnd Loss of mutual learning structureml. By minimizing the final Loss function, the domain invariant characteristics of the source domain and the target domain can be obtained through training, and a good migration learning fault diagnosis effect is achieved; wherein LossclsA loss function that tags data for the source domain, since this loss function is supervised, can ensure that the entire model is valid.
In one embodiment of the invention, LossdomThe method comprises k Softmax classification domain discriminators, wherein k is the number of fault categories, compared with the traditional domain confrontation classifier, the source domain feature and the target domain feature are aligned from the global angle, and in consideration of the fact that the features of different fault types are different, the method adopts a plurality of local domain classifiers to respectively align the source domain feature and the target domain feature of each category.
In one embodiment of the invention, LossmlThe distance Wasserstein is adopted to measure the distribution difference of the prediction of the pseudo label by the proxy classifier of the network 1 and the prediction of the pseudo label by the label classifier of the network 2, and the distance Wasserstein is also adopted to measure the distribution difference of the prediction of the pseudo label by the proxy classifier of the network 2 and the prediction of the pseudo label by the label classifier of the network 2.
Loss＝Losscls+Lossdom+Lossml
In one embodiment of the invention, migration of the same data set under different working conditions is carried out, bearing or gear vibration signals collected under different working conditions are different, such as different loads and different rotating speeds, considering that the rotating machine is mostly operated under variable working conditions, and the data volume collected under some working conditions is possibly insufficient, if enough marking data exist under one working condition, the marking data can be used as source domain data, and a good diagnosis effect can be achieved by migrating diagnosis knowledge from source domain data to the working condition with insufficient data volume through the method.
In other words, based on the first, second and third assumptions, the diagnostic knowledge learned from the source domain data is migrated to the condition of insufficient data volume to improve the diagnostic effect; the first assumption includes: bearing or gear vibration signals collected under different working conditions are different; the second assumption includes: the rotating machinery operates under the condition of variable working conditions; the third assumption includes: insufficient data is collected under at least one condition, wherein enough mark data is available as source domain data under one condition; the differences of the vibration signals of the bearings or gears collected under different working conditions comprise different loads and/or different rotating speeds.
In one embodiment of the invention, migration between different data sets is carried out, faults are rarely generated in a real industrial scene, and collected data are rarely used for training a good model. Bearing data collected by laboratory simulation can be adopted, diagnosis knowledge from laboratory bearings can be helpful for identifying the health state of the bearings in an engineering scene, and the scheme can be used for bearing fault diagnosis in the engineering scene.
In summary, the invention provides a fault diagnosis method based on mutual local countermeasure transfer learning, and aims to solve the problem that a good diagnosis effect can be obtained through the method under the condition that the data volume is insufficient in a real industrial scene. The scheme mainly combines the ideas of mutual learning and local countermeasure domain adaptation and designs a transfer learning fault diagnosis algorithm. Its advantages include:
1. the domain invariant features are automatically learned completely through a neural network without prior knowledge.
2. 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.
3. By adopting a structure that two networks learn each other, the two classification models can synchronously deduce the label distribution of the target domain to guide each other. The method considers that the label information of the target domain can be utilized in the training process, and the classifier is trained by utilizing the source domain target domain data in a mode of generating the pseudo label, so that the method is more robust than the classifier which is obtained by only adopting the source domain labeled data in the similar scheme, and the problem that the target domain data in the existing migration learning fault diagnosis algorithm is not fully utilized is solved.
4. A plurality of local anti-domain classifiers are designed to replace a global anti-domain classifier, the source domain and target domain features of each category are respectively aligned, better similar features can be obtained, and the problem of poor alignment effect of the similar features of the source domain and the target domain is solved.
5. The fault diagnosis algorithm with good diagnosis effect is provided for the scenes that fault marking data are insufficient or no mark fault data exist completely.
6. Compared with the traditional fault diagnosis algorithm of a single network structure, the scheme of the invention can obtain a more robust model by mutual cooperation due to mutual guidance of two networks.
7. Compared with some migration fault diagnosis algorithms for aligning the characteristics of the source domain and the target domain from the overall angle, the method and the device respectively align the characteristics aiming at each type of the fault, greatly reduce the probability of negative migration and further improve the migration effect.
The invention provides a rotary machine fault diagnosis method based on mutual local countermeasure transfer learning, which comprises the following steps: considering that the label information hidden in the target domain data is not fully utilized, a mutual learning network structure is designed to learn reliable pseudo labels, and further the target domain data without labels is fully utilized, so that the model performance is improved. Is more robust than the model than a single network structure. Considering the problem that alignment deviation may occur when the data of the source domain and the target domain are aligned directly from the global perspective by ignoring the category information of the fault, a plurality of local anti-domain classifiers are provided to replace the global anti-domain classifier, and each local anti-domain classifier aligns the source domain and the target domain feature of each category respectively to obtain better similar features.
In summary, the above embodiments have described in detail different configurations of the rotating machine fault diagnosis method based on mutual local countermeasure migration learning, but it goes without saying 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.
The embodiments in the present description 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 rotating machine fault diagnosis method based on mutual local countermeasure transfer learning is characterized by comprising the following steps:
the mutual learning network structure is utilized to carry out fault diagnosis, and the two networks mutually guide to obtain a more reliable pseudo label, so that target domain data can be fully utilized, and the model effect and the robustness thereof are improved;
and providing a plurality of local contrast domain classifiers in place of the global contrast domain classifier, each local contrast domain classifier aligning the source domain and target domain features of each class separately for improved similarity features.
2. The method of claim 1, wherein the fault diagnosis of the rotating machine based on the mutual local countermeasure transfer learning comprises:
providing a mutual learning network structure, wherein the mutual learning network structure comprises at least two networks, and the at least two networks are mutual partner networks;
the generalization capability of each network is improved by referring to the learning experience of the partner network while marked source domain fault data is adopted in the training process; and
learning experience is shared between the two networks to carry out mutual learning, and fault diagnosis is carried out according to mutual learning results.
3. The rotating machine fault diagnosis method based on mutual local countermeasure transfer learning according to claim 2, characterized by further comprising:
the class probability estimation has a data essence rule extracted by a network, and reflects the intrinsic essence characteristic of the data;
the mutual learning structure carries out category probability estimation interaction to transmit the learned data distribution characteristics among the networks, so that the generalization performance of the networks is improved; and
and improving generalization performance based on the network, interchanging the data distribution characteristics of each network transmission study by adopting a mutual study structure, and mutually guiding and studying until convergence.
4. The rotating machine fault diagnosis method based on mutual local countermeasure transfer learning according to claim 3, characterized by further comprising:
the distribution difference of class probability estimation is constrained to be minimum by adopting Wasserstein distance, so that the regularization effect is achieved;
when the class probability estimation of one network is transmitted to the other network, additional class prior constraint is provided, the label classifier is prevented from being over-fitted to the 0-1 distribution of the truth label, and the over-fitting occurrence probability is effectively reduced.
5. The rotating machine fault diagnosis method based on mutual local countermeasure transfer learning according to claim 4, characterized by further comprising:
the proxy classifier serves as a bridge between the label classifier and the feature extractor, and the proxy classifier, the label classifier and the feature extractor are ensured to use label information in the target domain;
the agent classifier ensures that the two label classifiers are not required to be aligned with each other forcibly, so that two independent networks are prevented from convincing the prediction result of the other side, and the agent classifier communicates the two networks in the training process to play a role in exchanging information learned by the two networks; and
the structure of the proxy classifier enables the following actions to be performed between the label classifier and the feature extractor:
and indicating how the other party obtains the pseudo label, wherein the distribution difference between the predicted value of each step and the predicted value of the agent classifier of the other party is minimum, so as to avoid the problem that the traditional mutual learning structure forces the two networks to convince the predicted result of the other party due to the fact that how the pseudo label is not indicated, and the diagnosis effect is not good.
6. The rotating machine fault diagnosis method based on mutual local countermeasure transfer learning according to claim 5, characterized by further comprising:
based on the fact that one target domain sample is only aligned with the most relevant class data of the target domain sample, the local domain discriminators conduct fine-grained alignment of different classes of fault data distribution, the probability of negative migration is reduced, and therefore the wrong alignment of discrimination structures in different distributions is avoided.
7. The rotating machine fault diagnosis method based on mutual local countermeasure transfer learning according to claim 3, characterized by further comprising:
the data preprocessing module is used for preprocessing the vibration signals collected on the rotating mechanical equipment;
the feature extraction module constructs a multilayer CNN convolution network extraction feature, and extracts features of different scales by setting different initialization parameters;
the domain adaptation module designs 3 Loss functions to form a final Loss function, wherein the 3 Loss functions comprise source domain label data classification Softmax multi-classification LossclsSoftmax two-classification domain discriminator LossdomAnd Loss of mutual learning structureml(ii) a And
training to obtain the domain invariant characteristics of the source domain and the target domain by minimizing the final Loss function so as to improve the fault diagnosis effect of the transfer learning;
wherein LossclsAnd a loss function with label data in the source domain has supervision characteristics based on the loss function, so that the effect of the whole model is ensured.
8. The rotating machine fault diagnosis method based on mutual local countermeasure transfer learning according to claim 7, characterized by further comprising:
Lossdomthe system comprises k Softmax two-classification domain discriminators, wherein k is the number of fault classes, and a plurality of local domain classifiers are adopted to align the source domain and the target domain characteristics of each class respectively based on different fault types so as to avoid the domain countermeasure classifier from aligning the source domain and the target domain characteristics from the global angle to generate deviation;
Lossmlthe pseudo label prediction by the agent classifier of the Wassertein distance measuring network 1 and the pseudo label prediction by the label classifier of the network 2 are adopted to obtain the distribution difference, and the pseudo label prediction by the agent classifier of the Wassertein distance measuring network 2 and the pseudo label prediction by the label classifier of the network 2 are adopted to obtain the mutual guidance and mutual learning effects of the two networks;
Loss＝Losscls+Lossdom+Lossdml。
9. a method as claimed in claim 3, further comprising migrating the same kind of data set under different conditions:
on the basis of the first hypothesis, the second hypothesis and the third hypothesis, migrating the diagnosis knowledge learned from the source domain data to the working condition with insufficient data quantity so as to improve the diagnosis effect;
the first assumption includes: bearing or gear vibration signals collected under different working conditions are different;
the second assumption includes: the rotating machinery operates under the condition of variable working conditions;
the third assumption includes: insufficient data is collected under at least one condition, wherein enough mark data is available as source domain data under one condition;
the differences of the vibration signals of the bearings or gears collected under different working conditions comprise different loads and/or different rotating speeds.
10. A method of diagnosing faults in a rotating machine based on mutual local countermeasure transfer learning as claimed in claim 3, further comprising transferring between different data sets:
and collecting bearing data and bearing diagnosis knowledge simulated in a laboratory, and identifying the health state of the bearing in an engineering scene by adopting the bearing data and the bearing diagnosis knowledge so as to be used for bearing fault diagnosis in the engineering scene.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111362333.5A CN114021285B (en) | 2021-11-17 | 2021-11-17 | Rotary machine fault diagnosis method based on mutual local countermeasure migration learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111362333.5A CN114021285B (en) | 2021-11-17 | 2021-11-17 | Rotary machine fault diagnosis method based on mutual local countermeasure migration learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114021285A true CN114021285A (en) | 2022-02-08 |
| CN114021285B CN114021285B (en) | 2024-04-12 |
Family
ID=80064692
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111362333.5A Active CN114021285B (en) | 2021-11-17 | 2021-11-17 | Rotary machine fault diagnosis method based on mutual local countermeasure migration learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114021285B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114580571A (en) * | 2022-04-01 | 2022-06-03 | 南通大学 | Small sample power equipment image classification method based on migration mutual learning |
Citations (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103868692A (en) * | 2014-03-18 | 2014-06-18 | 电子科技大学 | Rotary machine fault diagnosis method based on kernel density estimation and K-L divergence |
| CN106813921A (en) * | 2016-12-24 | 2017-06-09 | 广东石油化工学院 | A kind of combined failure of rotating machinery diagnostic method |
| CN106980822A (en) * | 2017-03-14 | 2017-07-25 | 北京航空航天大学 | A kind of rotary machinery fault diagnosis method learnt based on selective ensemble |
| CN107817787A (en) * | 2017-11-29 | 2018-03-20 | 华南理工大学 | A kind of intelligent producing line robotic failure diagnostic method based on machine learning |
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN109947086A (en) * | 2019-04-11 | 2019-06-28 | 清华大学 | Mechanical breakdown migration diagnostic method and system based on confrontation study |
| CN110189278A (en) * | 2019-06-06 | 2019-08-30 | 上海大学 | A kind of binocular scene image repair method based on generation confrontation network |
| CN110287837A (en) * | 2019-06-17 | 2019-09-27 | 上海大学 | Sea obstacle detection method based on prior estimate network and space constraint mixed model |
| CN110414383A (en) * | 2019-07-11 | 2019-11-05 | 华中科技大学 | Convolutional neural networks based on Wasserstein distance fight transfer learning method and its application |
| CN110428004A (en) * | 2019-07-31 | 2019-11-08 | 中南大学 | Component of machine method for diagnosing faults under data are unbalance based on deep learning |
| CN110751207A (en) * | 2019-10-18 | 2020-02-04 | 四川大学 | Fault diagnosis method for anti-migration learning based on deep convolution domain |
| CN110907176A (en) * | 2019-09-30 | 2020-03-24 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
| CN111060318A (en) * | 2020-01-09 | 2020-04-24 | 山东科技大学 | Bearing fault diagnosis method based on deep countermeasure migration network |
| CN111709448A (en) * | 2020-05-20 | 2020-09-25 | 西安交通大学 | Mechanical fault diagnosis method based on migration relation network |
| CN111860677A (en) * | 2020-07-29 | 2020-10-30 | 湖南科技大学 | Rolling bearing transfer learning fault diagnosis method based on partial domain confrontation |
| CN111898634A (en) * | 2020-06-22 | 2020-11-06 | 西安交通大学 | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption |
| CN112101085A (en) * | 2020-07-22 | 2020-12-18 | 西安交通大学 | Adaptive intelligent fault diagnosis method based on importance weighted domain impedance |
| CN112183581A (en) * | 2020-09-07 | 2021-01-05 | 华南理工大学 | Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network |
| CN112906644A (en) * | 2021-03-22 | 2021-06-04 | 重庆大学 | Mechanical fault intelligent diagnosis method based on deep migration learning |
| CN113076834A (en) * | 2021-03-25 | 2021-07-06 | 华中科技大学 | Rotating machine fault information processing method, processing system, processing terminal, and medium |
| CN113159126A (en) * | 2021-03-22 | 2021-07-23 | 河海大学 | Industrial Internet of things equipment fault diagnosis method based on universal domain adaptation |
| CN113176776A (en) * | 2021-03-03 | 2021-07-27 | 上海大学 | Unmanned ship weather self-adaptive obstacle avoidance method based on deep reinforcement learning |
| CN113392881A (en) * | 2021-05-27 | 2021-09-14 | 重庆大学 | Rotary machine fault diagnosis method based on transfer learning |
| CN113505664A (en) * | 2021-06-28 | 2021-10-15 | 上海电力大学 | Fault diagnosis method for planetary gear box of wind turbine generator |
-
2021
- 2021-11-17 CN CN202111362333.5A patent/CN114021285B/en active Active
Patent Citations (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103868692A (en) * | 2014-03-18 | 2014-06-18 | 电子科技大学 | Rotary machine fault diagnosis method based on kernel density estimation and K-L divergence |
| CN106813921A (en) * | 2016-12-24 | 2017-06-09 | 广东石油化工学院 | A kind of combined failure of rotating machinery diagnostic method |
| CN106980822A (en) * | 2017-03-14 | 2017-07-25 | 北京航空航天大学 | A kind of rotary machinery fault diagnosis method learnt based on selective ensemble |
| CN107817787A (en) * | 2017-11-29 | 2018-03-20 | 华南理工大学 | A kind of intelligent producing line robotic failure diagnostic method based on machine learning |
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN109947086A (en) * | 2019-04-11 | 2019-06-28 | 清华大学 | Mechanical breakdown migration diagnostic method and system based on confrontation study |
| CN110189278A (en) * | 2019-06-06 | 2019-08-30 | 上海大学 | A kind of binocular scene image repair method based on generation confrontation network |
| CN110287837A (en) * | 2019-06-17 | 2019-09-27 | 上海大学 | Sea obstacle detection method based on prior estimate network and space constraint mixed model |
| CN110414383A (en) * | 2019-07-11 | 2019-11-05 | 华中科技大学 | Convolutional neural networks based on Wasserstein distance fight transfer learning method and its application |
| CN110428004A (en) * | 2019-07-31 | 2019-11-08 | 中南大学 | Component of machine method for diagnosing faults under data are unbalance based on deep learning |
| CN110907176A (en) * | 2019-09-30 | 2020-03-24 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
| CN110751207A (en) * | 2019-10-18 | 2020-02-04 | 四川大学 | Fault diagnosis method for anti-migration learning based on deep convolution domain |
| CN111060318A (en) * | 2020-01-09 | 2020-04-24 | 山东科技大学 | Bearing fault diagnosis method based on deep countermeasure migration network |
| CN111709448A (en) * | 2020-05-20 | 2020-09-25 | 西安交通大学 | Mechanical fault diagnosis method based on migration relation network |
| CN111898634A (en) * | 2020-06-22 | 2020-11-06 | 西安交通大学 | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption |
| CN112101085A (en) * | 2020-07-22 | 2020-12-18 | 西安交通大学 | Adaptive intelligent fault diagnosis method based on importance weighted domain impedance |
| CN111860677A (en) * | 2020-07-29 | 2020-10-30 | 湖南科技大学 | Rolling bearing transfer learning fault diagnosis method based on partial domain confrontation |
| CN112183581A (en) * | 2020-09-07 | 2021-01-05 | 华南理工大学 | Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network |
| CN113176776A (en) * | 2021-03-03 | 2021-07-27 | 上海大学 | Unmanned ship weather self-adaptive obstacle avoidance method based on deep reinforcement learning |
| CN112906644A (en) * | 2021-03-22 | 2021-06-04 | 重庆大学 | Mechanical fault intelligent diagnosis method based on deep migration learning |
| CN113159126A (en) * | 2021-03-22 | 2021-07-23 | 河海大学 | Industrial Internet of things equipment fault diagnosis method based on universal domain adaptation |
| CN113076834A (en) * | 2021-03-25 | 2021-07-06 | 华中科技大学 | Rotating machine fault information processing method, processing system, processing terminal, and medium |
| CN113392881A (en) * | 2021-05-27 | 2021-09-14 | 重庆大学 | Rotary machine fault diagnosis method based on transfer learning |
| CN113505664A (en) * | 2021-06-28 | 2021-10-15 | 上海电力大学 | Fault diagnosis method for planetary gear box of wind turbine generator |
Non-Patent Citations (3)
| Title |
|---|
| SHAO, JIAJIE ET AL.: "Rotating machinery fault diagnosis by deep adversarial transfer learning based on subdomain adaptation", 《ADVANCES IN MECHANICAL ENGINEERING》, vol. 13, no. 8, 29 September 2021 (2021-09-29) * |
| 彭艳;陈加宏;李小毛;罗均;谢少荣;刘畅;蒲华燕;: "时空上下文融合的无人艇海面目标跟踪", 中国科学:技术科学, no. 12, 23 November 2018 (2018-11-23) * |
| 文成林;吕菲亚;: "基于深度学习的故障诊断方法综述", 电子与信息学报, no. 01, 15 January 2020 (2020-01-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114580571A (en) * | 2022-04-01 | 2022-06-03 | 南通大学 | Small sample power equipment image classification method based on migration mutual learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114021285B (en) | 2024-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hao et al. | An end-to-end architecture for class-incremental object detection with knowledge distillation | |
| WO2021008328A1 (en) | Image processing method and device, terminal, and storage medium | |
| CN110907176B (en) | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network | |
| CN110163236B (en) | Model training method and device, storage medium and electronic device | |
| CN101907681B (en) | Analog circuit dynamic online failure diagnosing method based on GSD-SVDD | |
| EP3767536A1 (en) | Latent code for unsupervised domain adaptation | |
| CN107316049A (en) | A kind of transfer learning sorting technique based on semi-supervised self-training | |
| Xu et al. | Bridging the domain gap for multi-agent perception | |
| CN114048568B (en) | Rotary machine fault diagnosis method based on multisource migration fusion shrinkage framework | |
| CN111368690A (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
| CN111950525B (en) | Fine-grained image classification method based on destructive reconstruction learning and GoogLeNet | |
| CN113076994A (en) | Open-set domain self-adaptive image classification method and system | |
| Yang et al. | Detecting 11k classes: Large scale object detection without fine-grained bounding boxes | |
| Li et al. | A review of deep learning methods for pixel-level crack detection | |
| CN112115993A (en) | Zero sample and small sample evidence photo anomaly detection method based on meta-learning | |
| CN114782752B (en) | Small sample image integrated classification method and device based on self-training | |
| CN114021285A (en) | Rotary machine fault diagnosis method based on mutual local countermeasure transfer learning | |
| Realpe et al. | Multi-sensor fusion module in a fault tolerant perception system for autonomous vehicles | |
| CN111209935A (en) | Unsupervised target detection method and system based on self-adaptive domain transfer | |
| Lonij et al. | Open-world visual recognition using knowledge graphs | |
| CN115630299A (en) | Rotary machine fault diagnosis method and system based on joint domain adaptive network | |
| Xue | Computational diagnostic classification model using deep feedforward network based semi-supervised learning | |
| CN114897063A (en) | Indoor positioning method based on-line pseudo label semi-supervised learning and personalized federal learning | |
| Katsamenis et al. | Evaluating YOLO transferability limitation for road infrastructures monitoring | |
| Li et al. | A Novel Semi-supervised Adaboost Technique Based on Improved Tri-training |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |