An intelligent fault diagnosis method and system for a rotating machine

Abstract

The present application belongs to the technical field of intelligent operation and maintenance of industrial equipment, and discloses an intelligent fault diagnosis method and system for rotating machinery, wherein the method comprises the following steps: S1, constructing a training sample set comprising sample signals of multiple source domains, each sample signal comprising a vibration signal, an acoustic signal and a fault label; S2, constructing a fault diagnosis model comprising: a vibration feature extractor for extracting vibration features, an acoustic feature extractor for extracting acoustic features, a feature fusion module for fusing the vibration features and the acoustic features to obtain connection features, a feature mapping module for extracting unified features from the connection features, and a classification module for classifying fault modes according to the extracted features; S3, training the fault diagnosis model based on the training sample set; and S4, performing online fault diagnosis by using the trained fault diagnosis model. The present application learns invariant unified fault representation from multi-source domain vibration and acoustic signals, thereby improving the diagnosis accuracy of the model across domains.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent operation and maintenance of industrial equipment, and more particularly to an intelligent fault diagnosis method and system for rotating machinery. BACKGROUND

[0002] Rotating machinery is widely used in modern industry. Rotating machinery failure can cause serious safety hazards and result in huge maintenance costs. To ensure normal operation of equipment, equipment operation and maintenance departments are often set up in industrial production to diagnose and maintain equipment failures. The fault diagnosis of equipment is gradually shifting from traditional signal analysis methods to intelligent diagnosis methods based on machine learning.

[0003] In recent years, with the development of sensors and artificial intelligence technology, deep learning has been increasingly applied to mechanical fault diagnosis and has achieved good results. Common diagnostic models include convolutional neural networks, recurrent neural networks, and autoencoders. These deep learning models with good results are inseparable from sufficient labeled training samples. That is, existing intelligent fault diagnosis technology relies on sufficient labeled training data, and the collected samples need to meet the same distribution condition. In order to accurately implement fault diagnosis, samples of the target working condition are usually collected for training, which also makes the trained diagnostic model only accurate for a specific target working condition. Once the equipment working condition changes, the model will be difficult to make accurate judgments.

[0004] Due to the variety of operating conditions of machine equipment and the difference in physical characteristics of different machine equipment, the difference in data distribution between the training stage and the application stage will reduce the effectiveness of the diagnostic model. That is, most of the existing intelligent fault diagnosis methods have poor generalization ability and poor diagnostic accuracy when the data distribution is different between the training stage and the application stage. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides an intelligent fault diagnosis method and system for rotating machinery, which solves the problem that most of the existing intelligent fault diagnosis methods have poor generalization ability and poor diagnostic accuracy when the data distribution is different between the training stage and the application stage.

[0006] To achieve the above purpose, according to one aspect of the present application, an intelligent fault diagnosis method for rotating machinery is provided, comprising:

[0007] Offline data processing and training stage:

[0008] S1, a training sample set is constructed, which includes sample signals of multiple source domains, and each sample signal includes a vibration signal, an acoustic signal, and a fault label;

[0009] S2, the fault diagnosis model comprises a vibration feature extractor, an acoustic feature extractor, a feature fusion module, a feature mapping module and a classification module, the vibration feature extractor is used for extracting vibration features from the vibration signals, the acoustic feature extractor is used for extracting acoustic features from the acoustic signals, the feature fusion module is used for fusing the vibration features and the acoustic features to obtain connection features, the feature mapping module is used for extracting unified features from the connection features, and the classification module is used for classifying fault modes according to the vibration features, the acoustic features and the unified features;

[0010] S3, training the fault diagnosis model based on the training sample set to obtain a trained fault diagnosis model;

[0011] The online fault diagnosis stage comprises:

[0012] S4, using the trained fault diagnosis model to perform online fault diagnosis on the rotating machinery.

[0013] According to the intelligent fault diagnosis method for rotating machinery provided by the application, the feature fusion module in S2 is used to fuse the vibration features and the acoustic features from the same source domain to obtain same-domain connection features, and fuse the vibration features and the acoustic features from different source domains to obtain cross-domain connection features.

[0014] Correspondingly, the feature mapping module is used to extract same-domain unified features from the same-domain connection features and extract cross-domain unified features from the cross-domain connection features.

[0015] According to the intelligent fault diagnosis method for rotating machinery provided by the application, the same-domain connection features z id Specifically,

[0016]

[0017] The cross-domain connection features z cd Specifically,

[0018]

[0019] Wherein, is the vibration feature belonging to the mth class of the ith source domain; is the acoustic feature belonging to the mth class of the jth source domain.

[0020] According to the intelligent fault diagnosis method for rotating machinery provided by the application, S3 comprises training the feature mapping module to maximize the mutual information between the same-domain unified features and the cross-domain unified features.

[0021] The intelligent fault diagnosis method for a rotating machine provided by the application trains the feature mapping module to maximize mutual information between the intra-domain unified features and the cross-domain unified features, and specifically comprises the following steps.

[0022] The mutual information between the intra-domain unified features s id and the cross-domain unified features s cd is specifically shown in the following formula:

[0023]

[0024] wherein I represents mutual information; is an expectation; W and N are batch numbers of samples; f is an evaluation function; is the i-th intra-domain unified feature; is the j-th cross-domain unified feature;

[0025] The mutual information maximization is realized by maximizing the lower bound of mutual information, and the supervised contrast loss shown in the following formula is used to maximize the lower bound of mutual information The feature mapping module is trained as follows:

[0026]

[0027] wherein k∈K={1,2,…,2W}; A(k)≡K / k; P(k) is a positive sample feature set with the same label as the sample k; s=s id ∪s cd ; s i is the i-th feature; s p is the p-th feature; s a is a negative sample feature; and τ is a scaling parameter.

[0028] According to the intelligent fault diagnosis method for a rotating machine provided by the application, the classification module in S2 comprises a vibration state classifier, an acoustic state classifier and a combined state classifier, the vibration state classifier is used to classify fault modes according to the vibration features, the acoustic state classifier is used to classify fault modes according to the acoustic features, and the combined state classifier is used to classify fault modes according to the unified features, and the output result of the classification module is obtained by applying the majority voting method to the three state classifiers.

[0029] According to the intelligent fault diagnosis method for a rotating machine provided by the application, S3 further comprises the following steps.

[0030] The classification losses of the vibration state classifier, the acoustic state classifier and the combined state classifier are respectively obtained, and then the total classification loss is obtained;

[0031] training the vibration state classifier, the acoustic state classifier and the combined state classifier based on the total classification loss.

[0032] According to the intelligent fault diagnosis method for a rotating machine provided by the application, the fault diagnosis model further comprises a modal converter, the modal converter is used to generate false acoustic features according to the vibration features, and generate false vibration features according to the acoustic features.

[0033] Correspondingly, S4 comprises: acquiring vibration signals and / or acoustic signals of the rotating machine, and performing online fault diagnosis on the basis of the vibration signals and / or acoustic signals by using the trained fault diagnosis model.

[0034] According to the intelligent fault diagnosis method for a rotating machine provided by the application, S3 further comprises: adopting a semantic supervision loss shown in the following formula training the modal converter:

[0035]

[0036] wherein, L CE is a cross-entropy function; is a category probability distribution output by the vibration state classifier based on the false acoustic features; is a category probability distribution output by the acoustic state classifier based on the false vibration features; is a label; and indicates that the parameters of the vibration state classifier and the acoustic state classifier are not updated in the training process.

[0037] According to another aspect of the application, an intelligent fault diagnosis system for a rotating machine is provided, the system comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the intelligent fault diagnosis method for a rotating machine according to any one of the above aspects.

[0038] Overall, compared with the prior art, the intelligent fault diagnosis method and system for a rotating machine provided by the application have the following advantages:

[0039] 1. The model training stage simultaneously utilizes vibration signals and acoustic signals to improve the resistance of the fault diagnosis model to data distribution differences, and the test stage uses multiple sensors to collect signals of different modalities (vibration and acoustic), which is beneficial for comprehensive monitoring of the equipment; the designed feature fusion module connects vibration features and acoustic features to make the connected features cover more information, overcoming the limitations of single-modality information, and then the feature mapping module maps the unified features, which is beneficial for better learning of invariant features, learning of potential commonalities in multi-modal and multi-field data from the decoupling perspective, exploration of an invariant unified feature space for capturing robust features from different modalities and cross-domain data, and improvement of the model generalization ability to maintain high diagnostic accuracy in cross-domain diagnosis;

[0040] 2. The design of the feature fusion module specifically generates intra-domain connected features and cross-domain connected features, which increases the diversity of features and improves the utilization rate of useful information;

[0041] 3. The intra-domain unified features after mapping of the intra-domain connected features maintain consistent potential field-specific information, so that they are similar to each other and thus close to each other in the unified feature space, on the contrary, the cross-domain unified features after mapping of the cross-domain connected features contain different field-specific information, resulting in relatively dispersed positions in the space, by maximizing the mutual information between the intra-domain and cross-domain unified features, the field-specific information is effectively removed and the intra-class difference is reduced; the feature mapping module projects the connected multi-modal features onto the invariant unified features by maximizing the mutual information between the intra-domain and cross-domain unified features, the invariant unified feature space sufficiently retains the features related to faults and robust to domain migration, and removes the redundant information that may interfere with model decision-making, thereby improving the model generalization ability and diagnostic accuracy;

[0042] 4. Three state classifiers are set, and a hybrid integrated diagnosis strategy is proposed, which is beneficial for improving the flexibility and adaptability of the diagnosis model to various tasks and environments;

[0043] 5. A modal converter is equipped to complete the conversion between modalities when some modal signals are missing, the modal converter can generate false data of the missing modality according to the available monitoring data, and through semantic supervision, the generated synthetic data can be correctly recognized by the respective state classifier, thereby improving the robustness and applicability of the diagnosis model to various external restrictions. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is the model structure diagram of the intelligent fault diagnosis method for a rotating machine provided by the application;

[0045] Figure 2 is a schematic diagram of the information enhancement connection method provided by the application;

[0046] Figure 3 is a mutual information maximization schematic diagram provided by the present application;

[0047] Figure 4 is a feature modal conversion schematic diagram provided by the present application. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0049] Please refer to Figure 1 The embodiment provides an intelligent fault diagnosis method for a rotating machine, which comprises the following steps of:

[0050] Offline data processing and training phase:

[0051] S1, a training sample set is constructed, wherein the training sample set comprises sample signals of multiple source domains, and each sample signal comprises a vibration signal, an acoustic signal and a fault label;

[0052] S2, a fault diagnosis model is constructed, which comprises a vibration feature extractor, an acoustic feature extractor, a feature fusion module, a feature mapping module and a classification module, the vibration feature extractor is used to extract vibration features from the vibration signal, the acoustic feature extractor is used to extract acoustic features from the acoustic signal, the feature fusion module is used to fuse the vibration features and the acoustic features to obtain connection features, the feature mapping module is used to extract unified features from the connection features, and the classification module is used to classify fault patterns according to the vibration features, the acoustic features and the unified features;

[0053] S3, the fault diagnosis model is trained based on the training sample set to obtain a trained fault diagnosis model;

[0054] Online fault diagnosis phase:

[0055] S4, the trained fault diagnosis model is used to perform online fault diagnosis on the rotating machine.

[0056] The training sample set in the embodiment is constructed for rotating machinery, which can be bearings, gearboxes, motors, etc. Different source domains correspond to different working conditions and / or devices to which the rotating machinery is applied. That is, different source domain data can be obtained by changing the working condition in one device, or the device can be changed to collect source domain data in different devices, as long as the sample data is for the same type of rotating machinery. Therefore, when the data of some devices is not easy to obtain, the sample set can be obtained from other devices with the same type of rotating machinery, and the obtained model is suitable for the same type of rotating machinery and is not limited to specific devices and working conditions.

[0057] Specifically, the feature fusion module in S2 is configured to fuse the vibration features and the acoustic features from the same source domain to obtain same-domain connection features, and fuse the vibration features and the acoustic features from different source domains to obtain cross-domain connection features.

[0058] Correspondingly, the feature mapping module is configured to extract same-domain unified features from the same-domain connection features and extract cross-domain unified features from the cross-domain connection features.

[0059] The classification module in S2 includes a vibration state classifier, an acoustic state classifier, and a combined state classifier. The vibration state classifier is configured to perform fault mode classification according to the vibration features, the acoustic state classifier is configured to perform fault mode classification according to the acoustic features, and the combined state classifier is configured to perform fault mode classification according to the unified features. The output result of the classification module is obtained by applying a majority voting method to the three state classifiers.

[0060] To improve the flexibility and adaptability of the diagnostic model to various tasks and environments, a hybrid integrated diagnostic strategy is proposed in the embodiment, which is equipped with three state classifiers: a vibration state classifier, an acoustic state classifier, and a combined state classifier. First, the invariant unified features output by the mapper are sent to the combined state classifier to capture the relationship between multi-modal monitoring data and machine health status. The vibration state classifier and the acoustic state classifier also mine the potential knowledge in their respective modal data. The result of the hybrid integrated diagnosis is obtained by applying a majority voting method to the three different state classifiers.

[0061] The fault diagnosis model in S2 further includes a modal converter configured to generate false acoustic features according to the vibration features, and generate false vibration features according to the acoustic features.

[0062] Correspondingly, S4 includes: obtaining vibration signals and / or acoustic signals of the rotating machinery, and performing online fault diagnosis based on the vibration signals and / or acoustic signals using the trained fault diagnosis model.

[0063] The embodiment specifically proposes a sound-vibration integrated generalization and conversion method for intelligent fault diagnosis of rotating machinery. Figure 1 As shown in the figure, the fault diagnosis model comprises a vibration feature extractor F v , an acoustic feature extractor F a , a vibration state classifier C v , an acoustic state classifier C a , a combined state classifier C c , a modal converter T av and T va , a feature fusion module, and a feature mapping module specifically an integrated feature mapper P n and P u .

[0064] In the model training phase, the vibration feature extractor F v and the acoustic feature extractor F a respectively take the original vibration signal x v and the original acoustic signal x a as input to perform feature extraction, and the feature extraction is as follows:

[0065] z v =F v (x v );

[0066] z a =F a (x a );

[0067] Wherein, z v is a high-dimensional vibration feature; z a is a high-dimensional acoustic feature.

[0068] Further, information enhancement connection and mutual information maximization are designed to learn invariant unified features cooperatively. The information enhancement connection is to generate same-domain connection features and cross-domain connection features through the feature fusion module. In order to increase the diversity of features and improve the utilization rate of useful information, same-domain connection features from different modal features in the same domain and cross-domain connection features from different modal features in different domains are generated.

[0069] The feature fusion module is configured to splice the vibration features and the acoustic features for fusion. Splicing high-dimensional features from different modalities together is a simple and effective method to overcome the limitations of single-modality information and thus improve the performance of the diagnostic model. In order to increase the diversity of features and enhance the use of useful information, in addition to generating intra-domain connection features from different modalities of features from the same field, cross-domain connection features from different modalities of features from different fields are also generated. Finally, through the synergistic effect of information enhancement connection and mutual information maximization, invariant unified features are obtained, which enhance the resistance of the diagnostic model to data distribution differences.

[0070] Specifically, as shown in Figure 2 , the intra-domain connection features z id after fusion are

[0071]

[0072] The cross-domain connection features z cd are

[0073]

[0074] wherein, is the vibration feature belonging to the m-th class of the i-th source domain; is the acoustic feature belonging to the m-th class of the j-th source domain.

[0075] Further, the integrated feature mapper P n and P u map the intra-domain connection features z id and the cross-domain connection features z cd to unified features as follows:

[0076] s id = P n (z id );

[0077] s cd = P u (z cd );

[0078] wherein, s id is the feature of the intra-domain connection features z id in the unified feature space after mapping by the integrated feature mapper. s cd is the feature of the cross-domain connection features z cd in the unified feature space after mapping by the integrated feature mapper.

[0079] Furthermore, by maximizing the mutual information between co-domain and cross-domain connected features within a unified feature space, fault-related, domain-specific, and modality-specific information can be deciphered.

[0080] That is, S3 includes: training the feature mapping module with the goal of maximizing the mutual information between the same-domain unified features and the cross-domain unified features.

[0081] Specifically, the feature mapping module is trained with the goal of maximizing the mutual information between the intra-domain unified features and the cross-domain unified features, including:

[0082] The same domain unified feature s id and the cross-domain unified feature s cd The mutual information between them is shown in the following formula:

[0083]

[0084] Since directly calculating mutual information is infeasible, maximizing mutual information is usually achieved by increasing its lower bound. Here, I represents mutual information; is the expected value; W and N are the batch sizes of the sample; f is the evaluation function; This is the i-th unified feature within the same domain; This is the j-th cross-domain unified feature.

[0085] Mutual information maximization is achieved by maximizing a lower bound on mutual information. To maximize this lower bound, a supervised contrast loss is used as shown in the following equation. To increase mutual information between features, such as Figure 3 As shown, this is based on supervised contrast loss. The feature mapping module is trained as follows:

[0086]

[0087] Where, k∈K={1,2,…,2W}; A(k)≡K / k; P(k) is the set of positive sample features with the same label as sample k; s=s id ∪s cd ;s i For the i-th feature; s p For the p-th feature; s a τ represents the negative sample features; τ is the scaling parameter.

[0088] The feature mapping module is trained with the goal of maximizing the mutual information between the unified features in the same domain and the unified features across domains. Specifically, this includes training the feature mapping module using supervised contrastive learning and iteratively optimizing the parameters of the feature mapping module based on supervised contrastive loss.

[0089] Further, S3 further comprises:

[0090] According to the classification results of the vibration state classifier, the acoustic state classifier and the combined state classifier, a classification loss is obtained respectively, and then a total classification loss is obtained;

[0091] The vibration state classifier, the acoustic state classifier and the combined state classifier are trained based on the total classification loss.

[0092] Specifically, first, the mapper P n and P u The output unified invariant features are input to the combined state classifier C c to capture the relationship between the multi-modal monitoring data and the machine health state:

[0093]

[0094] Wherein, is the loss function of the combined state classifier; L CE is a cross-entropy loss function; C c (S) is the prediction result of the combined state classifier; is the label; S is s id and s cd .

[0095] The vibration state classifier C v and the acoustic state classifier C a also respectively mine the potential knowledge in their respective modal data:

[0096]

[0097]

[0098] Wherein, is the loss function of the vibration state classifier; C v (F v (X v )) is the prediction result of the vibration state classifier; is the loss function of the acoustic state classifier; C a (F a (X a )) is the prediction result of the acoustic state classifier;

[0099] The total classification loss is defined as:

[0100]

[0101] The results of the hybrid ensemble diagnostics are obtained through majority voting on three different state classifiers, as defined below:

[0102]

[0103] Where H(x) is the final prediction result of sample x; m is the m-th dimension in the label space; The probability of the m-th dimension output by the vibration state classifier based on vibration features; Let m be the probability of the acoustic state classifier output based on acoustic features in the m-th dimension. Let m be the probability of the combined state classifier for the m-th dimension of the unified feature.

[0104] like Figure 4 As shown, this embodiment includes a mode converter T in the model. av and T va This enables mode transitions when certain modes are missing. Real vibration characteristics. False vibration characteristics True acoustic characteristics and false acoustic features The calculation can be performed as follows:

[0105]

[0106] in, It exhibits high-dimensional vibration characteristics; High-dimensional acoustic characteristics; These are false acoustic features, derived from high-dimensional vibration signals converted by a mode converter; These are spurious vibration characteristics, derived from high-dimensional acoustic signals converted by a mode converter.

[0107] S3 also includes: semantic supervision loss as shown in the following formula. The mode converter is trained as follows:

[0108]

[0109] Among them, L CE It is the cross-entropy function; The probability distribution of the class output by the vibration state classifier based on spurious acoustic features; The class probability distribution output by the acoustic state classifier based on spurious vibration features; For tags; and The parameters of the vibration state classifier and the acoustic state classifier are not updated during training. In this way, only the mode conversion translator is penalized during training to establish connections between different modes.

[0110] Further, the embodiment provides an intelligent fault diagnosis system for a rotating machine, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the intelligent fault diagnosis method for the rotating machine.

[0111] The application discloses a sound-vibration integrated generalization and conversion method for intelligent fault diagnosis of a rotating machine, which comprises two signal input modules (a vibration feature extractor and an acoustic feature extractor), a feature fusion module, a feature extraction module, a modal converter and a state recognition module (a vibration state classifier, an acoustic state classifier and a combined state classifier). The two signal input modules extract features from vibration signals and acoustic signals respectively. The feature fusion module and the feature extraction module aggregate the vibration and acoustic features into unified features based on information enhancement connection and mutual information maximization. The modal converter converts the vibration and acoustic features to each other. The state recognition module integrates multiple output results to determine the health state of the equipment. The method effectively utilizes the vibration and acoustic information in the monitoring of the rotating equipment and is expected to improve the equipment diagnosis accuracy.

[0112] The method solves the technical problem of fault diagnosis based on multi-modal field generalization, that is, the vibration signal and the acoustic signal are used simultaneously in the training stage to improve the resistance of the diagnosis model to distribution differences and the robustness of the diagnosis model to sensor faults; and the purpose is to use the multi-source domain vibration and acoustic signals to learn the unified fault representation that is invariant, so as to improve the diagnosis accuracy of the model. Meanwhile, the model training data usually uses vibration and acoustic signals from different operating conditions or other machines of the same type. The fault diagnosis model based on field generalization is trained offline on multi-modal multi-source data (vibration and acoustic signals from different operating conditions or equipment). After the training is completed, the model is applied to online equipment monitoring. The method has low requirements for training data.

[0113] Those skilled in the art will easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for intelligent fault diagnosis of rotating machinery, characterized in that, include: Offline data processing and training phase: S1, Construct a training sample set, which includes sample signals from multiple source domains, and each sample signal includes vibration signals, acoustic signals, and fault labels; S2, The fault diagnosis model is constructed by including: a vibration feature extractor, an acoustic feature extractor, a feature fusion module, a feature mapping module, and a classification module. The vibration feature extractor is used to extract vibration features from the vibration signal. The acoustic feature extractor is used to extract acoustic features from the acoustic signal. The feature fusion module is used to fuse the vibration features and the acoustic features to obtain connection features. The feature mapping module is used to extract unified features from the connection features. The classification module is used to classify fault modes based on the vibration features, the acoustic features, and the unified features. S3, Train the fault diagnosis model based on the training sample set to obtain the trained fault diagnosis model; Online fault diagnosis phase: S4. Use the trained fault diagnosis model to perform online fault diagnosis on rotating machinery; The feature fusion module in S2 is used to fuse the vibration features and acoustic features from the same source domain to obtain co-domain connectivity features, and to fuse the vibration features and acoustic features from different source domains to obtain cross-domain connectivity features. Accordingly, the feature mapping module is used to extract homo-domain unified features from the homo-domain connection features and to extract cross-domain unified features from the cross-domain connection features; S3 includes: training the feature mapping module with the goal of maximizing the mutual information between the same-domain unified features and the cross-domain unified features; The feature mapping module is trained with the goal of maximizing the mutual information between the intra-domain unified features and the cross-domain unified features, specifically including: The same domain unified feature and the aforementioned cross-domain unified features The mutual information between them is shown in the following formula: ； in, Represents mutual information; For expectations; and This represents the batch size of the sample. For evaluation functions; For the first i Uniform characteristics within the same domain; For the first j A unified feature across domains; Mutual information is maximized by maximizing a lower bound on mutual information, which is achieved using the supervised contrast loss as shown in the following formula. The feature mapping module is trained as follows: ； in, ; ; To be consistent with the sample A set of positive sample features with the same label; ; For the first i One feature; For the first p One feature; Features of negative samples; These are scaling parameters.

2. The intelligent fault diagnosis method for rotating machinery as described in claim 1, characterized in that, The same domain connection feature Specifically: ； The cross-domain connection feature Specifically: ； in, For belonging to the first The source domain number Vibrational characteristics of the class; For belonging to the first j The source domain number Acoustic characteristics of the class.

3. The intelligent fault diagnosis method for rotating machinery as described in claim 1 or 2, characterized in that, The classification module in S2 includes a vibration state classifier, an acoustic state classifier, and a combined state classifier. The vibration state classifier is used to classify fault modes based on the vibration characteristics. The acoustic state classifier is used to classify fault modes based on the acoustic characteristics. The combined state classifier is used to classify fault modules based on the unified characteristics. The output of the classification module is obtained by applying a majority voting method to the three state classifiers.

4. The intelligent fault diagnosis method for rotating machinery as described in claim 3, characterized in that, S3 also includes: The classification loss is obtained from the classification results of the vibration state classifier, the acoustic state classifier and the combined state classifier, and then the total classification loss is obtained. The vibration state classifier, the acoustic state classifier, and the combined state classifier are trained based on the total classification loss.

5. The intelligent fault diagnosis method for rotating machinery as described in claim 3, characterized in that, The fault diagnosis model in S2 further includes a mode converter, which is used to generate false acoustic features based on the vibration features and to generate false vibration features based on the acoustic features. Accordingly, S4 includes: acquiring vibration signals and / or acoustic signals of rotating machinery, and performing online fault diagnosis based on the vibration signals and / or acoustic signals using a trained fault diagnosis model.

6. The intelligent fault diagnosis method for rotating machinery as described in claim 5, characterized in that, S3 also includes: semantic supervision loss as shown in the following formula. The mode converter is trained as follows: ； in, It is the cross-entropy function; The probability distribution of the class output by the vibration state classifier based on spurious acoustic features; The class probability distribution output by the acoustic state classifier based on spurious vibration features; For tags; and This indicates that the parameters of the vibration state classifier and the acoustic state classifier were not updated during training.

7. An intelligent fault diagnosis system for rotating machinery, characterized in that, The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it performs the intelligent fault diagnosis method for rotating machinery as described in any one of claims 1-6.

Images