Fault Diagnosis Method, System, Electronic Device and Medium Based on Digital Twin

Abstract

The present application discloses a fault diagnosis method, system, electronic device and medium based on digital twin. The method includes: obtaining real-domain data and simulation-domain data, and dividing the real-domain data into a training set and a test set; establishing a fault diagnosis model; respectively inputting the training set and the simulation-domain data into a first feature extractor and a second feature extractor in the fault diagnosis model for feature extraction, and outputting real-domain features and simulation-domain features; training the fault diagnosis model according to the real-domain features and the simulation-domain features, and testing the trained fault diagnosis model through the test set to obtain a target model; inputting a to-be-diagnosed fault set containing fault data into the target model for fault identification, and outputting a target fault type. In the embodiments of the present application, the problem that the recognition performance of the fault diagnosis model is low due to data imbalance can be improved, the recognition of single faults can be realized, and at the same time, the accurate decoupling of compound faults can be realized, and the accuracy of fault recognition can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular, to a fault diagnosis method, system, electronic device, and storage medium based on digital twin. Background Art

[0002] Rotating machinery is the most important part of mechanical equipment. According to statistics, 80% of mechanical equipment is rotating machinery, and its health status plays a crucial role in the safe and reliable operation of the entire mechanical system. Rolling bearings are one of the most widely used parts in rotating machinery, and key components of rolling bearings induce 30% of faults. The phenomenon of multiple faults occurring simultaneously in rolling bearings is relatively common. A single fault inside the bearing gradually evolves into multiple faults, becoming a compound fault. The compound fault signal contains multiple fault characteristics, which may exist in forms such as non-independence and coupling. The number of faults and the types of faults included cannot be judged before fault diagnosis. Compared with single faults, compound faults are more common in actual engineering, and at the same time, the diagnostic difficulty and the harm to mechanical equipment are greater. Therefore, the effective detection and fault diagnosis of compound faults are becoming more and more important.

[0003] However, traditional compound fault diagnosis methods regard compound faults as an independent fault mode during the diagnosis process. In fact, compound faults are not completely a new type of fault because the compound fault information is composed of the corresponding characteristics of single faults. If compound faults are only regarded as an independent fault mode during the diagnosis process, the connection between compound faults and single faults will be ignored, resulting in problems such as missed diagnosis in the diagnostic application process. In the prior art, deep learning is usually used to decouple compound faults. However, deep learning depends on a large amount of labeled data to train the model. It is very difficult to collect data on compound faults, and there are extremely few labeled compound fault diagnosis signals. It is unrealistic to collect complete labeled fault data, which brings the problem of data imbalance and further affects the accuracy of decoupling and classifying compound faults of rolling bearings. Summary of the Invention

[0004] The embodiments of the present application provide a fault diagnosis method, system, electronic device, and storage medium based on digital twin, which can improve the problem of low recognition performance of the fault diagnosis model caused by data imbalance and improve the accuracy of fault recognition and classification.

[0005] In a first aspect, the embodiments of the present application provide a fault diagnosis method based on digital twin, and the method includes:

[0006] Obtain real-domain data and simulation-domain data, and divide the real-domain data into a training set and a test set, where the real-domain data is used to represent single-fault data and compound-fault data of a rolling bearing, and the simulation-domain data is obtained by performing simulated data sampling on the single-fault data and the compound-fault data;

[0007] Build a fault diagnosis model, where the fault diagnosis model includes a first feature extractor and a second feature extractor;

[0008] Input the training set into the first feature extractor for feature extraction, output real-domain features, and input the simulation-domain data into the second feature extractor for feature extraction, output simulation-domain features;

[0009] Iteratively train the fault diagnosis model based on the real-domain features and the simulation-domain features, and perform model testing on the iteratively trained fault diagnosis model through the test set to obtain a target model;

[0010] Obtain a to-be-diagnosed fault set containing fault data, and input the to-be-diagnosed fault set into the target model for fault identification, output the target fault type.

[0011] The fault diagnosis method based on digital twin provided by the embodiments of the present application has at least the following beneficial effects: First, real-domain data including single-fault data and compound-fault data of rolling bearings, as well as simulation-domain data obtained by simulating data sampling of the single-fault data and compound-fault data, are obtained, so that real-domain data and simulation-domain data containing rich fault feature information are obtained, providing sufficient source-domain data for model training, effectively improving the problem of low recognition performance of the fault diagnosis model caused by data imbalance, and dividing the real-domain data into a training set and a test set. Then, a fault diagnosis model including a first feature extractor, a second feature extractor, and a multi-label decoupling classifier is established to enhance the robustness of the fault diagnosis model. After that, the training set is input into the first feature extractor for feature extraction to output real-domain features, and the simulation-domain data is input into the second feature extractor for feature extraction to output simulation-domain features, so that automatic feature learning and extraction can be performed on the training set and the simulation-domain data respectively, improving the accuracy of feature extraction. Then, the real-domain features and simulation-domain features are input into the fault diagnosis model for iterative training to extract various single-fault features contained in the compound-fault signal and classify them by using the relationship between the compound fault and the single fault, realizing accurate prediction of the fault, and the iterative training of the fault diagnosis model with the real-domain features and simulation-domain features can improve the prediction accuracy of the fault diagnosis model. And the iterative-trained fault diagnosis model is tested by the test set to obtain a target model, a to-be-diagnosed fault set containing fault data is obtained, and the to-be-diagnosed fault set is input into the target model for fault identification to output the target fault type, improving the diagnosis accuracy of the compound fault. The embodiments of the present application obtain simulation-domain data by simulating data sampling of single-fault data and compound-fault data, so that data containing rich fault feature information can be obtained, and the fault diagnosis model is trained with the real-domain data and simulation-domain data, improving the defect that the training model seriously depends on a large amount of labeled data, avoiding the problem of low recognition performance of the fault diagnosis model caused by data imbalance, realizing the identification of single faults, and at the same time realizing the accurate decoupling of compound faults, and further improving the diagnosis accuracy of fault categories.

[0012] In some embodiments, the simulation-domain data is obtained by the following steps:

[0013] Establish a rolling bearing digital twin model according to the actual data of the rolling bearing;

[0014] Simulate the operating conditions of the rolling bearing through the rolling bearing digital twin model to obtain vibration signal data;

[0015] Classify the vibration signal data to obtain simulation-domain data.

[0016] In some embodiments, establishing the fault diagnosis model includes:

[0017] Build an intelligent diagnosis model;

[0018] Build a first feature extractor and a second feature extractor based on the multi-head self-attention mechanism;

[0019] Conduct transfer learning design and edge-aware regularization design on the intelligent diagnosis model;

[0020] Construct a multi-label decoupled classifier based on the intelligent diagnosis model;

[0021] Build a fault diagnosis model according to the first feature extractor, the second feature extractor and the multi-label decoupled classifier.

[0022] In some embodiments, the conducting transfer learning design and edge-aware regularization design on the intelligent diagnosis model includes:

[0023] Align the features of the real domain data and the simulation domain data based on the subdomain adaptation mechanism of local maximum mean discrepancy measurement to align the conditional distributions of the real domain data and the simulation domain data;

[0024] Obtain the real sample labels of the rolling bearing, where the real sample labels are used to characterize the real fault data of the rolling bearing;

[0025] Design edge-aware regularization based on the real sample labels.

[0026] In some embodiments, the inputting the training set into the first feature extractor for feature extraction and outputting real domain features includes:

[0027] Input the training set into the first feature extractor to enable the first feature extractor to divide the one-dimensional vibration signals in the training set and output a real patch sequence;

[0028] Map the real patch sequence to a latent vector through the linear transformation layer of the first feature extractor to generate a real block embedding;

[0029] Retain the position information of the real block embedding and output real domain features;

[0030] The inputting the simulation domain data into the second feature extractor for feature extraction and outputting simulation domain features includes:

[0031] Input the simulation domain into the second feature extractor to enable the second feature extractor to divide the one-dimensional vibration signals in the simulation domain and output a simulation patch sequence;

[0032] Map the simulation patch sequence to a latent vector through the linear transformation layer of the second feature extractor to generate a simulation block embedding;

[0033] Retain the position information of the simulation block embedding and output the simulation domain features.

[0034] In some embodiments, the fault diagnosis model includes a multi-label decoupled classifier; iteratively training the fault diagnosis model based on the real domain features and the simulation domain features includes:

[0035] Perform transfer learning on the real domain features and the simulation domain features based on the subdomain adaptation mechanism to obtain the local maximum mean discrepancy metric;

[0036] Input the real domain features and the simulation domain features into the multi-label decoupled classifier for fault diagnosis and output the predicted labels;

[0037] Calculate the local maximum mean discrepancy metric and the predicted labels based on edge-aware regularization to iteratively train the fault diagnosis model.

[0038] In some embodiments, testing the iteratively trained fault diagnosis model through the test set to obtain the target model includes:

[0039] Evaluate the performance of the fault diagnosis model through the test set to obtain the evaluation metrics;

[0040] When the evaluation metrics meet the preset metric conditions, use the trained fault diagnosis model as the target model.

[0041] In a second aspect, an embodiment of the present application further provides a fault diagnosis system based on digital twin, the system includes:

[0042] A data acquisition module, configured to acquire real domain data and simulation domain data, and divide the real domain data into a training set and a test set, where the real domain data is used to represent single fault data and composite fault data of a rolling bearing, and the simulation domain data is obtained by simulating data sampling of the single fault data and the composite fault data;

[0043] A model establishment module, configured to establish a fault diagnosis model, the fault diagnosis model includes a first feature extractor and a second feature extractor;

[0044] A feature extraction module, configured to input the training set into the first feature extractor for feature extraction, output real domain features, and input the simulation domain data into the second feature extractor for feature extraction, output simulation domain features;

[0045] A model training module, configured to iteratively train the fault diagnosis model based on the real domain features and the simulation domain features, and perform model testing on the iteratively trained fault diagnosis model through the test set to obtain a target model;

[0046] A fault identification module, configured to obtain a fault set to be diagnosed including fault data, and input the fault set to be diagnosed into the target model for fault identification, and output a target fault type.

[0047] In a third aspect, an embodiment of the present application further provides an electronic device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the method for fault diagnosis based on digital twin as described in the first aspect is implemented.

[0048] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, storing computer-executable instructions, where the computer-executable instructions are used to cause a computer to execute the method for fault diagnosis based on digital twin as described in the first aspect.

[0049] Other features and advantages of the present application will be described in the subsequent description, and part of them will become obvious from the description, or be understood by implementing the present application. The purpose and other advantages of the present application can be realized and obtained through the structures specifically pointed out in the description and the drawings. Description of the Drawings

[0050] The drawings are used to provide a further understanding of the technical solutions of the present application, and constitute a part of the description. Together with the examples of the present application, they are used to explain the technical solutions of the present application, and do not constitute a limitation to the technical solutions of the present application.

[0051] Figure 1 It is a flowchart of the specific method of the fault diagnosis method based on digital twin provided by the embodiment of the present application;

[0052] Figure 2 It is a specific flowchart of obtaining simulation domain data provided by the embodiment of the present application;

[0053] Figure 3 It is a specific flowchart of step S102 provided by the embodiment of the present application;

[0054] Figure 4 It is a specific flowchart of step S303 provided by the embodiment of the present application;

[0055] Figure 5 It is a specific flowchart of inputting a training set into a first feature extractor for feature extraction provided by the embodiment of the present application;

[0056] Figure 6It is a specific flowchart for inputting simulation domain data into a second feature extractor for feature extraction provided by an embodiment of the present application;

[0057] Figure 7 It is a schematic diagram of a feature extractor provided by an embodiment of the present application;

[0058] Figure 8 It is a specific flowchart for inputting real domain features and simulation domain features into a fault diagnosis model for iterative training provided by an embodiment of the present application;

[0059] Figure 9 It is a schematic diagram of a multi-label decoupled classifier provided by an embodiment of the present application;

[0060] Figure 10 It is a schematic diagram of a sub-domain adaptation mechanism provided by an embodiment of the present application;

[0061] Figure 11 It is a specific flowchart for inputting a test set into an iteratively trained fault diagnosis model for model testing provided by an embodiment of the present application;

[0062] Figure 12 It is a model schematic diagram of a fault diagnosis model provided by an embodiment of the present application;

[0063] Figure 13 It is a system schematic diagram of a fault diagnosis system based on digital twin provided by an embodiment of the present application;

[0064] Figure 14 It is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0065] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0066] It should be noted that although functional module division is performed in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order from the module division in the device or the flowchart in the flowchart. Terms such as "first" and "second" in the specification, claims and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0068] The fault diagnosis method based on digital twin provided by the embodiments of the present application can be applied to a terminal, or to a server side, or can also be software running on the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart watch, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms; the software can be an application implementing the above method, etc., but is not limited to the above forms.

[0069] The embodiments of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet-type devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer computer devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0070] Rotating machinery is the most important part of mechanical equipment. According to statistics, 80% of mechanical equipment is rotating machinery, and its health status plays a crucial role in the safe and reliable operation of the entire mechanical system. Rolling bearings are one of the most widely used parts in rotating machinery, and 30% of the faults in rolling bearings are induced by key components. The phenomenon of multiple faults occurring simultaneously in rolling bearings is relatively common, and a single fault in the bearing gradually evolves into multiple faults, becoming a compound fault. The compound fault signal contains multiple fault characteristics, which may exist in forms such as non-independence and coupling, and the number of faults and the types of faults included cannot be judged before fault diagnosis. Compared with single faults, compound faults are more common in actual engineering, and at the same time, the diagnostic difficulty and the harm to mechanical equipment are greater. Therefore, the effective detection and fault diagnosis of compound faults are becoming more and more important.

[0071] However, traditional compound fault diagnosis methods regard compound faults as an independent fault mode during the diagnosis process. In fact, compound faults are not entirely a new type of fault because the compound fault information is composed of the corresponding characteristics of single faults. If compound faults are only regarded as an independent fault mode during the diagnosis process, the connection between compound faults and single faults will be ignored, resulting in problems such as missed diagnosis in the process of diagnostic applications.

[0072] Although current deep learning models are constantly developing, the research on intelligent decoupling methods for deep learning is relatively scarce. Most intelligent compound fault diagnosis models regard compound faults as an independent new type of fault and cannot achieve the decoupling effect. Moreover, deep learning relies on a large amount of labeled data to train the model. It is difficult to collect data on compound faults, and the labeled compound fault diagnosis signals are extremely rare. It is unrealistic to collect complete labeled fault data.

[0073] To solve the above problems, the embodiments of the present application provide a fault diagnosis method, system, electronic device, and storage medium based on digital twin. First, real-domain data including single-fault data and compound-fault data of rolling bearings, as well as simulation-domain data obtained by simulating data sampling for the single-fault data and compound-fault data, are acquired, thereby obtaining real-domain data and simulation-domain data containing rich fault feature information, providing sufficient source-domain data for model training, effectively improving the problem of low recognition performance of the fault diagnosis model caused by data imbalance, and dividing the real-domain data into a training set and a test set. Then, a fault diagnosis model including a first feature extractor, a second feature extractor, and a multi-label decoupling classifier is established to enhance the robustness of the fault diagnosis model. After that, the training set is input into the first feature extractor for feature extraction to output real-domain features, and the simulation-domain data is input into the second feature extractor for feature extraction to output simulation-domain features, so as to be able to automatically learn and extract features from the training set and the simulation-domain data respectively, improving the accuracy of feature extraction. Then, the real-domain features and the simulation-domain features are input into the fault diagnosis model for iterative training to extract various single-fault features contained in the compound-fault signal and classify them by using the relationship between the compound fault and the single fault, realizing accurate prediction of the fault, and the iterative training of the fault diagnosis model by the real-domain features and the simulation-domain features can improve the prediction accuracy of the fault diagnosis model. And the iterative-trained fault diagnosis model is tested by the test set to obtain a target model, a to-be-diagnosed fault set containing fault data is acquired, and the to-be-diagnosed fault set is input into the target model for fault identification to output the target fault type, improving the diagnosis accuracy of the compound fault. The embodiments of the present application can obtain data containing rich fault feature information by simulating data sampling for the single-fault data and compound-fault data, and train the fault diagnosis model by the real-domain data and the simulation-domain data, improving the defect that the training model seriously depends on a large amount of labeled data, avoiding the problem of low recognition performance of the fault diagnosis model caused by data imbalance, realizing the identification of single faults, and at the same time realizing the accurate decoupling of compound faults, and further being able to improve the diagnosis accuracy of fault categories.

[0074] The following further elaborates on the embodiments of the present invention in conjunction with the accompanying drawings.

[0075] Refer to Figure 1 , Figure 1 which is a flowchart of the specific method of the fault diagnosis method based on digital twin provided by the embodiments of the present application. In some embodiments, the fault diagnosis method based on digital twin includes but is not limited to steps S101 to S105.

[0076] Step S101, acquire real-domain data and simulation-domain data, and divide the real-domain data into a training set and a test set.

[0077] It should be noted that the real - domain data is used to characterize the single - fault data and compound - fault data of the rolling bearing, and the simulation - domain data is obtained by performing simulated data sampling on the single - fault data and compound - fault data.

[0078] In some embodiments, after a defect occurs at a certain position in the rolling bearing, it causes a system chain reaction. As the operation continues and the operating environment changes, the single fault in the bearing gradually evolves into multiple faults, becoming a compound fault. To improve the accuracy of judging the fault type of the rolling bearing, the embodiments of the present application first obtain the real - domain data including single - fault data and compound - fault data, and obtain the simulation - domain data obtained by performing simulated data sampling on the single - fault data and compound - fault data, so as to obtain fault data containing rich fault - feature information, provide sufficient source - domain data for model training, and divide the real - domain data into a training set and a test set, facilitating subsequent training of the model and improving the accuracy of the model for fault classification.

[0079] It should be noted that the compound - fault data in the embodiments of the present application includes, but is not limited to, combinations such as inner - race fault, outer - race fault, rolling - element fault, etc.

[0080] It is worth noting that the characteristics of the compound - fault signal are not a simple linear superposition of multiple single faults, but a complex coupling result of multiple single - fault characteristics. If, during the diagnosis process, the compound fault is merely regarded as an independent fault mode, the connection between the compound fault and the single fault will be ignored, resulting in the inability to reveal its essential characteristics. This is not a true decoupling and identification of the compound fault, and problems such as missed diagnosis may even occur during its diagnostic application process.

[0081] In some embodiments, the real - domain data in the embodiments of the present application can use the rolling - bearing dataset in the open - source database. Among them, the real - domain data contains data of various working conditions such as single faults and compound faults of the rolling bearing, that is, the real - domain data includes vibration - signal data of inner - race faults, outer - race faults, rolling - element faults, and compound faults composed of two or more of them. The embodiments of the present application select a small amount of vibration signals of single faults and compound faults as the acceleration vibration signals of the real - domain data. In addition, the simulation - domain data in the embodiments of the present application is only used for subsequent training of the fault - diagnosis model.

[0082] It should be noted that the real-domain data in the embodiments of this application includes a large amount of normal data and a small amount of faulty data (single-fault data and compound-fault data). Among them, the normal data is the data when the rolling bearing is working normally. At this time, based on the above real-domain data, a data imbalance scenario is simulated, and a digital twin model of the rolling bearing is established through digital twin technology, so as to obtain a sufficient amount of simulation-domain data through simulation, achieve the purpose of balancing training data, and further solve the problem of data imbalance in the process of training the model.

[0083] Step S102, establish a fault diagnosis model, which includes a first feature extractor and a second feature extractor.

[0084] In some embodiments, traditional compound fault diagnosis methods regard compound faults as an independent fault mode during the diagnosis process. In fact, compound faults are not completely a new type of fault. The characteristics of compound fault signals are not a simple linear superposition of multiple single faults, but a complex coupling result of multiple single fault characteristics. Just regarding compound faults as an independent fault mode will ignore the connection between compound faults and single faults, resulting in the inability to reveal their essential characteristics. This is not a real decoupling and identification of compound faults, and problems such as missed diagnosis may even occur in the process of its diagnostic application. To solve the above problems, the embodiments of this application will establish a fault diagnosis model, so that the fault diagnosis model can utilize the connection between compound faults and single faults, extract various single fault characteristics contained in the compound fault signal and classify them, thereby improving the diagnostic accuracy of compound faults.

[0085] It should be noted that the models and shared parameters of the first feature extractor and the second feature extractor in the embodiments of this application are the same.

[0086] Step S103, input the training set into the first feature extractor for feature extraction, output real-domain features, and input the simulation-domain data into the second feature extractor for feature extraction, output simulation-domain features.

[0087] In some embodiments, input the training set into the first feature extractor for feature extraction, output real-domain features, to achieve the extraction of features in the training set, and input the simulation-domain data into the second feature extractor for feature extraction, output simulation-domain features, to achieve the extraction of features in the simulation domain. By separately extracting features from real-domain data and simulation-domain data, the accuracy of feature extraction can be improved.

[0088] Step S104, based on the real-domain features and simulation-domain features, input them into the fault diagnosis model for iterative training, and use the test set to test the fault diagnosis model after iterative training to obtain the target model.

[0089] In some embodiments, the fault diagnosis model is iteratively trained by real-domain features and simulation-domain features to improve the diagnostic accuracy of the fault diagnosis model for faults. By optimizing the network learning parameters, the model with the lowest training set loss value can be selected as the well-trained model during the iteration process, and the iteratively trained fault diagnosis model is tested by the test set to obtain the target model, ensuring the quality, reliability, and practicability of the model, and improving the accuracy of decoupling faults by the fault diagnosis model.

[0090] It should be noted that the embodiments of the present application construct a fault diagnosis model based on unbalanced transfer learning, which only uses a small amount of real data, and improves the difference between the source domain and the target domain based on sufficient simulation-domain data and transfer learning, effectively solving the problem that the model performance is limited due to data imbalance.

[0091] Step S105, obtain the to-be-diagnosed fault set containing fault data, and input the to-be-diagnosed fault set into the target model for fault identification, and output the target fault type.

[0092] In some embodiments, obtain the to-be-diagnosed fault set containing fault data, and input the to-be-diagnosed fault set into the target model for fault identification, and output the target fault type, realizing the rapid diagnosis of the fault types in the diagnosed fault set, improving the fault response speed, reducing human errors, and overcoming the defect that traditional fault diagnosis can only output one label. At the same time, deep learning is used to automatically extract features, saving manpower and material resources, reducing errors in the intermediate links, and improving the diagnosis accuracy and efficiency.

[0093] Refer to Figure 2 , Figure 2 is the specific flowchart for obtaining simulation-domain data provided by the embodiments of the present application. In some embodiments, the method includes but is not limited to step S201 and step S203.

[0094] Step S201, establish a digital twin model of the rolling bearing according to the actual data of the rolling bearing.

[0095] Step S202, simulate the operating conditions of the rolling bearing through the digital twin model of the rolling bearing to obtain vibration signal data.

[0096] Step S203, classify the vibration signal data to obtain simulation-domain data.

[0097] In steps S201 to S203 of some embodiments, in the process of obtaining simulation domain data, first, a high-fidelity digital twin model of the rolling bearing is established based on the actual data of the rolling bearing, that is, a high-fidelity digital twin model of the rolling bearing is established according to the physical entity of the rolling bearing. Then, the operating conditions of the rolling bearing are simulated through the digital twin model of the rolling bearing. At this time, single faults and compound faults of the rolling bearing are included, for example, inner race fault, outer race fault, rolling element fault, etc. Based on the vibration response of the digital twin model of the rolling bearing, one-dimensional vibration signal data is generated to obtain the vibration signal data. After that, the vibration signal data is sorted and classified to obtain the simulation domain data, so that the simulation domain data includes various fault types such as single faults and compound faults. The embodiments of the present application can combine digital twin into compound fault decoupling, establish its dynamic simulation model through digital twin technology, obtain rich fault feature information, and provide sufficient source domain data for model training.

[0098] Specifically, in the process of establishing the digital twin model of the rolling bearing and obtaining the simulation domain data, first, the physical entity of the rolling bearing and its related characteristics are obtained, for example, geometric characteristics, mechanical properties, material characteristics, etc. Then, finite element analysis is performed on the physical entity and related characteristics to establish a dynamic model. By analyzing the dynamic model under the motion state, the force condition under the motion state is obtained. Finally, the defect condition of the rolling bearing is simulated, and its fault model is constructed to obtain the simulation domain data.

[0099] Refer to Figure 3 , Figure 3 is the specific flowchart of step S102 provided by the embodiments of the present application. In some embodiments, step S102 specifically includes but is not limited to step S301 and step S305.

[0100] Step S301, establish an intelligent diagnosis model.

[0101] Step S302, establish a first feature extractor and a second feature extractor based on the multi-head self-attention mechanism.

[0102] In some embodiments, in the process of establishing the fault diagnosis model, the embodiments of the present application need to design a shared feature extractor, that is, establish a first feature extractor and a second feature extractor based on the multi-head self-attention mechanism.

[0103] Specifically, in the process of establishing the first feature extractor and the second feature extractor in the embodiments of the present application, a feature extractor based on Transformer is designed. The Transformer-based network is built into the feature extractor. The Transformer module consists of a multi-head self-attention (MSA) and a multi-layer perceptron (MLP) module. Layer normalization (Layer Norm) is applied before each block, and a residual connection is applied after each block. The main module of the Transformer module is the MSA, and the MSA has the ability to capture long-range dependencies between sequences. The multi-head self-attention mechanism consists of multiple heads of the self-attention function, and the formula calculation process of the attention is as follows.

[0104] Assume a sequence of input embedding vectors x, where each element has a dimension of d model , and each element has its own query q, key k, and value v. The dimensions of the query and the key are d k , and the dimension of the value is d v , which is transformed into d model , and the query matrix Q j (queries), key matrix K j (keys), and value matrix V j (values) are expressed as the following formulas:

[0105] Q j = xW j q ;

[0106] K j = xW j k ;

[0107] V j = xW j v ;

[0108] Among them, and are both learnable linear mapping matrices. The single-head scaled dot-product attention is calculated as:

[0109]

[0110] Among them, is the scaling factor.

[0111] Then, multi-head attention is introduced to jointly focus on information from different representation subspaces at different positions. It performs single-head attention H times, and the multi-head attention is defined as:

[0112]

[0113] head j = Attention(Q, K, V) j , j = 1, 2, ..., H;

[0114] Among them, Concat represents concatenation, is a learnable linear mapping matrix, and H is the number of attention heads.

[0115] In some embodiments, Transformer relies on the internal multi-head attention mechanism, which can accurately capture long-range correlation knowledge. The multi-head attention jointly focuses on information from different representation subspaces at different positions. In addition to multi-head self-attention, Transformer also contains an MLP module. The MLP module includes two fully connected layers and a GELU non-linear activation function, and can be defined by the following formula:

[0116] MLP(x) = GELU(0, xW 1 + b 1 )W 2 + b 2 ;

[0117] Among them, W 1 and W 2 are weight parameters, and b 1 and b 2 are bias terms.

[0118] Step S303, perform transfer learning design and edge-aware regularization design on the intelligent diagnosis model.

[0119] In some embodiments, perform transfer learning design on the intelligent diagnosis model. By minimizing the difference in fine-grained features between the simulated domain fault data and the real domain fault data, align the conditional distributions of the fault data in the two domains, and design edge-aware regularization to impose a significant regularization penalty on the fault data margin, facilitating subsequent enhancement of the fault diagnosis robustness of the model.

[0120] Step S304, construct a multi-label decoupled classifier based on the intelligent diagnosis model.

[0121] In some embodiments, the embodiment of the present application designs a Sigmoid classifier in the last layer of the intelligent diagnosis model. Specifically, select the Sigmoid function to construct a multi-label classifier, ensuring that the probability values of each output label are independent of each other.

[0122] It should be noted that the embodiments of the present application can also set a threshold in the intelligent diagnosis model. When the predicted probability of a certain health state is greater than the threshold, the corresponding label is output. For example, the threshold is set to the average value of the three single faults in the compound fault, the threshold is set to the average value of the two single faults in the compound fault, etc. The embodiments of the present application do not make specific limitations.

[0123] It can be understood that, assuming a test sample x is given, after the feedforward operation of the network model, the output value of the last fully connected layer is y i , and the probability prediction value of the sample belonging to a certain category is calculated through the Sigmoid function. Among them, the expression of the Sigmoid function is represented by the following formula:

[0124]

[0125] Step S305, establish a fault diagnosis model according to the first feature extractor, the second feature extractor and the multi-label decoupling classifier.

[0126] In some embodiments, a fault diagnosis model is established according to the first feature extractor, the second feature extractor and the multi-label decoupling classifier, so that the fault diagnosis model includes a shared feature extractor, a sub-domain adaptive mechanism based on local maximum mean discrepancy measurement, edge-aware regularization and a decoupled multi-label classifier, which is convenient for subsequent decoupling of compound faults and improves the accuracy of the model in predicting faults.

[0127] Based on the above description of the fault diagnosis model, the feature extraction process and the fault decoupling process in the embodiments of the present application are specifically described.

[0128] Refer to Figure 4 , Figure 4 is the specific flowchart of step S303 provided by the embodiments of the present application. In some embodiments, step S303 specifically includes but is not limited to step S401 and step S403.

[0129] Step S401, perform feature alignment on the real domain data and the simulation domain data through a sub-domain adaptive mechanism based on local maximum mean discrepancy measurement to align the conditional distributions of the real domain data and the simulation domain data.

[0130] Step S402, obtain the real sample labels of the rolling bearing.

[0131] It should be noted that the real sample labels are used to characterize the real fault data of the rolling bearing.

[0132] Step S403, design edge-aware regularization based on the real sample labels.

[0133] In steps S401 to S403 of some embodiments, in the process of performing transfer learning design and edge-aware regularization design on the intelligent diagnosis model, first, a subdomain adaptive mechanism (SAM) based on the local maximum mean discrepancy (MMD) metric is used to perform feature alignment on the real domain data and the simulation domain data. Specifically, the embodiment of the present application aligns the conditional distribution of the real domain data and the simulation domain data by minimizing the difference in fine-grained features between the simulation domain and the real domain fault data, thereby reducing the difference in distribution space between the simulation domain data and the real domain data. Then, edge-aware regularization is designed. Specifically, the real sample labels of the rolling bearings are obtained, and then edge-aware regularization (Modality-Aware Regu l ar izat ion, MAR) is designed based on the real sample label distribution. MAR is essentially a cross-entropy loss function with an optimized decision boundary, which imposes a significant regularization penalty on the fault data margin to enhance the fault diagnosis robustness of the model.

[0134] It is worth noting that the scope of transfer learning is introduced in the embodiment of the present application, and a subdomain adaptive mechanism based on the local maximum mean difference metric is adopted to align the conditional distribution of fault data in the two domains and reduce the difference in distribution space between the simulation domain and the real domain. In addition, edge-aware regularization is designed using the label distribution of the real world, and a significant regularization penalty is imposed on the margin of fault data, reducing the difference in classifier output and enhancing the fault diagnosis robustness of the model. The above combination can use the empirical knowledge of the source domain to improve the feature learning of the target domain, making it suitable for fault diagnosis under various working conditions or various equipment. In addition, transfer learning is not limited by the difference in distribution between the training set and the test set. It can mine common features in the same or different fields, improve the diversity of source domain data, speed up the training process and improve the classification accuracy to a certain extent, while reducing tedious steps such as labeling the test set.

[0135] Reference Figure 5 , Figure 5 It is a specific flow chart of inputting a training set into a first feature extractor for feature extraction provided by an embodiment of the present application. The method includes but is not limited to step S501 and step S503.

[0136] Step S501: input the training set into the first feature extractor, so that the first feature extractor divides the one-dimensional vibration signal in the training set and outputs a true patch sequence.

[0137] Step S502 , mapping the real patch sequence to a latent vector through a linear transformation layer of the first feature extractor to generate a real block embedding.

[0138] Step S503: Preserve the position information of the real block embedding and output the real domain features.

[0139] In steps S501 to S503 of some embodiments, during the process of extracting the real domain features of the training set, the training set is input into the first feature extractor, so that the first feature extractor divides the one-dimensional vibration signals in the training set, splits the long sequence into shorter segments to meet the input requirements of the Transformer model, and outputs the real patch sequence. Then, the real patch sequence is mapped into the latent vector through the linear transformation layer of the first feature extractor, thereby generating the real block embedding, and the size of the generated real block embedding is the same as that of the original patch. Finally, since the Transformer itself does not have the ability to capture the order of elements in the sequence, in this application embodiment, the position information of the real block embedding is preserved. Specifically, a randomly trainable embedding class token is introduced into the real block embedding, so as to be able to preserve the position information in the Transformer block, allowing the model to distinguish different positions in the sequence, and thus being able to capture the order dependencies in the sequence, outputting the real domain features, realizing the extraction of features in the training set, and improving the accuracy of feature extraction.

[0140] It can be understood that since the Transformer is usually designed to process sequences of fixed length, segmentation can enable the model to process long sequences more effectively and improve the computational efficiency. At the same time, shorter sequences may be easier to capture local dependencies.

[0141] Refer to Figure 6 , Figure 6 is the specific flowchart of inputting the simulation domain data into the second feature extractor for feature extraction provided by the embodiments of this application. The method includes but is not limited to steps S601 and S603.

[0142] Step S601: Input the simulation domain into the second feature extractor, so that the second feature extractor divides the one-dimensional vibration signals in the simulation domain and outputs the simulation patch sequence.

[0143] Step S602: Map the simulation patch sequence into the latent vector through the linear transformation layer of the second feature extractor to generate the simulation block embedding.

[0144] Step S603: Preserve the position information of the simulation block embedding and output the simulation domain features.

[0145] In steps S601 to S603 of some embodiments, during the process of extracting the simulation domain features of the simulation domain, the simulation domain is input into the second feature extractor, so that the second feature extractor divides the one-dimensional vibration signal in the simulation domain, splits the long sequence into shorter segments to meet the input requirements of the Transformer model, and outputs a simulation patch sequence. Then, the simulation patch sequence is mapped to a latent vector through the linear transformation layer of the second feature extractor to generate a simulation block embedding, and the size of the generated simulation block embedding is the same as that of the original patch. Finally, position information is retained for the simulation block embedding. For the specific process, reference can be made to the relevant description in step S503, which will not be elaborated herein in the embodiments of the present application. The simulation domain features are output to achieve the extraction of features in the simulation domain and improve the accuracy of feature extraction.

[0146] Referring to Figure 7 , Figure 7 is a schematic diagram of the feature extractor provided by the embodiments of the present application.

[0147] It can be understood that the embodiments of the present application design two feature extractors (the first feature extractor and the second feature extractor) with the same model and shared parameters as the shared feature extractor. During training, a small amount of real composite fault one-dimensional vibration signals (one-dimensional vibration signals in the training set) collected and the one-dimensional vibration signals generated by the digital twin model (one-dimensional vibration signals in the simulation domain) are respectively input into the feature extractor for automatic feature learning and extraction.

[0148] Specifically, assuming that the input sample x from a signal sequence of each length L is divided into C segments of length S to form a patch sequence, the following formula can be referred to:

[0149]

[0150] where C is the number of patches.

[0151] Then, these patches are mapped to latent vectors through the linear transformation layer to generate block embeddings. The size of the generated patch embeddings remains the same as that of the original patches. In addition, since the Transformer itself does not have the ability to capture the order of elements in the sequence, the embodiments of the present application introduce a randomly trainable embedding class token position encoding in the patch embedding to retain the position information of each element in the sequence.

[0152] Referring to Figure 8 , Figure 8It is a specific flowchart for iteratively training a fault diagnosis model by inputting real - domain features and simulation - domain features provided by an embodiment of the present application. In some embodiments, the method includes but is not limited to step S701 and step S703.

[0153] Referring to Figure 9 , Figure 9 It is a schematic diagram of a multi - label decoupled classifier provided by an embodiment of the present application.

[0154] It should be noted that the fault diagnosis model includes a multi - label decoupled classifier.

[0155] Step S701: Perform transfer learning on real - domain features and simulation - domain features based on a sub - domain adaptation mechanism to obtain a local maximum mean discrepancy metric.

[0156] Step S702: Input real - domain features and simulation - domain features into the multi - label decoupled classifier for fault diagnosis and output predicted labels.

[0157] Step S703: Calculate the local maximum mean discrepancy metric and the predicted labels based on edge - aware regularization to iteratively train the fault diagnosis model.

[0158] In steps S701 to S703 of some embodiments, during the process of training the fault diagnosis model, transfer learning is performed on real - domain features and simulation - domain features based on a sub - domain adaptation mechanism to obtain a local maximum mean discrepancy metric, so as to obtain the distance expected value under each fault mode. Then, real - domain features and simulation - domain features are input into the multi - label decoupled classifier for fault diagnosis. Specifically, when the multi - label decoupled classifier identifies that the feature is a single fault, it outputs a label of the single fault; when the multi - label decoupled classifier identifies that the feature is a compound fault, it decouples the compound fault and outputs multiple fault labels. After the multi - label decoupled classifier outputs the predicted labels, the local maximum mean discrepancy metric and the predicted labels are calculated based on edge - aware regularization to improve the diagnostic accuracy of the fault diagnosis model, determine the loss function of the fault diagnosis model, and iteratively train the fault diagnosis model according to this loss function, so that the model with the lowest training set loss value in the subsequent iteration process is selected as the well - trained model.

[0159] Specifically, the calculation process of the local maximum mean discrepancy metric can refer to the following formula:

[0160]

[0161] where P c and Q c respectively represent the probability distributions of a certain fault category C in the simulation domain and the real domain; E[·] represents the mathematical expectation; ψ(·) represents the data feature mapping process; xs and x t are samples from the simulation domain and the real domain respectively; H is the feature mapping space, and the above formula focuses on the expected local distance of each fault mode.

[0162] Refer to Figure 10 , Figure 10 which is a schematic diagram of the sub-domain adaptive mechanism provided by the embodiments of the present application.

[0163] In addition, the process of transfer learning for real-domain features and simulation-domain features based on the sub-domain adaptive mechanism can be referred to the following formula:

[0164]

[0165] where n s and n t are the number of samples in the simulation domain and the real domain respectively; ω c is the weight, and represent the possibility that x s and x t samples belong to the fault category C; k(·) represents the kernel function for data feature mapping; G f (·) is the feature extractor. By minimizing this equation, the alignment of the conditional distributions of each sub-domain is achieved, that is, the alignment of the conditional distributions of the real domain and the simulation domain.

[0166] In some embodiments, the specific process of calculating the local maximum mean difference index and the predicted label based on edge-aware regularization can be referred to the following formula:

[0167]

[0168]

[0169] where x is the sample, y is the label value corresponding to the sample x, n is all samples in the real domain, and n c is the number of samples of type C faults in the real domain faults. represents the output value of type y i , and z c = G c (G f (x i )) c represents the output value of type C, and Δ c is the regularization penalty coefficient of type C.

[0170] The loss function of the final model can be referred to the following formula:

[0171]

[0172] Among them, is a weight coefficient that changes with time; t and t m are the current iteration number and the maximum iteration number respectively. θ f is the learning parameter of the feature extractor G f ; θ c is the learning parameter of the multi-label decoupling classifier G c .

[0173] It should be noted that the feature extractor in the above formula includes a first feature extractor and a second feature extractor.

[0174] In some embodiments, during the training iteration process, the embodiments of the present application use an Adam optimizer with weight decay regularization for minimization to optimize the network learning parameters, and the model with the lowest training set loss value during the iteration is selected as the well-trained model.

[0175] Referring to Figure 11 , Figure 11 is the specific flowchart of model testing for inputting the test set into the fault diagnosis model after iterative training provided by the embodiments of the present application. In some embodiments, the method includes but is not limited to step S801 and step S802.

[0176] Step S801, perform performance evaluation on the fault diagnosis model after iterative training by inputting the test set, and obtain evaluation metrics.

[0177] Step S802, when the evaluation metrics meet the preset metric conditions, use the trained fault diagnosis model as the target model.

[0178] In steps S801 to S802 of some embodiments, during the process of model testing for the fault diagnosis model after iterative training by inputting the test set, input the actually measured test set into the fault diagnosis model after iterative training for model testing to obtain evaluation metrics, where the evaluation metrics are used to evaluate the quality of the fault diagnosis model, so as to verify whether the model can accurately identify and decouple faults in actual applications. When the evaluation metrics meet the preset metric conditions, it indicates that the fault diagnosis model after iterative training meets the decoupling requirements, and the trained fault diagnosis model can be directly used as the target model to ensure the quality, reliability and practicability of the model, and improve the accuracy of decoupling faults of the fault diagnosis model.

[0179] It should be noted that the evaluation metrics in the embodiments of the present application include but are not limited to parameters such as precision, recall rate, F1 score, etc. The metric conditions can be set according to the parameters included in the evaluation metrics, and the embodiments of the present application do not make specific limitations.

[0180] Specifically, accuracy is used to characterize the percentage of correctly predicted results in the total samples, and its formula is as follows:

[0181]

[0182] Among them, TP means that the sample prediction result is a positive sample while the true sample is a positive sample, FP means that the sample prediction result is a positive sample while the true sample is a negative sample, TN means that the sample prediction result is a negative sample while the true sample is a negative sample, and FN means that the sample prediction result is a negative sample while the true sample is a positive sample.

[0183] Recall is used to characterize the probability that a sample actually being positive is predicted as a positive sample, and its formula is as follows:

[0184]

[0185] The F1 score is used to characterize the relationship between precision and recall, and its formula is as follows:

[0186]

[0187] Specifically, referring to Figure 12 , Figure 12 is a schematic diagram of the fault diagnosis model provided by the embodiments of the present application.

[0188] In some embodiments, the embodiments of the present application first obtain real-domain data and simulation-domain data, divide the real-domain data into a training set and a test set, establish a fault diagnosis model, and the fault diagnosis model includes a first feature extractor and a second feature extractor. Then, the real-domain data is respectively input into the first feature extractor for feature extraction to output real-domain features, and the simulation-domain data is input into the second feature extractor for feature extraction to output simulation-domain features. After that, transfer learning is performed on the real-domain features and the simulation-domain features based on the subdomain adaptation mechanism to obtain the local maximum mean discrepancy index. The real-domain features and the simulation-domain features are input into a multi-label decoupled classifier for fault diagnosis to output prediction labels. Then, based on edge-aware regularization, calculations are performed on the local maximum mean discrepancy index and the prediction labels to iteratively train the fault diagnosis model. After iteratively training the fault diagnosis model, the test set is input into the iteratively trained fault diagnosis model for performance evaluation, that is, the test set is input into the feature extractor to obtain relevant features, and then the relevant features are input into the multi-label decoupled classifier for fault diagnosis to output the labels corresponding to the test set, so as to be able to evaluate the pros and cons of the fault diagnosis model, ensure the quality, reliability and practicability of the model, and improve the accuracy of decoupling faults by the fault diagnosis model.

[0189] Please refer to Figure 13, An embodiment of the present application also provides a digital-twin-based fault diagnosis system, which can implement the above digital-twin-based fault diagnosis method. The system includes:

[0190] A data acquisition module 901, configured to acquire real-domain data and simulation-domain data, and divide the real-domain data into a training set and a test set. The real-domain data is used to represent single-fault data and compound-fault data of a rolling bearing, and the simulation-domain data is obtained by performing simulated data sampling on the single-fault data and the compound-fault data;

[0191] A model establishment module 902, configured to establish a fault diagnosis model, where the fault diagnosis model includes a first feature extractor and a second feature extractor;

[0192] A feature extraction module 903, configured to input the training set into the first feature extractor for feature extraction, output real-domain features, and input the simulation-domain data into the second feature extractor for feature extraction, outputting simulation-domain features;

[0193] A model training module 904, configured to iteratively train the fault diagnosis model based on the real-domain features and the simulation-domain features, and perform model testing on the iteratively trained fault diagnosis model through the test set to obtain a target model;

[0194] A fault identification module 905, configured to acquire a to-be-diagnosed fault set containing fault data, and input the to-be-diagnosed fault set into the target model for fault identification, outputting a target fault type.

[0195] The digital-twin-based fault diagnosis system of the embodiment of the present application is used to execute the digital-twin-based fault diagnosis method in the above embodiment. The specific processing process is the same as that of the digital-twin-based fault diagnosis method in the above embodiment, and will not be elaborated here one by one.

[0196] An embodiment of the present application also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor is used to execute the digital-twin-based fault diagnosis method in the above embodiment of the present application.

[0197] Please refer to Figure 14 , Figure 14 , which schematically shows the hardware structure of the electronic device provided by the embodiment of the present application. The electronic device includes:

[0198] The processor 1001 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0199] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), etc. The memory 1002 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1002 and are called by the processor 1001 to execute the digital twin-based fault diagnosis method of the embodiments of the present application;

[0200] The input / output interface 1003 is used to implement information input and output;

[0201] The communication interface 1004 is used to implement communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or through wireless means (such as mobile network, WI FI, Bluetooth, etc.);

[0202] The bus 1005 transmits information between various components of the device (such as the processor 1001, the memory 1002, the input / output interface 1003, and the communication interface 1004);

[0203] Among them, the processor 1001, the memory 1002, the input / output interface 1003, and the communication interface 1004 achieve communication connections with each other inside the device through the bus 1005.

[0204] The embodiments of the present application also provide a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned digital twin-based fault diagnosis method is implemented.

[0205] A memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0206] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

[0207] Those skilled in the art can understand that Figure 1-14 the technical solutions shown in do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figure, or combine certain steps, or different steps.

[0208] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0209] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices can be implemented as software, firmware, hardware, and appropriate combinations thereof.

[0210] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0211] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (item) of the following" or its similar expression refers to any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0212] In several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the above division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of systems or units can be in electrical, mechanical or other forms.

[0213] The units described above as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0214] In addition, each functional unit in various embodiments of this application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0215] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes: various media that can store programs, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0216] The preferred embodiments of the embodiments of this application have been described above with reference to the accompanying drawings, and this does not limit the scope of rights of the embodiments of this application. Any modifications, equivalent replacements, and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of this application shall be within the scope of rights of the embodiments of this application.

Claims

1. A fault diagnosis method based on digital twins, characterized in that: The method comprises: Acquire real domain data and simulation domain data, and divide the real domain data into a training set and a test set, wherein the real domain data is used to characterize single fault data and compound fault data of a rolling bearing, and the simulation domain data is obtained by performing simulated data sampling on the single fault data and the compound fault data; Establishing a fault diagnosis model, wherein the fault diagnosis model includes a first feature extractor and a second feature extractor; Input the training set into the first feature extractor for feature extraction, and output real domain features, and input the simulation domain data into the second feature extractor for feature extraction, and output simulation domain features; Iteratively training the fault diagnosis model based on the real domain features and the simulation domain features, and performing model testing on the iteratively trained fault diagnosis model through the test set to obtain a target model; Acquire a to-be-diagnosed fault set containing fault data, input the to-be-diagnosed fault set into the target model for fault identification, and output a target fault type; Wherein, the establishment of a fault diagnosis model includes: Establish intelligent diagnostic models; Establish a first feature extractor and a second feature extractor based on a multi-head self-attention mechanism; Performing feature alignment on the real domain data and the simulation domain data based on a subdomain adaptation mechanism of a local maximum mean difference metric to align conditional distributions of the real domain data and the simulation domain data; Acquire a real sample label of a rolling bearing, wherein the real sample label is used to characterize real fault data of the rolling bearing; Designing edge-aware regularization based on the real sample labels; Constructing a multi-label decoupling classifier based on the intelligent diagnosis model; A fault diagnosis model is established according to the first feature extractor, the second feature extractor and the multi-label decoupling classifier.

2. The fault diagnosis method according to claim 1, characterized in that: The simulation domain data is obtained by the following steps: Establish a digital twin model of rolling bearings based on the actual data of rolling bearings; The rolling bearing digital twin model is used to simulate the operating condition of the rolling bearing to obtain vibration signal data; The vibration signal data is classified to obtain simulation domain data.

3. The fault diagnosis method according to claim 1, characterized in that: The step of inputting the training set into the first feature extractor for feature extraction and outputting real domain features comprises: inputting the training set into the first feature extractor so that the first feature extractor divides the one-dimensional vibration signal in the training set and outputs a true patch sequence; Mapping the true patch sequence to a latent vector through a linear transformation layer of the first feature extractor to generate a true block embedding; Preserving the position information of the real block embedding and outputting real domain features; The step of inputting the simulation domain data into the second feature extractor for feature extraction and outputting simulation domain features comprises: inputting the simulation domain into the second feature extractor so that the second feature extractor divides the one-dimensional vibration signal in the simulation domain and outputs a simulation patch sequence; Mapping the simulated patch sequence to a latent vector via a linear transformation layer of the second feature extractor to generate a simulated block embedding; The position information of the simulation block embedding is retained, and simulation domain features are output.

4. The fault diagnosis method according to claim 1, characterized in that: The fault diagnosis model includes a multi-label decoupling classifier; iteratively training the fault diagnosis model based on the real domain features and the simulation domain features includes: Based on the subdomain adaptive mechanism, transfer learning is performed on the real domain features and the simulation domain features to obtain a local maximum mean difference index; Inputting the real domain features and the simulation domain features into the multi-label decoupling classifier for fault diagnosis, and outputting a predicted label; The local maximum mean difference index and the predicted label are calculated based on edge-aware regularization to iteratively train the fault diagnosis model.

5. The fault diagnosis method according to claim 1, characterized in that: The method of performing a model test on the iteratively trained fault diagnosis model through the test set to obtain a target model includes: Performing performance evaluation on the fault diagnosis model through the test set to obtain evaluation indicators; When the evaluation index meets the preset index conditions, the trained fault diagnosis model is used as the target model.

6. A fault diagnosis system based on digital twins, characterized in that: The system comprises: A data acquisition module, used to acquire real domain data and simulation domain data, and divide the real domain data into a training set and a test set, wherein the real domain data is used to characterize single fault data and compound fault data of a rolling bearing, and the simulation domain data is obtained by performing simulated data sampling on the single fault data and the compound fault data; A model building module, used to build a fault diagnosis model, the fault diagnosis model includes a first feature extractor and a second feature extractor; wherein, the building of the fault diagnosis model includes: building an intelligent diagnosis model; building the first feature extractor and the second feature extractor based on a multi-head self-attention mechanism; performing feature alignment on the real domain data and the simulation domain data based on a subdomain adaptive mechanism of a local maximum mean difference metric to align the conditional distribution of the real domain data and the simulation domain data; obtaining real sample labels of rolling bearings, wherein the real sample labels are used to characterize the real fault data of the rolling bearings; designing edge-aware regularization based on the real sample labels; constructing a multi-label decoupling classifier based on the intelligent diagnosis model; and building a fault diagnosis model based on the first feature extractor, the second feature extractor and the multi-label decoupling classifier; A feature extraction module, used for inputting the training set into the first feature extractor for feature extraction and outputting real domain features, and inputting the simulation domain data into the second feature extractor for feature extraction and outputting simulation domain features; A model training module, used for iteratively training the fault diagnosis model based on the real domain features and the simulation domain features, and performing model testing on the iteratively trained fault diagnosis model through the test set to obtain a target model; The fault identification module is used to obtain a set of to-be-diagnosed faults containing fault data, input the set of to-be-diagnosed faults into the target model for fault identification, and output a target fault type.

7. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the digital twin-based fault diagnosis method as described in any one of claims 1 to 6 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the digital twin-based fault diagnosis method as described in any one of claims 1 to 6.

Images