Fault diagnosis system, model training method, model, chip and electronic equipment

Abstract

Embodiments of the present application provide a fault diagnosis system, a model training method, a model, a chip and an electronic device. The fault diagnosis system comprises a feature learning module and a diagnosis module. The feature learning module comprises a first neural network model, which is trained by a MoCo algorithm. The diagnosis module comprises a second neural network model. The feature learning module is configured to obtain feature data of a running process of a component to be diagnosed, and perform feature extraction on the feature data by using the first neural network model to obtain a feature vector corresponding to the feature data. The diagnosis module is configured to perform feature classification on the feature vector by using the second neural network model to obtain a fault diagnosis result of the running process of the component to be diagnosed. By the above technical solution, the fault diagnosis of the mechanical component can be realized. Moreover, the above technical solution does not need to rely on expert experience, and the model automatically learns the features of different fault types and realizes fault diagnosis, and has stronger noise resistance.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a fault diagnosis system, model training method, model, chip, and electronic device. Background Technology

[0002] Bearings are critical components of mechanical equipment, playing vital roles in rotation, support, and force transmission, and are used in various types of machinery. Bearing failure can not only trigger a series of equipment malfunctions but also damage other components, thereby shortening equipment lifespan and even causing personal injury or significant economic losses. Therefore, bearing fault diagnosis is crucial. However, current bearing fault diagnosis technologies have limitations. Summary of the Invention

[0003] In view of this, one of the technical problems solved by the embodiments of this application is to provide a fault diagnosis system, a model training method, a model, a chip, and an electronic device to solve the above-mentioned technical problems.

[0004] In a first aspect, embodiments of this application provide a fault diagnosis system, including: a feature learning module and a diagnosis module, wherein the feature learning module includes a first neural network model, which is trained by the momentum comparison learning MoCo algorithm; and the diagnosis module includes a second neural network model.

[0005] The feature learning module is used to acquire feature data of the operating process of the component to be diagnosed, and to extract features from the feature data using the first neural network model to obtain the feature vector corresponding to the feature data.

[0006] The diagnostic module is used to perform feature classification on the feature vector using the second neural network model to obtain fault diagnosis results for the operation process of the component to be diagnosed.

[0007] Secondly, embodiments of this application provide a model training method, including:

[0008] Obtain a first training sample, which includes feature data of the target component operating in a target state; the target state includes a non-fault state and various different fault states.

[0009] The first training sample is used as the model input, and the first initial neural network model is trained based on the MoCo algorithm to obtain the trained first neural network model; the trained first neural network model is used to extract features from the input data.

[0010] Obtain a second training sample, which includes feature data of the target component operating in the target state and fault diagnosis labels corresponding to the feature data;

[0011] The second training sample is used as the model input to train the second initial neural network model, resulting in a trained second neural network model; the trained second neural network model is used to perform fault diagnosis on the input data.

[0012] Thirdly, embodiments of this application provide a model obtained based on the model training method described in the second aspect.

[0013] Fourthly, embodiments of this application provide a chip including the system described in the first aspect.

[0014] Fifthly, embodiments of this application provide an electronic device, including a processor, a communication interface, a memory, and a communication bus. The processor, the memory, and the communication interface communicate with each other through the communication bus. The memory is used to store computer programs, and when the processor executes the programs stored in the memory, it implements the model training method as described in the second aspect.

[0015] In a sixth aspect, embodiments of this application provide a computer storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the model training method as described in the second aspect.

[0016] This application provides a fault diagnosis system, a model training method, a model, a chip, and an electronic device. The fault diagnosis system provided in this application can diagnose faults in mechanical components. Furthermore, the above technical solution does not rely on expert experience; the model automatically learns the characteristics of different fault types and achieves fault diagnosis, resulting in stronger noise resistance. Attached Figure Description

[0017] The following describes some specific embodiments of the present application in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or components. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0018] Figure 1 A schematic flowchart illustrating a model training method provided in an embodiment of this application;

[0019] Figure 2 This is a schematic diagram of the structure of the fault diagnosis system provided in the embodiments of this application.

[0020] List of reference numerals in the attached diagram:

[0021] 101: Obtain the first training sample;

[0022] 102: Using the first training sample as the model input, train the first initial neural network model based on the MoCo algorithm to obtain the trained first neural network model;

[0023] 103: Obtain the second training sample, which includes the feature data of the target component operating in the target state and the fault diagnosis labels corresponding to the feature data;

[0024] 104: Use the second training sample as the model input to train the second initial neural network model and obtain the trained second neural network model;

[0025] 20: Fault diagnosis system;

[0026] 201: Feature learning module;

[0027] 202: Diagnostic module;

[0028] 203: Data acquisition module. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of this application.

[0030] As critical components in mechanical equipment, the failure of parts such as bearings and gearboxes can severely affect the normal operation of the equipment. Therefore, it is necessary to diagnose faults in these critical components to ensure timely detection and repair during equipment operation.

[0031] In some technical solutions, fault diagnosis of relevant components in mechanical equipment can be achieved based on signal processing methods. Specifically, signals from relevant components, such as vibration data, can be collected during the operation of the mechanical equipment. Then, feature extraction can be performed on the collected signals, such as extracting the amplitude of specific frequencies. Finally, the corresponding fault diagnosis results can be obtained through analysis of the signal features.

[0032] However, the above-mentioned technical solution relies heavily on expert experience, requiring experts to manually calculate fault characteristics and set relevant fault judgment thresholds. Furthermore, this technical solution is significantly affected by noise; when the noise is excessive, effective feature extraction will be impossible.

[0033] Based on the above issues, this application is submitted.

[0034] The specific implementation of the embodiments of this application will be further described below with reference to the accompanying drawings.

[0035] This application provides a model training method. Figure 1 A flowchart illustrating a model training method provided in this application embodiment is shown below. Figure 1 As shown, the method includes the following steps:

[0036] Step 101: Obtain the first training sample.

[0037] In this embodiment, the first training sample may include feature data of the target component operating in a target state. The target state may include a non-fault state and various different fault states. Thus, the first training sample may simultaneously include feature data of the target component under both the non-fault state and the feature data under different fault states. The target component may be any component in the target device, such as a bearing or gearbox.

[0038] In one exemplary implementation, the feature data of the target component's operation process may include vibration data; that is, vibration data can be used as training samples to train the model. However, in real-world scenarios, due to the structure of the equipment, vibration signal changes are usually abrupt, and vibration signals collected in adjacent time periods may have similar characteristics. Therefore, when training the model based on vibration signals, it may cause the model to repeatedly learn similar samples, leading to data imbalance or overfitting.

[0039] To address the aforementioned issues, another exemplary implementation can include both vibration and temperature data as feature data for the target component's operation. Since temperature data is typically continuously changing, introducing temperature data allows the model to learn from two different samples, even if vibration signals from two adjacent time periods are similar, as long as the temperature data differs. This prevents the model from blindly following trends due to similar vibration data. Furthermore, adding temperature data can compensate for the lag in fault diagnosis based on vibration data. Specifically, in the early stages of a fault, vibration characteristics may not yet show significant changes (not reaching a step point), but temperature data may already exhibit abnormal trends. Introducing temperature data allows the model to learn the relationship between temperature trends and faults, enabling it to detect risks through temperature trends even before vibration data reveals obvious fault characteristics, thus achieving early fault warning.

[0040] Specifically, temperature and vibration data of the target component can be collected during its operation in the target state. Then, a first training sample can be generated based on the temperature and vibration data. During each data acquisition operation, the acquisition frequency of temperature data can be lower than that of vibration data. For example, during each data acquisition operation, one temperature data point can be acquired within one second, and vibration data can be acquired within one second at a sampling frequency of 10240 Hz, resulting in 10240 discrete vibration signals. The acquired temperature and vibration data can then be aligned to form a first training sample. Specifically, a time-frequency transformation can be performed on the vibration data to obtain its frequency spectrum. For example, a Fast Fourier Transform can be performed on the vibration data to obtain the corresponding spectrum. Then, the first training sample can be obtained based on the frequency spectrum and temperature data. By transforming the vibration data from the time domain to the frequency domain, the model can better capture spectral features and learn the relationship between the peak value change trend of the vibration data spectrum and the temperature change trend, which can be used for feature recognition.

[0041] Step 102: Use the first training sample as the model input, and train the first initial neural network model based on the Momentum Contrast (MoCo) algorithm to obtain the trained first neural network model.

[0042] In real-world scenarios, components such as bearings are in normal condition most of the time, resulting in a small sample size of fault data and an imbalance between positive and negative samples. Therefore, in this embodiment, the first initial neural network model can be trained using the MoCo algorithm. The trained first neural network model can then be used as a "feature extractor" to extract features from the input data. Since the MoCo algorithm is a self-supervised learning algorithm, the first training samples do not require fault labels. The MoCo algorithm can effectively utilize a large amount of unlabeled data through comparative learning, bringing similar feature data closer together and dissimilar feature data further apart in the feature space, thereby learning highly discriminative features. Furthermore, the MoCo algorithm, through data augmentation techniques (such as noise superposition, amplitude scaling, and cyclic translation), enables the model to learn more essential features to adapt to fault diagnosis under different operating conditions.

[0043] The initial neural network model can include a Transformer model. By training the initial neural network model based on the MoCo algorithm, it can be trained as a "feature extractor" to extract features from the input data. For details on how to train the model using the MoCo algorithm, please refer to relevant technical documentation.

[0044] Step 103: Obtain the second training sample, which includes feature data of the target component operating in the target state and the fault diagnosis labels corresponding to the feature data.

[0045] In this embodiment, for example, a fault diagnosis label can be added to each sample data based on the first training sample to obtain a second training sample. The fault diagnosis label can be determined, for example, based on the target state of the target component from which the sample data originates. In real-world scenarios, there are usually more sample data in non-fault states than in fault states. To prevent data imbalance, in this embodiment, the training samples with different fault diagnosis labels can also be downsampled to balance the amount of data in the training samples with different fault diagnosis labels. Specifically, based on the fault diagnosis label, the feature data of the target component operating in the target state can be clustered to obtain multiple sample sets. Then, each sample set can be downsampled to obtain the second training sample.

[0046] Step 104: Use the second training sample as the model input to train the second initial neural network model and obtain the trained second neural network model.

[0047] Specifically, the second initial neural network model can be connected to the backend of the trained first neural network model to obtain the complete neural network model to be trained. Then, the second training samples can be used as the model input of the complete neural network model to train the second initial neural network model in the complete neural network model, thus obtaining the trained second neural network model.

[0048] In one exemplary implementation, the trained first neural network model in the complete neural network model can be frozen first. Then, the second training sample is used as the model input to the complete neural network model to train the second initial neural network model in the complete neural network model, resulting in a trained second neural network model. In this way, the trained first neural network model does not need to undergo backpropagation, thereby improving the training speed.

[0049] Alternatively, the trained first neural network model can be left unfrozen. In this way, the complete neural network model can be trained, not only training the second initial neural network model, but also fine-tuning the trained first neural network model, thereby correcting any deviations that may exist during the training of the first neural network model.

[0050] The trained second neural network model can be used as a "classification head" to diagnose faults in the input data and output specific fault categories and corresponding confidence levels.

[0051] The above technical solution allows for the training of a first neural network model and a second neural network model. The first neural network model can be used to extract features from the characteristic data of the target component's operation process, while the second neural network model can be used for further fault identification based on the extracted feature vectors, thereby achieving fault diagnosis of mechanical components such as bearings and gearboxes.

[0052] This application also provides a fault diagnosis system. Figure 2 This is a schematic diagram of the system structure provided in the embodiments of this application, such as... Figure 2 As shown, the fault diagnosis system 20 provided in this embodiment includes a feature learning module 201 and a diagnosis module 202. The feature learning module 201 includes a first neural network model, which is trained using the MoCo algorithm and may include a Transformer model. The diagnosis module 202 includes a second neural network model.

[0053] The following explains how the fault diagnosis system is implemented.

[0054] The feature learning module 201 is used to acquire feature data of the operation process of the component to be diagnosed, and to extract features from the feature data using the first neural network model to obtain the feature vector corresponding to the feature data.

[0055] The component to be diagnosed can be any part of a mechanical device, such as a bearing or gearbox. Characteristic data may include, for example, temperature and vibration data.

[0056] One possible implementation is, such as Figure 2As shown, the fault diagnosis system provided in this application embodiment may further include a data acquisition module 203. The data acquisition module 203 can be used to acquire temperature data and vibration data of the component to be diagnosed during its operation. It can also generate characteristic data of the component's operation process based on the temperature data and vibration data. Each time the data acquisition module 203 performs a data acquisition operation, the acquisition frequency of temperature data can be less than the acquisition frequency of vibration data. For example, the data acquisition module 203 can perform data acquisition once every 10 seconds, with each acquisition lasting for 1 second. During this period, the acquisition frequency of temperature data can be once per second, and the acquisition frequency of vibration data can be 10240Hz. When performing the step of generating characteristic data of the component's operation process based on temperature data and vibration data, the data acquisition module 203 can specifically be used to: perform time-frequency transformation on the vibration data to obtain the frequency spectrum data of the vibration data in the frequency domain. Then, based on the frequency spectrum data of the vibration data in the frequency domain and the temperature data, characteristic data of the component's operation process is obtained. For example, the frequency spectrum data corresponding to each acquired vibration data can be associated with the temperature data to obtain a set of characteristic data.

[0057] The diagnostic module 202 is used to perform feature classification on the feature vector using a second neural network model to obtain fault diagnosis results for the operating process of the component to be diagnosed. The fault diagnosis results may include, for example, the specific fault type and its corresponding confidence level.

[0058] The second neural network model may include a classification head, which can be used to classify the feature vectors output by the first neural network model to obtain fault diagnosis results, such as normal, inner circle fault, outer circle fault, etc.

[0059] The above technical solution enables fault diagnosis of mechanical components. Furthermore, this solution does not rely on expert experience; the model automatically learns the characteristics of different fault types and performs fault diagnosis, exhibiting stronger noise resistance. Simultaneously, the input data for fault diagnosis includes both vibration and temperature data. Based on the continuously changing nature of temperature data, it prevents the model's feature recognition and fault diagnosis accuracy from being affected by local similarities in vibration data.

[0060] This application also provides a model, which can be a model obtained based on the aforementioned model training method, including a first neural network model, a second neural network model, and any other neural network model that includes the first neural network model and / or the second neural network model.

[0061] The model provided in this application embodiment can be used to diagnose faults in related components of mechanical equipment, such as bearings and gearboxes.

[0062] This application also provides a chip that may include the fault diagnosis system provided in this application.

[0063] Based on the model training method described in any of the above embodiments, this application provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The memory is used to store computer programs, and the processor, when executing the program stored in the memory, implements the method described in any of the above embodiments.

[0064] Based on the model training method described in any of the above embodiments, this application provides a computer storage medium storing a computer program that, when executed by a processor, implements the method described in any of the above embodiments.

[0065] It should be noted that, depending on the implementation needs, the various components / steps described in the embodiments of this application can be broken down into more components / steps, or two or more components / steps or parts of the operation of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of this application.

[0066] The methods described in the embodiments of this application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code downloaded over a network that is originally stored in a remote recording medium or a non-transitory machine-readable medium and will be stored in a local recording medium. Thus, the methods described herein can be stored as software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the control methods of the adjustment machine described herein are implemented. Furthermore, when a general-purpose computer accesses code used to implement the control methods of the adjustment machine shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the control methods of the adjustment machine shown herein.

[0067] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0068] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this application.

[0069] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0070] The above embodiments are only used to illustrate the embodiments of this application, and are not intended to limit the embodiments of this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of this application. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of this application, and the patent protection scope of the embodiments of this application should be defined by the claims.

Claims

1. A fault diagnosis system, comprising: The system includes a feature learning module and a diagnostic module. The feature learning module includes a first neural network model, which is trained using the MoCo algorithm for momentum comparison learning. The diagnostic module includes a second neural network model. The feature learning module is used to acquire feature data of the operating process of the component to be diagnosed, and to extract features from the feature data using the first neural network model to obtain the feature vector corresponding to the feature data. The diagnostic module is used to perform feature classification on the feature vector using the second neural network model to obtain fault diagnosis results for the operation process of the component to be diagnosed.

2. The system according to claim 1, wherein, The system also includes a data acquisition module, which is used for: During the operation of the component to be diagnosed, temperature data and vibration data of the component to be diagnosed are collected; Based on the temperature and vibration data, characteristic data of the operating process of the component to be diagnosed are generated.

3. The system according to claim 2, wherein, Each time the data acquisition module performs a data acquisition operation, the acquisition frequency of the temperature data is less than the acquisition frequency of the vibration data.

4. The system according to claim 3, wherein, When the data acquisition module performs the step of generating characteristic data of the operating process of the component to be diagnosed based on the temperature data and vibration data, it is specifically used for: The vibration data is subjected to time-frequency transformation to obtain the frequency spectrum data of the vibration data in the frequency domain; Based on the frequency spectrum data of the vibration data and the temperature data, characteristic data of the operating process of the component to be diagnosed are obtained.

5. The system according to claim 1, wherein, The first neural network model includes the Transformer model.

6. A model training method, comprising: Obtain a first training sample, which includes feature data of the target component operating in the target state; The target state includes a non-fault state and various different fault states; The first training sample is used as the model input, and the first initial neural network model is trained based on the MoCo algorithm to obtain the trained first neural network model; the trained first neural network model is used to extract features from the input data. Obtain a second training sample, which includes feature data of the target component operating in the target state and fault diagnosis labels corresponding to the feature data; The second training sample is used as the model input to train the second initial neural network model, resulting in a trained second neural network model; the trained second neural network model is used to perform fault diagnosis on the input data.

7. The method according to claim 6, wherein, The process of obtaining the first training sample includes: During the operation of the target component in the target state, temperature data and vibration data of the target component are collected; Based on the temperature and vibration data, a first training sample is generated.

8. The method according to claim 7, wherein, Each time a data acquisition operation is performed, the frequency of acquiring the temperature data is less than the frequency of acquiring the vibration data; Based on the temperature and vibration data, a first training sample is generated, including: The vibration data is subjected to time-frequency transformation to obtain the frequency spectrum data of the vibration data in the frequency domain; Based on the frequency spectrum data of the vibration data and the temperature data, a first training sample is obtained.

9. The method according to claim 6, wherein, Obtain the second training sample, including: Based on the fault diagnosis labels, the feature data of the target component operating in the target state are clustered to obtain multiple sample sets; Each sample set is downsampled to obtain the second training samples.

10. The method according to claim 6, wherein, Using the second training sample as model input, the second initial neural network model is trained to obtain the trained second neural network model, including: The second initial neural network model is connected to the back end of the trained first neural network model to obtain the complete neural network model to be trained; The second training sample is used as the model input of the complete neural network model to train the second initial neural network model in the complete neural network model, thereby obtaining the trained second neural network model.

11. The method according to claim 10, wherein, Before training the second initial neural network model in the complete neural network model by using the second training sample as the model input, the method further includes: Freeze the trained first neural network model in the complete neural network model.

12. A model, said model being obtained based on the model training method as described in any one of claims 6-11.

13. A chip, comprising: The system as described in any one of claims 1-5.

14. An electronic device comprising: The processor, communication interface, memory, and communication bus are connected, with the processor, communication interface, and memory communicating with each other via the communication bus. Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method described in any one of claims 6-11.

15. A computer storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the method as described in any one of claims 6-11.

Images