A fault diagnosis method for multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions

Abstract

The present invention provides a method for diagnosing equipment faults using multi-sensor time-frequency fusion under class-imbalanced conditions, which belongs to the technical field of intelligent equipment fault diagnosis. The method comprises the following steps: constructing an equipment class-balanced data set based on the original class-imbalanced sensor data set through a denoising diffusion probability model; training an equipment fault identification model through the equipment class-balanced data set; and receiving multi-source sensor data using the pre-trained equipment fault identification model to output the type of equipment fault. The present invention expands the unbalanced data set to a balanced state by generating high-quality virtual data, thereby optimizing the training effect of the equipment fault identification model; and introduces a frequency domain feature extraction branch into the fault identification model, and designs a multi-scale gated convolution module and an efficient adaptive fusion module to achieve adaptive fusion of the time-frequency domain features of multi-source sensors, significantly improving the prediction accuracy of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent equipment fault diagnosis, and in particular to a multi-sensor time-frequency fusion equipment fault diagnosis method under quasi-unbalanced conditions. Background Art

[0002] With the acceleration of industrial intelligence, fault prediction and health management technologies play a core role in ensuring the reliable operation of critical equipment. Multi-sensor information fusion technology, by integrating monitoring data from various sources (such as vibration, temperature, and acoustic emission signals), can effectively overcome the limitations of single-sensor signals, such as limited coverage and weak interference immunity, significantly improving the accuracy and robustness of fault diagnosis. In recent years, intelligent diagnostic methods based on deep learning have further promoted the development of this field, becoming a key research direction for achieving accurate perception of equipment status and early warning of faults.

[0003] Currently, multi-sensor information fusion fault diagnosis methods primarily rely on time-domain signal features to construct classification models and achieve multi-source data fusion through simple feature addition or tensor concatenation. Some studies use convolutional neural networks or long-short-term memory networks to extract time series signal features, or introduce attention mechanisms to optimize feature weights during the fusion stage.

[0004] However, existing methods still face three significant limitations: First, existing model training uses real datasets, but fault samples are scarce in actual industrial scenarios and the data distribution is highly unbalanced, resulting in a sharp decline in the model's recognition performance for minority fault types; second, feature extraction is overly focused on the time domain, ignoring the fault characteristics contained in the video domain signal, and is prone to misjudgment in strong noise environments; finally, traditional fusion strategies such as splicing lack the ability to explore deep correlations between features, and redundant parameters increase the computational burden, limiting the model's applicability in industrial scenarios with high real-time requirements.

[0005] Therefore, a multi-sensor time-frequency fusion equipment fault diagnosis method under quasi-unbalanced conditions is needed. Summary of the Invention

[0006] In view of this, the present invention provides a multi-sensor time-frequency fusion equipment fault diagnosis method under class imbalance conditions, which constructs a device class balanced data set through a denoising diffusion probability model for equipment fault identification model training; the equipment fault identification model integrates the time domain and frequency domain diversity characteristics of multi-sensor data to improve its diagnostic performance, solving the problems of imbalanced training data in industrial practice and insufficient equipment fault identification accuracy caused by the prediction model ignoring frequency domain characteristics in the existing technology.

[0007] To this end, the present invention provides the following technical solutions:

[0008] A method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions, comprising:

[0009] Based on the original class-imbalanced sensor dataset, a device-class balanced dataset is constructed using a denoising diffusion probability model.

[0010] Train the device fault identification model using a balanced data set of device classes;

[0011] The pre-trained equipment fault identification model is used to receive multi-source sensor data and output the equipment fault type.

[0012] Furthermore, the equipment fault identification model includes:

[0013] Receive time domain data from multiple sensors and convert the time domain data of each sensor into frequency domain data using Fourier transform;

[0014] Through a multi-branch backbone network, dual-domain features of each sensor data are extracted based on the time domain data and frequency domain data of each sensor;

[0015] The second-stage efficient adaptive fusion module is used to fuse the dual-domain features of each sensor to obtain the multi-source features of the device.

[0016] Furthermore, each branch backbone network includes: a time domain network, a frequency domain network and a first-stage efficient adaptive fusion module;

[0017] The time domain network extracts time domain features;

[0018] The frequency domain network extracts frequency domain features;

[0019] The first-stage efficient adaptive fusion module fuses time domain features and frequency domain features to obtain dual-domain features of the sensor data.

[0020] Furthermore, the method of constructing a device-class balanced dataset based on the original class-unbalanced sensor dataset through a denoising diffusion probability model includes:

[0021] Collect raw class-imbalanced sensor data from the device;

[0022] Through the denoising diffusion probability model, Gaussian noise is gradually added to the original class-imbalanced sensor data, which is converted into pure noise after a preset step, forming a Markov chain and destroying the original data distribution;

[0023] Through neural network learning, the reverse denoising path is used to reconstruct the original data distribution and generate various types of balanced distributed sensor data.

[0024] Furthermore, the neural network includes:

[0025] Improved DiffWave network;

[0026] The improvements to the DiffWave network include:

[0027] Based on the DiffWave network, efficient multi-scale attention is introduced to enhance local multi-scale feature extraction;

[0028] And add two linear layers to suppress high-frequency noise.

[0029] Furthermore, the time domain network includes:

[0030] Several downsampling modules and multi-scale gated convolution modules;

[0031] The downsampling modules and the multi-scale gated convolution modules are arranged alternately;

[0032] The downsampling module extracts the transformation size and outputs multi-scale features;

[0033] The multi-scale gated convolution module fuses multi-scale features and outputs multi-scale fused features.

[0034] Furthermore, the time domain network and the frequency domain network have the same structure.

[0035] Furthermore, the first-stage efficient adaptive fusion module includes:

[0036] element-wise addition of the time domain features and the frequency domain features, and performing grouping operations to obtain fused features;

[0037] Convolution and pooling are performed on the fused features to obtain local features and global features;

[0038] Integrate local features and global features to obtain intermediate features;

[0039] Use the intermediate features to determine the corresponding weights of time domain features and frequency domain features;

[0040] Based on the corresponding weights, the time domain features and the frequency domain features are weighted and summed to obtain dual-domain features.

[0041] Furthermore, the multi-scale gated convolution module includes:

[0042] Obtain the corresponding weights of each scale feature through the gating mechanism;

[0043] Based on the weights corresponding to the scale features, the multi-scale features are weightedly fused to obtain multi-scale fused features;

[0044] The multi-scale fusion features are used to output the multi-scale gated convolution module output features through a linear layer.

[0045] Furthermore, the output feature of the last multi-scale gated convolution module in the time domain network is used as the time domain feature;

[0046] The output features of the last multi-scale gated convolution module in the frequency domain network are used as the frequency domain features.

[0047] Advantages and positive effects of the present invention:

[0048] The present invention expands the unbalanced data set to a balanced state by generating high-quality virtual data, thereby optimizing the training effect of the equipment fault identification model; and introduces frequency domain features into the fault identification model. Through a multi-scale gated convolution module and an efficient adaptive fusion module, the adaptive fusion of time-frequency domain features of multi-source sensors is achieved, significantly improving the prediction accuracy of the model.

[0049] 1) This paper introduces a denoising diffusion probability model for data augmentation, expanding unbalanced datasets to a balanced state by generating virtual data. Based on the existing denoising diffusion probability model, the DiffWave network is improved. This improved DiffWave network introduces a linear layer to filter high-frequency noise and employs a dual-branch architecture to regulate the dynamic global characteristics of bidirectional augmented convolution, thereby enhancing focus on key information and improving the quality of generated virtual data.

[0050] 2) The device fault identification model of this invention adds a frequency domain feature branch to enhance the classification model's recognition accuracy. It also utilizes a multi-scale gated convolution module, employing parallel large-core multi-scale deep convolution to adjust linear features and adaptive gating for weighted multi-scale fusion. This addresses the limited diversity of generated virtual data and the inconsistent distribution and duration of key events in the time and frequency domains.

[0051] 3) The equipment fault identification model of the present invention is equipped with an efficient adaptive fusion module to realize the two-stage fusion of time domain and frequency domain features of multiple sensors; and the attention feature fusion technology is combined with an efficient grouping mechanism and cross-space learning, which can save computing resources and enhance local-global information integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0053] Figure 1 This is a flow chart of the fault diagnosis method for multi-sensor time-frequency fusion equipment under quasi-imbalance conditions in Example 1;

[0054] Figure 2 This is a structural diagram of the multi-source and multi-scale time-frequency fusion convolutional neural network in Example 1;

[0055] Figure 3 This is a flow chart of the downsampling module and the multi-scale gated convolution module in Example 2;

[0056] Figure 4 This is a structural diagram of the efficient adaptive fusion module in Example 2;

[0057] Figure 5 This is the improved DiffWave network structure diagram in Example 3;

[0058] Figure 6 This is the module diagram of the rotating equipment test bench in the experiment;

[0059] Figure 7 This is the data generated by sensor 1 in the experiment based on the denoised diffusion probability model;

[0060] Figure 8 This is the data generated by sensor 2 in the experiment based on the denoised diffusion probability model;

[0061] Figure 9 This is a diagram of the model classification results based on different sensor data before and after generating data for the denoised diffusion probability model in the experiment. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0063] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0064] The present invention provides a method for fault diagnosis of multi-sensor time-frequency fusion equipment under class imbalance conditions. The method utilizes a denoising diffusion probability model to perform data enhancement on minority class samples from multiple sensors to suppress data imbalance. The model injects progressive Gaussian noise during the forward diffusion process and then uses an improved DiffWave (Diffusion Waveform Network) to learn the reverse generation process. Regarding the classification network: 1) A multi-scale gated convolution module is used to enhance the network's ability to extract diverse features from time and frequency domain data. 2) An efficient and adaptive feature fusion module is constructed to assign learnable weights to features from different branches in the channel and position dimensions, achieving feature-level data fusion. 3) The constructed multi-source, multi-scale, time-frequency fusion convolutional neural network utilizes a two-stage, efficient and adaptive feature fusion module to perform data fusion deep within the network, integrating the time and frequency domain information from different sensor data for fault diagnosis.

[0065] Example 1

[0066] Combine Figure 1 As shown, the multi-sensor time-frequency fusion device fault diagnosis method under quasi-imbalance conditions of this embodiment includes the following steps:

[0067] S1, based on the original class-imbalanced sensor dataset, a device class-balanced dataset is constructed through a denoising diffusion probability model;

[0068] S2. Train the equipment fault identification model using a balanced data set of equipment classes;

[0069] S3. Utilize the pre-trained fault identification model to receive multi-source sensor data and output the device fault type.

[0070] The equipment fault identification model in this embodiment is a multi-source, multi-scale, time-frequency fusion convolutional neural network. It improves diagnostic performance by fusing the time-domain and frequency-domain diversity features of multi-sensor data. The steps include:

[0071] 1) Receive time domain data from multiple sensors and convert each sensor’s time domain data into frequency domain data using Fourier transform;

[0072] 2) Through a multi-branch backbone network, dual-domain features of each sensor data are extracted based on the time domain data and frequency domain data of each sensor;

[0073] Each backbone network includes: time domain network, frequency domain network and the first stage efficient adaptive fusion module; the time domain network extracts time domain features; the frequency domain network extracts frequency domain features;

[0074] The time domain network and the frequency domain network have the same structure, both including: multiple multi-scale gated convolution modules and downsampling modules; abstract features are extracted through the multi-scale gated convolution module, and the transformation size is extracted through the downsampling module.

[0075] From the perspective of data change trends, time domain data and frequency domain data differ in the occurrence time and duration of key events. Using a fixed-size convolution kernel to obtain local information may be limited. Therefore, the multi-scale gated convolution module uses multiple different channel-by-channel convolutions to obtain spatial information of different scales, and combines the gated unit mechanism to assign learnable weights to information of different scales.

[0076] 3) Integrate the time and frequency domain features of individual sensor data through the first-stage efficient adaptive fusion module; achieve feature-level data fusion by assigning learnable weights to the time and frequency domain features of individual sensors;

[0077] 4) Through the second-stage efficient adaptive fusion module, the time domain and frequency domain features of multiple sensor data are integrated; learnable weights are assigned to each sensor data to achieve multi-source data fusion.

[0078] Example 2

[0079] Combine Figure 2 As shown, the backbone network is further explained:

[0080] In this embodiment, preferably, two sensor data are received;

[0081] Multi-scale features are obtained by downsampling, and the ability to extract features from time-frequency information is enhanced by a multi-scale gated convolution module. The structure is as follows: Figure 3 As shown;

[0082] 1) Downsampling module;

[0083] Downsampling obtains multi-scale features, and the formula is expressed as:

[0084]

[0085]

[0086]

[0087] in, Represents sensor time domain data or frequency domain data, and the size is expressed as B×C×L, which is 7×11×15 in this embodiment; 、 and are the characteristics of three scales, is a one-dimensional channel-wise convolution with dimensions of 7, 11, and 15, is a linear layer.

[0088] 2) Multi-scale gated convolution module;

[0089] Use the gating mechanism to obtain the corresponding weights of each scale feature:

[0090]

[0091]

[0092] Among them, Stack is a stacking operation. is a linear layer, represents the pre-fusion feature, Indicates the corresponding weights of multi-scale features; 3 indicates the number of multi-scale features; Represents the sum index variable.

[0093] The features of the three scales are Hadamard products with the weights and added together to obtain the multi-scale fusion features and the output features of the multi-scale gated convolution module. The formula is expressed as:

[0094]

[0095]

[0096] in, is the output feature of the multi-scale gated convolution module, is a linear layer, is the Hadamard product, and A represents the multi-scale fusion feature.

[0097] The output features of the last multi-scale gated convolution module of the time domain network are used as time domain features; the output features of the last multi-scale gated convolution module of the frequency domain network are used as frequency domain features.

[0098] 3) Combination Figure 4 As shown, the efficient adaptive fusion module in the method of the present invention combines an efficient grouping mechanism and cross-space learning, which not only saves computing resources but also improves the ability of attention features to fuse local information and global information.

[0099] Taking the first-stage efficient adaptive fusion module, which fuses the time domain features and frequency domain features of the same sensor, as an example, the efficient adaptive fusion module is further explained:

[0100] First, feature fusion, the formula is expressed as:

[0101]

[0102] Among them, Group represents the grouping operation, which merges some channel dimensions into the batch dimension to reduce the computational cost; In this embodiment, represents the time domain characteristics, represents the frequency domain feature; P is the fusion feature;

[0103] Next, local information and global information are obtained and cross-space interaction is performed. The formula is expressed as:

[0104]

[0105]

[0106]

[0107] in, Represents local features, Represents global features, is a one-dimensional channel-by-channel convolution of size 7, GAP is a global average pooling operation, represents a linear layer, is the Sigmoid function, It is an intermediate feature that integrates local-global information.

[0108] Finally, two weights are used to redistribute the input features:

[0109]

[0110]

[0111] Among them, Reshape is the operation of restoring the channel and batch dimension sizes. , are learnable weights, It is the output of the first stage efficient adaptive fusion module.

[0112] The output of the first-stage efficient adaptive fusion module in this implementation is the dual-domain features of a single sensor.

[0113] Similarly, the structure of the second stage efficient adaptive fusion module is the same, as shown in the following example: Figure 4 As shown; that is, input 1 and input 2 are the dual-domain features of the two sensors, and the output is the multi-source feature of the device.

[0114] Example 3

[0115] Based on the original class-imbalanced sensor dataset, a device-class balanced dataset is constructed using a denoising diffusion probability model.

[0116] In this embodiment, multiple sensors are used to simultaneously collect state monitoring vibration time domain signals of the equipment under various fault conditions to construct an original imbalance-like sensor data set.

[0117] Existing data augmentation methods have difficulty generating high-quality fault samples with realistic distribution.

[0118] Therefore, in this embodiment, a denoising diffusion probability model is used to generate virtual data to construct a device-class balanced data set.

[0119] The DiffWave network in the existing denoising diffusion probability model has deficiencies in noise suppression and feature preservation;

[0120] Therefore, in this embodiment, an improved DiffWave network is used in the inverse process of the denoising diffusion probability model to generate a balanced virtual-real mixed data set for training the fault recognition model.

[0121] The denoising diffusion probability model includes two stages: diffusion and inverse diffusion:

[0122] In the first stage, Gaussian noise is gradually added to the original data through the Markov chain. After the preset steps, the data degenerates into pure noise, forming a forward diffusion chain.

[0123] In the second stage, the neural network is trained to learn the reverse denoising path. The neural network model gradually reconstructs the original data distribution from the random noise by predicting the noise distribution at the current time step.

[0124] In this embodiment, the neural network is an improved DiffWave network with the following structure: Figure 4 Improvements include: introducing efficient multi-scale attention to enhance local multi-scale feature extraction, and adding two linear layers to suppress high-frequency noise.

[0125] The formula for two linear layers is expressed as:

[0126]

[0127] Among them, Linear represents the linear layer, and BiDilatedConv represents the bidirectional dilated convolution; represents the output of the efficient multi-scale attention module, y Represents the output after processing by convolution modulation technology.

[0128] Different from the traditional method of using a data set composed entirely of real data for training, in this embodiment, a virtual-real mixed balanced data set after data enhancement is used for training to achieve multi-sensor collaborative fault diagnosis under unbalanced conditions.

[0129] The effect of the present invention is further verified by experiments:

[0130] S1. Collect data of the gearbox under various operating conditions through the rotating equipment test bench.

[0131] like Figure 6 The test bench shown here includes a controller, brake, brake controller, gearbox, motor, and two accelerometers. The two accelerometers collect vertical and horizontal vibration signals, respectively. The experiment simulated five operating conditions, with a small number of samples for faulty operation and a 2% balance ratio. The detailed classification of class labels and samples is shown in Table 1.

[0132] Table 1

[0133]

[0134] S2. Set the sample length to 1024. In the denoising diffusion probability model, set the time step to 4000, the learning rate to 0.0002, and the step size T to 200. In the equipment fault identification model, set the number of cycles to 50, use the Adam optimizer, and set the learning rate to 0.001, which decreases as the number of iterations decreases.

[0135] S3. Experimental Procedure: First, data was generated using a denoising diffusion probability model. A dataset with a 2% balance ratio was restored to a virtual-real mixed dataset with a 100% balance ratio. At this point, the number of training samples for each category was 500. Next, a multi-source, multi-scale, time-frequency fusion convolutional neural network was trained using the virtual-real mixed dataset. Finally, the model performance was tested using real samples.

[0136] Experimental results:

[0137] Figure 7 The generated results of the denoised diffusion probability model for sensor 1 are shown. Figure 8 The generated results for sensor 2 using the denoised diffusion probability model are shown; the red and blue lines represent the generated data and the real data, respectively. As can be seen from the time and frequency domain plots, the generated virtual samples closely resemble the real samples, yet still exhibit a different diversity than the real samples. This provides preliminary evidence for the effectiveness of the denoised diffusion probability model.

[0138] Figure 9Model classification results based on different sensor data before and after data generation for the denoised diffusion probability model. When the balance ratio was 2%, the diagnostic accuracy based on sensor 1 and sensor 2 data was 89.7% and 87.3%, respectively, and the accuracy based on sensor 1+2 data was 96.4%. When the balance ratio was 100%, these results increased to 97.9%, 97.1%, and 99.1%, respectively. The denoised diffusion probability model significantly improved diagnostic performance. On both the unbalanced dataset and the balanced virtual-real mixed dataset, the accuracy of the multi-source, multi-scale time-frequency fusion convolutional neural network based on sensor 1+2 data was higher than that based on single sensor data. Therefore, the experimental results verify the effectiveness of the proposed method.

[0139] S4. Further ablation experiments were conducted to verify the effectiveness and rationality of the key steps in the method of the present invention. The experimental results are shown in Table 2.

[0140] In order to verify the effectiveness of combining frequency domain features, method A was constructed, which removed the frequency domain network of the equipment fault identification model.

[0141] To verify the role of multi-scale gated convolution, method B was constructed, which removes the multi-scale convolution and gating mechanism of the equipment fault identification model.

[0142] In order to verify the role of the efficient adaptive fusion module, method C was constructed. This method removes the efficient adaptive module of the equipment fault identification model and only uses the conventional addition operation for fusion.

[0143] In order to improve the role of DiffWave network in denoising diffusion probability model, method D is constructed, which uses the original DiffWave network to generate data.

[0144] Table 2 Ablation experiment results

[0145]

[0146] As can be seen, the proposed method outperforms other methods, while the accuracy of the method with several key operations removed is significantly reduced. The improved DiffWave network can improve diagnostic performance by 2.69%. Furthermore, experimental results demonstrate the effectiveness of the proposed multi-scale gated convolution, efficient adaptive fusion module, and frequency domain branch.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for fault diagnosis of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions, characterized in that: include: Based on the original class-imbalanced sensor dataset, a device-class balanced dataset is constructed using a denoising diffusion probability model. Train the equipment fault identification model using a balanced dataset of equipment classes; Use the pre-trained equipment fault identification model to receive multi-source sensor data and output the equipment fault type; The multi-source sensor data includes: an acceleration sensor collecting a vibration signal in the vertical direction, and another acceleration sensor collecting a vibration signal in the horizontal direction; The equipment fault identification model is a multi-source and multi-scale time-frequency fusion convolutional neural network, including: Receive time domain data from multiple sensors and convert the time domain data of each sensor into frequency domain data using Fourier transform; Through a multi-branch backbone network, dual-domain features of each sensor data are extracted based on the time domain data and frequency domain data of each sensor; Each branch backbone network includes: first-stage efficient adaptive fusion module; The first-stage efficient adaptive fusion module includes: Add the time domain features and frequency domain features element by element and perform grouping operation to obtain fusion features; Convolution and pooling are performed on the fused features to obtain local features and global features; Integrate local features and global features to obtain intermediate features. The formula is expressed as: in, Represents local features, Represents global features, is the Sigmoid function, It is an intermediate feature; Use the intermediate features to determine the corresponding weights of time domain features and frequency domain features; Based on the corresponding weights, performing weighted summation on the time domain features and the frequency domain features to obtain dual-domain features; The structure of the second-stage efficient adaptive fusion module is the same as that of the first-stage efficient adaptive fusion module. The input is the dual-domain features of the two sensors, and the output is the multi-source features of the device.

2. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 1, characterized in that: Each branch backbone network further comprises: a time domain network and a frequency domain network; the time domain network extracts time domain features; The frequency domain network extracts frequency domain features.

3. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 1, characterized in that: The method of constructing a device-class balanced dataset based on the original class-unbalanced sensor dataset by using a denoising diffusion probability model includes: Collect raw class-imbalanced sensor data from the device; Through the denoising diffusion probability model, Gaussian noise is gradually added to the original class-imbalanced sensor data, which is converted into pure noise after a preset step, forming a Markov chain and destroying the original data distribution; Through neural network learning, the reverse denoising path is used to reconstruct the original data distribution and generate various types of balanced distributed sensor data.

4. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 3, characterized in that: The neural network comprises: Improvements to the DiffWave network, including: Based on the DiffWave network, we introduce efficient multi-scale attention to enhance local multi-scale feature extraction, and add two linear layers to suppress high-frequency noise. The output of the efficient multi-scale attention is passed through a linear layer and then input into the bidirectional dilated convolution. At the same time, the output of the efficient multi-scale attention is convolved with the output of the bidirectional dilated convolution after passing through another linear layer. The formula is expressed as: Among them, Linear represents the linear layer, and BiDilatedConv represents the bidirectional dilated convolution; represents the output of the efficient multi-scale attention module, y Represents the output after processing by convolution modulation technology.

5. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 2, characterized in that: The time domain network includes: Several downsampling modules and multi-scale gated convolution modules; The downsampling modules and the multi-scale gated convolution modules are arranged alternately; The downsampling module extracts the transformation size and outputs multi-scale features; The multi-scale gated convolution module fuses multi-scale features and outputs multi-scale fused features.

6. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 5, characterized in that: The time domain network has the same structure as the frequency domain network.

7. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 5, characterized in that: The multi-scale gated convolution module includes: Obtain the corresponding weights of each scale feature through the gating mechanism; Based on the weights corresponding to the features at each scale, the multi-scale features are weightedly fused to obtain the multi-scale fusion features and the output features of the multi-scale gated convolution module. The formula is expressed as follows: in, is the output feature of the multi-scale gated convolution module, is a linear layer, It is Hadamard. represents multi-scale fusion features, Represents sensor time domain data, represents the pre-fusion feature, Indicates the weights corresponding to multi-scale features.

8. The method for diagnosing faults of multi-sensor time-frequency fusion equipment under quasi-unbalanced conditions according to claim 7, characterized in that: The output feature of the last multi-scale gated convolution module in the time domain network is used as the time domain feature; The output features of the last multi-scale gated convolution module in the frequency domain network are used as the frequency domain features.

Images