Rotating machine fault diagnosis device and method

Abstract

The invention discloses a rotating machinery fault diagnosis device and method of a double-branch multi-scale fusion network based on a time-frequency image, and the device comprises a data collection module which obtains a vibration signal of rotating machinery equipment; the data preprocessing module is used for carrying out division preprocessing on the vibration signal data by adopting a sliding window piecewise function; the data conversion module is used for converting the vibration signal data into a time-frequency image and a time-frequency feature by adopting continuous wavelet transform; the feature extraction module comprises a time-frequency feature extractor and a multi-scale feature extractor, and is used for respectively analyzing and processing the time-frequency features to extract numerical branch features and analyzing and processing the time-frequency images to extract multi-scale visual branch features; and the dynamic weighted fusion module is used for carrying out weighted fusion on the numerical branch features and the multi-scale visual branch features through dynamically generated weights and outputting a fault diagnosis result. The rotating machine fault diagnosis device and method have high rotating machine fault diagnosis efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a rotating machinery fault diagnosis device and method based on a dual-branch multi-scale fusion network of time-frequency images. Background Art

[0002] Rotating machinery is widely used in various fields such as industry, energy, and transportation, such as steam turbines, generators, compressors, pumps, and other equipment. These mechanical equipment will inevitably have various faults in long-term operation, such as bearing damage, imbalance, gear wear, and rotor eccentricity. If these faults are not diagnosed and handled in time, they will not only lead to equipment performance degradation, but also may cause serious safety accidents and economic losses. With the development of industrial automation and intelligence, higher requirements are placed on the real-time monitoring and fault diagnosis of the operating status of rotating machinery.

[0003] Traditional fault diagnosis methods usually rely on experienced technicians to judge the equipment status through manual observation and analysis of data such as vibration signals. However, this method has the problems of strong subjectivity and low diagnostic efficiency, and it is difficult to meet the needs of the complex production environment of modern industry. Therefore, developing a high-precision and high-efficiency fault diagnosis method is of great practical significance for comprehensively preventing equipment failures, minimizing downtime, and significantly reducing maintenance costs.

[0004] In order to solve the problem that the fault diagnosis effect is not ideal under complex working conditions, some fault diagnosis methods for non-stationary and nonlinear signal working condition data sets have been proposed. For example, the Chinese patent document with publication number CN119494022A discloses a method for constructing a mechanical equipment fault diagnosis model, a diagnosis method, a system and a device, which relates to the technical field of mechanical fault diagnosis. The fault diagnosis method includes: obtaining the vibration signal data of the rotating machinery in real time; inputting the vibration signal data into the fault diagnosis model to obtain the fault type of the rotating machinery. The Chinese patent document with publication number CN113177537A discloses a fault diagnosis method for rotating mechanical equipment, including: extracting the time domain feature data, frequency domain feature data and time-frequency domain feature data of the original vibration signal of the equipment to be diagnosed; performing curve secondary encoding on the time-frequency domain feature data to obtain a curve secondary encoding sequence; calculating the distance between the curve secondary encoding sequence and the clustering centers of multiple training samples of known fault types to obtain multi-dimensional distance feature data of the original vibration signal; inputting the time domain feature data, frequency domain feature data and distance feature data of the original vibration signal of the equipment to be diagnosed into the trained fault diagnosis model to obtain the fault type of the equipment to be diagnosed.

[0005] However, there are significant deficiencies in directly processing vibration signals. Vibration signals are usually one-dimensional and it is difficult to express the change information of time and frequency at the same time. The present invention converts the one-dimensional vibration signal into a two-dimensional time-frequency image by using continuous wavelet transform (CWT), in which the Morlet mother wavelet function can more effectively capture the instantaneous characteristics and dynamic change laws of the signal, especially for non-stationary signals, which makes the ability to capture the characteristics of non-stationary signals weaker; at the same time, direct processing of vibration signals is more sensitive to noise and easily interferes with the diagnosis results; in addition, it often requires complex manual feature extraction, relies on expert experience, is inefficient and highly subjective.

[0006] Therefore, it is necessary to find an effective method to solve the problems faced in processing vibration signals in fault diagnosis and improve the efficiency and accuracy of fault diagnosis methods. Summary of the invention

[0007] The present invention provides a rotating machinery fault diagnosis device and method based on a dual-branch multi-scale fusion network of time-frequency images, which has high rotating machinery fault diagnosis efficiency and accuracy.

[0008] The technical solution of the present invention is as follows: A rotating machinery fault diagnosis device based on a dual-branch multi-scale fusion network of time-frequency images, comprising: A data acquisition module, which acquires the vibration signal data to be detected of the rotating mechanical equipment; The data preprocessing module uses a sliding window piecewise function to divide the vibration signal data to be detected and performs normalization and standardization preprocessing; A data conversion module, which uses continuous wavelet transform to convert the pre-processed vibration signal data to be detected into a time-frequency image and corresponding time-frequency features; The feature extraction module includes a parallel time-frequency feature extractor and a multi-scale feature extractor; the time-frequency feature extractor analyzes and processes the time-frequency features to extract numerical branch features, and the multi-scale feature extractor analyzes and processes the time-frequency image to extract multi-scale visual branch features; The dynamic weighted fusion module performs weighted fusion of numerical branch features and multi-scale visual branch features through dynamically generated weights and outputs fault diagnosis results.

[0009] When capturing the dynamic changes and nonlinear characteristics of the signal, the present invention adopts a dual-branch feature extraction module to process the time-frequency image and the time-frequency features respectively. While considering the processing of the time-frequency image, the multi-scale feature branch is used to take into account the multi-scale characteristics of the vibration signal. Finally, the dynamic weighted fusion module is used to merge them, which can more finely capture the dynamic changes and nonlinear characteristics of the vibration signal, thereby improving the accuracy of fault diagnosis and the robustness in complex fault modes and noise environments.

[0010] Preferably, the segment length of the sliding window is 1024 and the overlap length is 512. It is ensured that the segmented signal covers multiple periods and retains sufficient time resolution.

[0011] Preferably, the continuous wavelet transform uses a Morlet mother wavelet with a scale range of 1-128.

[0012] The preprocessed vibration signal data to be detected is converted into time-frequency images and corresponding time-frequency features through continuous wavelet transform. The time-frequency images are used to extract the visual branch features of the model, and the characteristic files are used to extract the numerical branch features of the model. By fusing the information in the time domain and frequency domain, the dynamic characteristics of non-stationary signals can be effectively captured, especially the laws of short-term transient changes when a fault occurs. The multi-scale characteristics of the signal can be more intuitively displayed, which helps the model to automatically extract more comprehensive and accurate features, and improve the model's ability to recognize fault modes under complex working conditions, thereby significantly improving the accuracy and robustness of fault diagnosis.

[0013] The time-frequency feature extractor is constructed based on a convolutional neural network and is used to extract and optimize the time-frequency features layer by layer.

[0014] The time-frequency feature extractor includes several convolutional pooling modules; each convolutional pooling module includes an upsampling layer, a convolutional layer, a BatchNorm2d layer, a PReLU activation layer, a Dropout layer and a one-dimensional MaxPool layer; an attention module is introduced after each convolutional pooling module.

[0015] The upsampling layer is used to adjust the input to a fixed dimension to adapt to the requirements of the subsequent network structure and enhance the uniformity of the input. The convolution layer is used to extract fault features in the time-frequency features. The BatchNorm2d layer is used to accelerate model convergence, improve the stability of the training process, and enhance the robustness of the model. The Dropout layer is used to improve training efficiency and generalization ability. The PReLU layer is used to adapt to the dynamic changes of different features. The one-dimensional MaxPool layer is used to downsample the extracted features. The features extracted by different convolutional layers are sent to the attention modules respectively. Each attention module generates the corresponding attention weights through the compression and reweighting mechanism of the global features, and dynamically adjusts the channel importance of the input features to obtain the feature representation after channel attention weighting. The output weighted features are then further compressed in size through the pooling operation, and a multi-level feature representation is formed after splicing or superposition.

[0016] The time-frequency feature extractor of the present invention can not only fully extract the deep features in the time-frequency features, but also highlight the role of key features through a dynamic weighting mechanism, thereby significantly improving the model's ability to analyze non-stationary signal fault modes.

[0017] The multi-scale feature extractor is used to extract multi-scale features of the time-frequency image.

[0018] The multi-scale feature extractor includes several multi-convolution kernel modules and a dimensionality reduction module; the multi-convolution kernel module includes multiple convolution kernels of different sizes, which are used to extract detail features, mid-scale spatial features, and global range features of time-frequency images; the dimensionality reduction module includes a global average pooling layer and a fully connected layer, which performs dimensionality reduction processing on the output results of the multi-convolution kernel module.

[0019] Each multi-convolution kernel module includes 1×1 convolution kernel, 3×3 convolution kernel and 5×5 convolution kernel. The 1×1 convolution kernel captures local detail features, the 3×3 convolution kernel extracts mid-scale spatial features, and the 5×5 convolution kernel obtains global feature representations, thereby effectively covering information at different scales in the time-frequency image and avoiding the risk of missing key features with a single convolution kernel. After extracting the multi-scale features, the output results are reduced in dimension through the dimensionality reduction module, including global average pooling and full connection operations, mapping multi-channel features to a low-dimensional unified representation to ensure that the feature dimensions match the subsequent weighted fusion operations.

[0020] The multi-scale feature extractor of the present invention can fully capture the local, detailed and global characteristics of the time-frequency image, and provide comprehensive and accurate feature support for the identification of complex fault modes.

[0021] The dynamic weighted fusion module includes a feature splicing unit, a weight generating unit and a weighted fusion unit; the feature splicing unit splices the feature vectors output by the time-frequency feature extractor and the multi-scale feature extractor in the channel dimension to form a joint feature representation; the weight generating unit processes the joint features through a fully connected network, generates the weight of each branch, and normalizes the weight using a Softmax function; the weighted fusion unit performs a weighted summation of the feature vectors of the two branches according to the generated weights, generates a fused feature representation, and outputs a fault diagnosis result.

[0022] Preferably, the training method of the feature extraction module and the dynamic weighted fusion module includes: The vibration signals generated by rotating machinery under different working conditions are collected, the vibration signals are processed by sliding window segmentation, and the time-frequency images and corresponding time-frequency features are generated by continuous wavelet transform to construct a training data set; The feature extraction module and the dynamic weighted fusion module are trained using a training data set. During the training process, the time-frequency feature extractor and the multi-scale feature extractor extract features from the input time-frequency features and time-frequency images respectively. The dynamic weighted fusion module performs weighted fusion on the feature vectors of the two branches and outputs the predicted fault category label. The deviation between the current output and the expected target is calculated through forward propagation, and the weights and biases of the parameters of each layer are continuously optimized and adjusted through the error back propagation algorithm.

[0023] Furthermore, the optimization process uses the cross entropy loss function as the objective function, updates the parameters through the Adam optimizer, sets the initial learning rate to 0.005, and iterates 50 rounds of training.

[0024] A rotating machinery fault diagnosis method based on a dual-branch multi-scale fusion network of time-frequency images, comprising: The data acquisition module acquires the vibration signal data to be detected of the rotating mechanical equipment; The data preprocessing module uses a sliding window piecewise function to divide the vibration signal data to be detected, and performs normalization and standardization preprocessing; The data conversion module uses continuous wavelet transform to convert the pre-processed vibration signal data to be detected into time-frequency images and corresponding time-frequency features; The feature extraction module includes a parallel time-frequency feature extractor and a multi-scale feature extractor; the time-frequency feature extractor is used to analyze and process the time-frequency feature to extract the numerical branch feature, and the multi-scale feature extractor is used to analyze and process the time-frequency image to extract the multi-scale visual branch feature; The dynamic weighted fusion module performs weighted fusion of numerical branch features and multi-scale visual branch features through dynamically generated weights and outputs fault diagnosis results.

[0025] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses a continuous wavelet transform module to convert vibration signals into time-frequency images and time-frequency features, which can fully combine the information in the time domain and the frequency domain, and effectively capture the key dynamic features in non-stationary signals, especially the laws of short-term transient changes when a fault occurs. Compared with directly processing one-dimensional signals, time-frequency images can more intuitively display the multi-scale characteristics of signals, which helps the model to automatically extract more comprehensive and accurate features, and improve the model's ability to identify fault modes under complex working conditions, thereby significantly improving the accuracy and robustness of fault diagnosis.

[0026] The present invention combines time-frequency feature analysis and processing with multi-scale feature analysis and processing. The time-frequency feature branch extracts high-dimensional time-frequency features through convolution layers and pooling layers, which can capture the dynamic change law of the signal; the multi-scale feature branch extracts features at different scales and enhances the robustness of the model through multi-scale convolution and pooling operations, making full use of the global features of the time-frequency image and the local detail information of the multi-scale features, thereby improving the fault diagnosis capability for non-stationary and complex signals; the multi-scale feature branch effectively reduces the computational complexity through the dimensionality reduction module while retaining key information; the dynamic weighted fusion module adaptively generates weights for the two branches, which can allocate the optimal fusion ratio according to the features of different input data, significantly improving the accuracy and generalization ability of the diagnosis results.

[0027] The present invention shows good performance in feature extraction, fusion and model robustness, and is suitable for rotating machinery fault diagnosis in non-stationary and nonlinear signal industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the structural principle of a rotating machinery fault diagnosis method and system based on a dual-branch multi-scale fusion network of time-frequency images of the present invention; Figure 2 This is a flow chart of the diagnostic model training of the present invention; Figure 3 This is a diagram of the fault identification accuracy of the fault diagnosis model of the present invention. DETAILED DESCRIPTION

[0029] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0030] The present invention provides a rotating machinery fault diagnosis method and system based on a dual-branch multi-scale fusion network of time-frequency images, such as Figure 1 As shown, the continuous wavelet transform module is used to convert the vibration data in the data set into a time-frequency image; a dual-branch multi-scale convolutional network architecture is constructed to extract features from the preprocessed time-frequency images, including a time-frequency feature analysis module and a multi-scale feature analysis module; a dynamic weighted fusion module is built for data integration and output; a time-frequency dual-branch multi-scale fusion network fault diagnosis model is built based on the dual-branch multi-scale convolutional network architecture and the dynamic weighted fusion module, and the training model is used for fault diagnosis.

[0031] The sequence data of the vibration signal is used as the input signal. After the input signal is standardized, the signal is segmented using the sliding window piecewise function. The length of each segment is set to 1024 and the overlapping length is 512 to ensure that the segmented signal covers multiple cycles and retains sufficient time resolution. Subsequently, the signal data is processed by continuous wavelet transform, the Morlet mother wavelet is selected, the scale range is set to 1 to 128, and the generated time-frequency image and the corresponding characteristic file are saved separately. The time-frequency image is used for model visual branch feature extraction, and the characteristic file is used for model numerical branch feature extraction. By fusing the information in the time domain and frequency domain, the dynamic characteristics of non-stationary signals can be effectively captured, especially the law of short-term transient changes when a fault occurs. The multi-scale characteristics of the signal can be more intuitively displayed, which helps the model to automatically extract more comprehensive and accurate features, and improves the model's ability to recognize fault modes under complex working conditions, thereby significantly improving the accuracy and robustness of fault diagnosis.

[0032] The time-frequency feature branch is based on a convolutional neural network module, which consists of a convolutional pooling module and an attention module, and is used to extract and optimize time-frequency features layer by layer. The convolutional pooling module includes an upsampling layer, a convolutional layer, a BatchNorm2d layer, a PReLU activation layer, a Dropout layer, and a one-dimensional MaxPool layer, wherein the upsampling layer adjusts the input time-frequency feature dimension to a fixed size, the convolutional layer extracts the local features and global patterns of the signal through a 3×3 convolution kernel, the BatchNorm2d layer accelerates model convergence and improves stability, the PReLU activation layer enhances the nonlinear expression ability of the feature, the Dropout layer reduces the risk of overfitting, and the MaxPool layer extracts key information by reducing the size of the feature map. In addition, an attention module is introduced after each layer of the convolutional pooling module, the channel importance is calculated through global average pooling, and dynamic weights are generated through a fully connected network to weighted optimize the feature channels, thereby enhancing the expression ability of key features. This design can not only fully extract the deep features in the time-frequency features, but also highlight the role of key features through a dynamic weighting mechanism, significantly improving the model's ability to analyze non-stationary signal fault modes.

[0033] Taking the dual-branch multi-scale fusion network architecture as an example, assuming that the time-frequency feature dimension after continuous wavelet transform (CWT) is (127, 1024), the number of categories (num_class) is set to 10, and the time-frequency feature branch network parameters are as follows in Table 1:

[0034] The multi-scale feature branch is based on a multi-scale convolution structure, including a multi-convolution kernel module composed of convolution kernels of different sizes (such as 1×1, 3×3 and 5×5) and a dimension reduction module with average pooling, which is used to extract multi-scale features of time-frequency images. The multi-convolution kernel module captures local detail features through a 1×1 convolution kernel, extracts mid-scale spatial features through a 3×3 convolution kernel, and obtains global feature representations through a 5×5 convolution kernel, thereby effectively covering information at different scales in the time-frequency image and avoiding the risk of missing key features with a single convolution kernel. After the multi-scale features are extracted, the output results are reduced in dimension through a dimension reduction module, including global average pooling and full connection operations, mapping multi-channel features to a low-dimensional unified representation to ensure that the feature dimensions match the subsequent weighted fusion operations. The multi-scale design of this branch can fully capture the local, detailed and global characteristics of the time-frequency image, providing comprehensive and accurate feature support for the identification of complex fault modes.

[0035] Taking the dual-branch multi-scale fusion network architecture as an example, assuming that the dimension of the time-frequency image after continuous wavelet transform (CWT) is (3, 127, 128), the number of categories (num_class) is set to 10, and the multi-scale feature branch network parameters are as follows in Table 2:

[0036] The dynamic weighted fusion module adaptively integrates the outputs of the time-frequency feature branch and the multi-scale feature branch through three steps: feature splicing, weight generation and weighted fusion. First, the feature vectors of the two branches are spliced ​​in the channel dimension to form a joint feature representation; then, the joint features are processed through a fully connected network to generate two fusion weights, where the first fully connected layer reduces the dimension to extract important features, and the second fully connected layer outputs the fusion weights. Finally, the weights are normalized using the Softmax function to ensure that their sum is 1. Finally, the feature vectors of the two branches are weighted and summed according to the generated weights to generate a fusion feature representation and output the fault diagnosis results. This module realizes adaptive optimization based on input data by dynamically adjusting the importance of branch features, which not only improves the robustness of the fusion features, but also enhances the fault diagnosis capability of the model under complex working conditions.

[0037] like Figure 2The fault diagnosis model of the present invention is shown in the figure. The vibration signals generated by the rotating machinery under different working conditions are collected, such as the vibration signals of the normal working state and the vibration signals of various fault states, and they are sorted into corresponding data sets. Subsequently, the vibration signal is processed by sliding window segmentation, each segment is 1024 points long, 512 points overlapped, and the time-frequency image and the corresponding time-frequency characteristic file are generated by continuous wavelet transform. The sorted data set is randomly extracted, 70% of which are used as training sets, 15% as validation sets, and 15% as test sets. The training set and validation set are used to train the model. During the training process, the diagnostic model network calculates the deviation value between the current output and the expected target by forward propagation, and continuously adjusts the weights and biases of the parameters of each layer by the error back propagation algorithm. During the training process, the time-frequency feature branch and the multi-scale feature branch extract features from the input time-frequency image respectively, and the dynamic weighted fusion module integrates and optimizes the feature vectors of the two branches, and outputs the predicted fault category label. The entire optimization process uses the cross entropy loss function as the objective function, updates the parameters through the Adam optimizer, the initial learning rate is set to 0.005, and the training iteration is 50 rounds. After multiple rounds of iterations, the performance of the network is continuously improved on the training set and validation set, and finally converges to the optimal solution, obtaining a model with high diagnostic accuracy. In order to verify the performance of the model, the reserved test set is input into the trained diagnostic model. The model extracts and fuses features layer by layer and outputs the corresponding fault type prediction label. By comparing the model prediction label with the actual label of the test set, the recognition accuracy of the fault diagnosis is calculated. Figure 3 The dynamic weighted fusion network of the present invention shows excellent fault diagnosis ability in the test set, with an average diagnosis accuracy of more than 95%, demonstrating good robustness and adaptability to complex working conditions.

[0038] Comparison and verification of other models: The diagnostic model (DBMS-FNet) of the present invention is compared with the prior art VIbration Analysis ML and LogReg-Diagnosis, and the results are shown in Table 3 below:

[0039] The results show that the model of the present invention can effectively improve the ability to analyze complex fault modes through a dual-branch multi-scale fusion network architecture and a dynamic weighted fusion module, showing significant advantages. It innovatively combines the extraction and optimization of time-frequency features and multi-scale features, overcoming the problem of insufficient fault diagnosis accuracy of traditional methods under complex working conditions. The dynamic weighting mechanism is used to achieve adaptive optimization of different feature branches, effectively improving the robustness and applicability of the model. In addition, the model design of the present invention is scalable and can adapt to the fault diagnosis needs of other mechanical systems, providing a new solution for industrial intelligent maintenance.

[0040] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A rotating machinery fault diagnosis device based on a dual-branch multi-scale fusion network of time-frequency images, characterized in that: include: A data acquisition module, which acquires the vibration signal data to be detected of the rotating mechanical equipment; The data preprocessing module uses a sliding window piecewise function to divide the vibration signal data to be detected and performs normalization and standardization preprocessing; A data conversion module, which uses continuous wavelet transform to convert the pre-processed vibration signal data to be detected into a time-frequency image and corresponding time-frequency features; A feature extraction module, including a parallel time-frequency feature extractor and a multi-scale feature extractor; The time-frequency feature extractor analyzes and processes the time-frequency features to extract numerical branch features, and the multi-scale feature extractor analyzes and processes the time-frequency images to extract multi-scale visual branch features; The dynamic weighted fusion module performs weighted fusion of numerical branch features and multi-scale visual branch features through dynamically generated weights and outputs fault diagnosis results.

2. The rotating machinery fault diagnosis device based on the dual-branch multi-scale fusion network of time-frequency images according to claim 1 is characterized in that: The segment length of the sliding window is 1024 and the overlap length is 512.

3. The rotating machinery fault diagnosis device based on the dual-branch multi-scale fusion network of time-frequency images according to claim 1 is characterized in that: The continuous wavelet transform uses the Morlet mother wavelet with a scale range of 1-128.

4. The rotating machinery fault diagnosis device based on the dual-branch multi-scale fusion network of time-frequency images according to claim 1 is characterized in that: The time-frequency feature extractor includes several convolutional pooling modules; each convolutional pooling module includes an upsampling layer, a convolutional layer, a BatchNorm2d layer, a PReLU activation layer, a Dropout layer and a one-dimensional MaxPool layer; an attention module is introduced after each convolutional pooling module.

5. The rotating machinery fault diagnosis device based on a dual-branch multi-scale fusion network of time-frequency images according to claim 1 is characterized in that: The multi-scale feature extractor includes a plurality of multi-convolution kernel modules and a dimensionality reduction module; the multi-convolution kernel module includes a plurality of convolution kernels of different sizes, which are used to extract detail features, mid-scale spatial features, and global range features of the time-frequency image; The dimension reduction module includes a global average pooling layer and a fully connected layer, which performs dimension reduction processing on the output results of the multi-convolution kernel module.

6. The rotating machinery fault diagnosis device based on the dual-branch multi-scale fusion network of time-frequency images according to claim 5 is characterized in that: Each multi-convolution kernel module includes 1×1 convolution kernel, 3×3 convolution kernel and 5×5 convolution kernel. The 1×1 convolution kernel captures local detail features, the 3×3 convolution kernel extracts mid-scale spatial features, and the 5×5 convolution kernel obtains global feature representation.

7. The rotating machinery fault diagnosis device based on a dual-branch multi-scale fusion network of time-frequency images according to claim 1 is characterized in that: The dynamic weighted fusion module includes a feature splicing unit, a weight generation unit and a weighted fusion unit; the feature splicing unit splices the feature vectors output by the time-frequency feature extractor and the multi-scale feature extractor in the channel dimension to form a joint feature representation; the weight generation unit processes the joint features through a fully connected network, generates the weight of each branch, and normalizes the weight using the Softmax function; the weighted fusion unit performs weighted summation on the feature vectors of the two branches according to the generated weights to generate a fused feature representation. Output fault diagnosis results.

8. The rotating machinery fault diagnosis device based on a dual-branch multi-scale fusion network of time-frequency images according to claim 1 is characterized in that: The training methods of the feature extraction module and the dynamic weighted fusion module include: The vibration signals generated by rotating machinery under different working conditions are collected, the vibration signals are processed by sliding window segmentation, and the time-frequency images and corresponding time-frequency features are generated by continuous wavelet transform to construct a training data set; The feature extraction module and the dynamic weighted fusion module are trained using a training data set. During the training process, the time-frequency feature extractor and the multi-scale feature extractor extract features from the input time-frequency features and time-frequency images respectively. The dynamic weighted fusion module performs weighted fusion on the feature vectors of the two branches and outputs the predicted fault category label. The deviation between the current output and the expected target is calculated through forward propagation, and the weights and biases of the parameters of each layer are continuously optimized and adjusted through the error back propagation algorithm.

9. The rotating machinery fault diagnosis device based on a dual-branch multi-scale fusion network of time-frequency images according to claim 8 is characterized in that: The optimization process uses the cross entropy loss function as the objective function, updates the parameters through the Adam optimizer, sets the initial learning rate to 0.005, and iterates 50 rounds of training.

10. A method for fault diagnosis based on the rotating machinery fault diagnosis device according to any one of claims 1 to 9, characterized in that: include: The data acquisition module acquires the vibration signal data to be detected of the rotating mechanical equipment; The data preprocessing module uses a sliding window piecewise function to divide the vibration signal data to be detected, and performs normalization and standardization preprocessing; The data conversion module uses continuous wavelet transform to convert the pre-processed vibration signal data to be detected into time-frequency images and corresponding time-frequency features; The feature extraction module includes a parallel time-frequency feature extractor and a multi-scale feature extractor; the time-frequency feature extractor is used to analyze and process the time-frequency feature to extract the numerical branch feature, and the multi-scale feature extractor is used to analyze and process the time-frequency image to extract the multi-scale visual branch feature; The dynamic weighted fusion module performs weighted fusion of numerical branch features and multi-scale visual branch features through dynamically generated weights and outputs fault diagnosis results.

Images