Mechanical intelligent fault diagnosis method and device based on multi-mode sensing signal

Abstract

The invention discloses a mechanical intelligent fault diagnosis method and device based on a multi-modal sensing signal, and belongs to the field of mechanical equipment fault diagnosis, and the method comprises the steps: multi-modal data collection and preprocessing: collecting three-channel original signals of two vibration channels and one current channel of mechanical rotating equipment under normal and multiple fault working conditions, intercepting equal-length fragments to form a sample; constructing and training a convolutional neural network model, and performing deep feature learning on the multi-modal sample; and inputting a to-be-detected multi-modal signal into the trained model, performing fault diagnosis reasoning, and outputting a corresponding diagnosis result. According to the invention, multi-sensor information and a deep learning algorithm are fused, and high-accuracy fault detection and health state intelligent evaluation of the mechanical rotating part are realized.

Description

Technical Field

[0001] This invention relates to a mechanical intelligent fault diagnosis method and device based on multimodal sensing signals, belonging to the field of mechanical equipment fault diagnosis. Background Technology

[0002] Rotating machinery (such as motors, gearboxes, and bearings) is a critical component in industrial production processes, and its operating status directly affects production safety and efficiency. However, these rotating parts operate at high speeds for extended periods and are subjected to alternating loads, making them prone to various faults such as bearing wear, gear damage, and motor short circuits. Failure of critical components can lead to equipment downtime, production interruptions, and even safety accidents and economic losses. Therefore, timely and effective detection and diagnosis of faults in rotating machinery is of paramount importance for ensuring stable mechanical operation.

[0003] Traditional mechanical fault diagnosis methods often rely on single sensor signals and manual feature extraction. For example, they may use only bearing vibration signals or motor current signals, determine characteristic frequencies or statistical features through experience, and then use traditional machine learning methods such as Support Vector Machines (SVM) for classification and diagnosis. While single-mode signal methods can be effective in some situations, the complexity of industrial environments and the influence of various factors on the characteristics of a single signal can lead to decreased diagnostic accuracy. For instance, when fault signals are overwhelmed by noise or different faults exhibit similar behavior in a single signal, single-mode signal-based diagnosis may struggle to distinguish between them. Furthermore, methods like SVM require manual feature selection, making the feature extraction process cumbersome and highly dependent on expert experience, and hindering the ability to adapt to new fault modes in a timely manner.

[0004] In recent years, with the development of sensor and data acquisition technologies, multiple forms of sensor data can be collected simultaneously in industrial settings. Furthermore, the improvement in computing power has led to the widespread application of deep learning in fault diagnosis. Deep learning models such as Convolutional Neural Networks (CNNs) can automatically learn fault characteristics from raw signals, achieving end-to-end intelligent diagnosis with higher robustness and accuracy compared to traditional methods. However, most current research only applies deep learning models to signals of a single modality, failing to fully utilize complementary information from different sensor sources. Therefore, there is an urgent need for an intelligent fault diagnosis method that can fuse multiple sensor signals to improve the accuracy and generalization ability of fault detection in complex machinery.

[0005] Invention Patent: A bearing fault diagnosis method and system (Inventors: Zhu Wangchun, Zhou Canwen, Zhai Jun, Yang Dihuan, Chen Yuxia, Wang Jing; Applicant: Guilin University of Electronic Technology; Application (Patent) No.: CN202311363625.X) The aforementioned invention proposes a bearing fault diagnosis method and system based on a deep learning network model with embedded multi-layer small convolutional structures. However, the fault diagnosis mode proposed in the above invention only addresses a single bearing object, ignoring the complex spatiotemporal coupling relationships of multimodal data of key components within the overall system's transmission chain, and thus cannot perform real-time fault diagnosis for multiple key components. Summary of the Invention

[0006] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide a mechanical intelligent fault diagnosis method and device based on multimodal sensing signals, so as to realize comprehensive and reliable fault monitoring and diagnosis of mechanical rotating parts and improve the accuracy of detection.

[0007] The technical solution of this invention is: Firstly, a mechanical intelligent fault diagnosis method based on multimodal sensing signals is provided, including: Multimodal sensor signal data of rotating mechanical components under different operating conditions are collected. The collected multimodal sensor signals are cut into multiple data segments of the same length according to the same time length to form samples, and a multimodal sample dataset with health status labels is constructed. A convolutional neural network model is constructed and trained using a multimodal sample dataset labeled with health status. During model training, the parameters of the convolutional neural network are adjusted through forward and backward propagation so that the model can learn the feature patterns of each health status based on the multimodal input. The multimodal sensor signal data of the machine under test is processed by truncating it to equal time lengths and then input into a trained convolutional neural network model. The model is used to analyze and judge the input data and output the diagnostic results corresponding to the current working condition of the machine. The diagnostic results include multiple predefined health or fault states.

[0008] Preferably, the multimodal sensing signals come from the rotating parts of the machine, including dual-channel bidirectional vibration signals collected by vibration sensors installed on the bearing and gearbox parts, and single-phase current signals collected by current sensors installed on the motor power supply line, with each channel signal being collected synchronously.

[0009] Preferably, the convolutional neural network model comprises, in sequence: five convolutional pooling units, one adaptive pooling layer, and two fully connected layers; wherein: Each convolutional pooling unit consists of a convolutional block and a max-pooling layer connected in sequence.

[0010] Preferably, each convolutional block comprises, in sequence: a one-dimensional convolutional operation layer, a batch normalization layer, and a non-linear activation function layer; l The computational form of a layered convolutional block is as follows:

[0011] in Indicates the first Layer input feature map aisle, For the first Layer The weights of each convolution kernel, This represents the convolution operation. σ is the bias term, and σ(.) is the nonlinear activation function.

[0012] Preferred, the first Output of max pooling layer for:

[0013] in, This represents the set of indices within the current pooling window. Indicates the first The feature value with index m in the feature map output by each convolutional block.

[0014] Preferably, the adaptive pooling layer receives the deep features output by the last max pooling layer and maps them to a fixed-length feature vector. z :

[0015] in The deep features output by the last convolutional pooling unit. This means mapping the extracted deep features to a feature vector of fixed length; K It is a fixed dimension and is a positive integer.

[0016] Preferably, in the two fully connected layers: The first fully connected layer receives a fixed-length feature vector output by the adaptive pooling layer, performs linear transformation and nonlinear mapping on the feature vector to achieve further fusion and compression of high-level semantic features, and passes the processed feature representation as input to the second fully connected layer for subsequent classification and discrimination of multiple working conditions. The second fully connected layer is used to output the classification results of the multiple operating conditions; The outputs of the two fully connected layers are:

[0017] in, For activation function, This represents the probability distribution vector for multiple operating conditions. For the bias term of the first fully connected layer, This is the bias term for the second fully connected layer. This is the weight matrix of the first fully connected layer. This is the feature vector extracted from the aforementioned convolutional and pooling layers.

[0018] Preferably, the activation function of the nonlinear activation function layer is the Mish function.

[0019] Secondly, a mechanical intelligent fault diagnosis device based on multimodal sensing signals is provided, comprising: a data acquisition and preprocessing module, a dataset storage module, and an intelligent fault diagnosis module, wherein: The data acquisition and preprocessing module is used to acquire dual-channel bidirectional vibration signals and single-phase current signals of mechanical rotating parts. After acquisition, the data is cut into multiple data segments of the same length according to the same time length to form multimodal sample data. The data is labeled and output to the dataset storage module or intelligent fault diagnosis module. The dataset storage module stores labeled multimodal sample data; The intelligent fault diagnosis module stores the convolutional neural network model and uses multimodal sample data from the dataset storage module to train the convolutional neural network model. After training, it can output the diagnostic results of the current working condition of the machine based on the multimodal sample data output by the data acquisition and preprocessing module.

[0020] Preferably, the convolutional neural network model comprises, in sequence: five convolutional pooling units, one adaptive pooling layer, and two fully connected layers; wherein: Each convolutional pooling unit consists of a convolutional block and a max pooling layer connected in sequence; Each convolutional block consists of, in sequence, a one-dimensional convolutional operation layer, a batch normalization layer, and a non-linear activation function layer; the activation function of the non-linear activation function layer is the Mish function.

[0021] Compared with the prior art, the present invention has the following advantages: This invention utilizes multimodal sensor signal fusion and deep convolutional neural networks to achieve efficient fault diagnosis of rotating components in machinery. By simultaneously analyzing bearing vibration, gear vibration, and motor current signals, it overcomes the limitation of missing fault information from single signals, significantly improving the accuracy and reliability of fault detection. Employing a five-layer convolutional network structure including the Mish activation function allows for the extraction of richer fault features and accelerates model convergence. Experimental results show that compared to traditional support vector machine methods and convolutional neural network models utilizing only a single sensor signal, the method of this invention significantly improves fault classification accuracy, enabling more accurate identification of potential equipment faults and demonstrating high engineering application value. Attached Figure Description

[0022] Figure 1This is a flowchart of the fault diagnosis method of the present invention; Figure 2 A confusion matrix diagram of diagnostic classification results for a certain embodiment. Detailed Implementation

[0023] To address the problems existing in current technologies, this invention proposes a mechanical intelligent fault diagnosis method based on multimodal sensing signals. The method comprises three stages. The first stage is the multimodal data acquisition and preprocessing stage: acquiring raw signals from two vibration channels and one current channel of the rotating mechanical equipment under normal and various fault conditions, and extracting equal-length segments to form samples; the second stage is the construction and training of a convolutional neural network model, performing deep feature learning on the multimodal samples; the third stage involves inputting the multimodal signal to be tested into the trained model, performing fault diagnosis inference, and outputting the corresponding diagnostic results.

[0024] To achieve the objective of this invention, the present invention provides a mechanical intelligent fault diagnosis method based on multimodal sensing signals, comprising the following steps: Step 1: Multimodal data acquisition and preprocessing. Collect raw sensor signal data of mechanical rotating parts under different operating conditions. The sensor signals include two sets of vibration acceleration signals and one set of motor current signals. Cut the collected signals into multiple data segments of the same length according to the same time length to form samples and obtain a multimodal sample dataset with health status labels. Step 2: Model training. Construct a convolutional neural network model and perform deep feature extraction and training on the multimodal sample data obtained in Step 1. During model training, adjust the parameters of the convolutional neural network through forward and backward propagation so that the model can learn the feature patterns of each health state based on the multimodal input. Step 3: Fault diagnosis reasoning. The multimodal sensor signal data of the machine under test is preprocessed in the same way as in Step 1 and then input into the trained convolutional neural network model. The model is used to analyze and judge the input data and output the diagnosis results corresponding to the current working condition of the machine. The diagnosis results include multiple predefined health or fault states.

[0025] Furthermore, the multimodal sensing signals mentioned in step 1 come from the rotating parts of the machine (motor, gearbox, and shaft box), including dual-channel bidirectional vibration signals collected by vibration sensors installed on the bearing and gearbox parts, and single-phase current signals collected by current sensors installed on the motor power supply line. The signals of each channel are collected synchronously, and the sampling rate is 64kHz.

[0026] Further, in step 2, the convolutional neural network model includes, in sequence: five convolutional pooling units, one adaptive pooling layer, and two fully connected layers; wherein: each convolutional pooling unit includes, in sequence: a convolutional block and a max pooling layer; each convolutional block includes, in sequence: a one-dimensional convolutional operation layer, a batch normalization layer, and a non-linear activation function layer; specifically: The convolutional neural network model includes five convolutional blocks, each of which includes a one-dimensional convolution operation layer, a batch normalization layer, and a non-linear activation function layer. The mathematical formula for the convolution operation is:

[0027] in Indicates the first Layer input feature map aisle, For the first Layer The weights of each convolution kernel, This represents the convolution operation. σ is the bias term, and σ(.) is the nonlinear activation function.

[0028] The convolutional block is followed by a max-pooling layer, the first... Output of max pooling layer Its mathematical definition is:

[0029] in, This represents the set of indices within the current pooling window. Indicates the first The convolutional layer outputs the feature value with index m in the feature map.

[0030] And after the max pooling layer in the last convolutional block, an adaptive pooling layer is set to map the extracted deep features to a fixed-length feature vector:

[0031] in The deep features are output by the last max pooling layer. This means mapping the extracted deep features to a fixed-length feature vector. K It is a fixed dimension.

[0032] The convolutional neural network further includes two fully connected layers, wherein: The first fully connected layer receives a fixed-length feature vector output by the adaptive pooling layer, performs linear transformation and nonlinear mapping on the feature vector to achieve further fusion and compression of high-level semantic features, and passes the processed feature representation as input to the second fully connected layer for subsequent classification and discrimination of multiple working conditions. The second fully connected layer is used to output the classification results of the multiple operating conditions. The calculation formula for the output layer is as follows:

[0033] in, For activation function, This represents the probability distribution vector for multiple operating conditions. For the bias term of the first fully connected layer, This is the bias term for the second fully connected layer. This is the weight matrix of the first fully connected layer. This is the feature vector extracted from the aforementioned convolutional and pooling layers.

[0034] In step 2, the Mish function is used as the activation function for the convolutional neural network to replace the traditional ReLU activation function, thereby improving the non-linear expressive power of feature extraction. The mathematical definition of the Mish function is:

[0035] Compared to the ReLU function, the Mish function is smooth and differentiable near the zero point, which can improve the continuity of gradient propagation and the nonlinear expressive power of feature extraction in deep networks, thereby improving the resolution of fault features and classification accuracy.

[0036] Furthermore, the present invention provides a mechanical intelligent fault diagnosis device based on multimodal sensing signals, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method.

[0037] Furthermore, the present invention provides a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method.

[0038] Example: To make the technical solutions and objectives of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific implementation steps described herein are only used to better illustrate the application of the present invention, but the technical features involved in the embodiments of the present invention are not limited thereto.

[0039] Please see Figure 1 The mechanical intelligent fault diagnosis method based on multimodal sensor signals provided by this invention includes the following steps: Step 1: Multimodal data acquisition and preprocessing. Raw sensor signal data of mechanical rotating components under different operating conditions are collected. The sensor signals include a set of gearbox unidirectional vibration acceleration signals, a set of axle box unidirectional vibration acceleration signals, and a set of motor single-phase current signals. The collected signals are truncated into multiple data segments of the same length according to equal time length to form samples, and a multimodal sample dataset with health status labels is obtained. The multimodal sample dataset is divided into training set and test set according to a certain ratio for subsequent model training and testing.

[0040] In some embodiments of the present invention, to facilitate the subsequent construction of the convolutional neural network model, the acquired multimodal signal is equally divided into multiple samples of length 32000, and the multimodal sample dataset is divided into a training set in a 7:3 ratio. } and test set { }

[0041] Step 2: Model training. Construct a convolutional neural network model using the multimodal sample data set obtained in Step 1. Deep feature extraction and training are performed; during model training, the parameters of the convolutional neural network are adjusted through forward and backward propagation, enabling the model to learn feature patterns under various health states based on multimodal inputs.

[0042] In some embodiments of the present invention, taking into account the characteristic that the input data is a three-channel one-dimensional signal, the constructed convolutional neural network model includes multiple convolutional blocks and a final fully connected output layer.

[0043] In this embodiment, the convolutional neural network model is configured with 5 convolutional blocks and 2 fully connected layers stacked sequentially. The network structure parameters are shown in Table 1.

[0044] It should be noted that the network model is trained using a small learning rate in this step (the learning rate is set to 0.0001), which can avoid oscillations or non-convergence in the network training results.

[0045] Step 3: Fault diagnosis reasoning, using the multimodal sensor signal data test set of the machine under test { After the same preprocessing as in step 1, the data is input into the trained convolutional neural network model. The model is used to analyze and judge the input data and output the diagnostic results corresponding to the current working condition of the machine. The diagnostic results include multiple predefined health or fault states.

[0046] Table 1. Convolutional Neural Network Model Structure Parameters

[0047] The overall mechanical intelligent diagnostic process is as follows Figure 1 As shown.

[0048] Furthermore, the present invention provides a mechanical intelligent fault diagnosis device based on multimodal sensing signals, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method.

[0049] Furthermore, the present invention provides a method comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method.

[0050] The invention will be further described below with reference to the accompanying drawings and experimental examples.

[0051] To evaluate the performance of the proposed method, experimental verification was conducted using rotating mechanical data under different health conditions. A dataset of a bogie transmission system was selected, with key rotating mechanical components including the traction motor, drive gearbox, and axle box. A single power transmission chain on the experimental setup comprises a motor, reduction gearbox, and axle box. The transmission chain is driven by a three-phase asynchronous AC motor, with the motor speed controlled by a frequency converter. The load is applied via a hydraulic system. The reduction gearbox uses helical gears, with the drive gear having 16 teeth and the driven gear having 107 teeth; the axle box bearing is model HRB352213.

[0052] This dataset uses operating condition 1 (WC1) for testing, where the motor speed is 20Hz and the lateral load is 0kN. The dataset collects four types of signals: three-phase vibration, three-phase current, speed, and sound, totaling 24 channels, with each channel sampling at 64kHz. To closely reflect the actual needs of this invention, one-phase current data, one-phase gearbox vibration data, and one-phase shaft box vibration data are selected for corresponding modeling and analysis. The corresponding health status labels for each component are shown in Table 2. Based on actual needs, a subset of health status data is selected for training and testing of the fault diagnosis model: normal state, motor short circuit, tooth surface wear, missing teeth, broken teeth, bearing inner ring fault, bearing outer ring fault, bearing rolling element fault, and bearing cage fault, totaling nine health statuses. After model training and testing, the confusion matrix of the classification results is as follows: Figure 2 As shown, the classification accuracy reaches 98.15%, which is fully suitable for industrial applications.

[0053] The experimental results show that the deep convolutional neural network model trained on the training set data can achieve high-accuracy classification of the test set data, indicating that the method of the present invention can effectively distinguish normal samples of motors, gears, and bearings, as well as eight other types of fault samples, thus achieving the objective of the present invention.

[0054] Table 2 Basic Health Status Codes for Each Component

[0055] This invention addresses the shortcomings of existing technologies and, in conjunction with practical industrial contexts, utilizes multi-source information fusion and deep learning techniques to overcome the deficiencies of insufficient diagnostic information and low accuracy associated with single-signal methods. It achieves comprehensive and reliable fault monitoring and diagnosis of rotating mechanical components. This invention acquires raw signals from two vibration channels and one current channel of rotating mechanical equipment under normal and various fault conditions, extracting equal-length segments to form samples. A convolutional neural network model is constructed and trained to perform deep feature learning on the multimodal samples. The multimodal signals to be tested are input into the trained model for fault diagnosis inference, outputting corresponding diagnostic results. Ultimately, this achieves high-accuracy fault detection and intelligent health status assessment of rotating mechanical components. Furthermore, this invention creatively provides real-time health status monitoring at the system level for key rotating components such as motors, gearboxes, and axle boxes, filling a gap in related research.

[0056] This invention integrates multimodal sensor signals such as motor current signals, bearing vibration signals, and gear vibration signals, and constructs a deep convolutional neural network model to achieve accurate diagnosis and health status assessment of mechanical rotating components. It has advantages such as high diagnostic accuracy, good real-time performance, and adaptability to complex working conditions. Furthermore, it innovatively enables real-time health status monitoring of key rotating components such as motors, gearboxes, and axle boxes at the system level, filling a gap in related research fields and contributing to the real-time maintenance of key rotating components in industrial field systems.

[0057] It should be noted that although the implementation of the present invention has been described in detail with reference to examples, those skilled in the art will readily understand that any modifications, substitutions, and improvements made without departing from the spirit and principles of the invention as set forth in the appended claims should be included within the scope of protection of the present invention.

[0058] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A mechanical intelligent fault diagnosis method based on multimodal sensing signals, characterized in that... include: Multimodal sensor signal data of rotating mechanical components under different operating conditions are collected. The collected multimodal sensor signals are cut into multiple data segments of the same length according to the same time length to form samples, and a multimodal sample dataset with health status labels is constructed. A convolutional neural network model is constructed and trained using a multimodal sample dataset labeled with health status. During model training, the parameters of the convolutional neural network are adjusted through forward and backward propagation so that the model can learn the feature patterns of each health status based on the multimodal input. The multimodal sensor signal data of the machine under test is processed by truncating it to equal time lengths and then input into a trained convolutional neural network model. The model is used to analyze and judge the input data and output the diagnostic results corresponding to the current working condition of the machine. The diagnostic results include multiple predefined health or fault states.

2. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 1, characterized in that: The multimodal sensing signals come from the rotating parts of the machine, including dual-channel bidirectional vibration signals collected by vibration sensors installed on bearings and gearboxes, and single-phase current signals collected by current sensors installed on the motor power supply line. The signals from each channel are collected synchronously.

3. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 1, characterized in that: The convolutional neural network model consists of, in sequence: five convolutional pooling units, one adaptive pooling layer, and two fully connected layers; wherein: Each convolutional pooling unit consists of a convolutional block and a max-pooling layer connected in sequence.

4. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 3, characterized in that: Each convolutional block consists of, in sequence: a one-dimensional convolutional operation layer, a batch normalization layer, and a non-linear activation function layer; l The computational form of a layered convolutional block is as follows: in Indicates the first Layer input feature map aisle, For the first Layer The weights of each convolution kernel, This represents the convolution operation. σ is the bias term, and σ(.) is the nonlinear activation function.

5. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 3, characterized in that: No. Output of max pooling layer for: in, This represents the set of indices within the current pooling window. Indicates the first The feature value with index m in the feature map output by each convolutional block.

6. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 3, characterized in that: The adaptive pooling layer receives the deep features output from the last max pooling layer and maps them to a fixed-length feature vector. z : in The deep features output by the last convolutional pooling unit. This means mapping the extracted deep features to a feature vector of fixed length; K It is a fixed dimension and is a positive integer.

7. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 3, characterized in that: In two fully connected layers: The first fully connected layer receives a fixed-length feature vector output by the adaptive pooling layer, performs linear transformation and nonlinear mapping on the feature vector to achieve further fusion and compression of high-level semantic features, and passes the processed feature representation as input to the second fully connected layer for subsequent classification and discrimination of multiple working conditions. The second fully connected layer is used to output the classification results of the multiple operating conditions; The outputs of the two fully connected layers are: in, For activation function, This represents the probability distribution vector for multiple operating conditions. For the bias term of the first fully connected layer, This is the bias term for the second fully connected layer. This is the weight matrix of the first fully connected layer. This is the feature vector extracted from the aforementioned convolutional and pooling layers.

8. The mechanical intelligent fault diagnosis method based on multimodal sensing signals according to claim 4, characterized in that: The activation function for the nonlinear activation function layer is the Mish function.

9. A mechanical intelligent fault diagnosis device based on multimodal sensing signals, characterized in that... include: The module includes a data acquisition and preprocessing module, a dataset storage module, and an intelligent fault diagnosis module, among which: The data acquisition and preprocessing module is used to acquire dual-channel bidirectional vibration signals and single-phase current signals of mechanical rotating parts. After acquisition, the data is cut into multiple data segments of the same length according to the same time length to form multimodal sample data. The data is labeled and output to the dataset storage module or intelligent fault diagnosis module. The dataset storage module stores labeled multimodal sample data; The intelligent fault diagnosis module stores the convolutional neural network model and uses multimodal sample data from the dataset storage module to train the convolutional neural network model. After training, it can output the diagnostic results of the current working condition of the machine based on the multimodal sample data output by the data acquisition and preprocessing module.

10. A mechanical intelligent fault diagnosis device based on multimodal sensing signals according to claim 9, characterized in that: The convolutional neural network model consists of, in sequence: five convolutional pooling units, one adaptive pooling layer, and two fully connected layers; wherein: Each convolutional pooling unit consists of a convolutional block and a max pooling layer connected in sequence; Each convolutional block consists of, in sequence, a one-dimensional convolutional operation layer, a batch normalization layer, and a non-linear activation function layer; the activation function of the non-linear activation function layer is the Mish function.

Images