Fault diagnosis method for multiple parts of rolling mill

Abstract

The invention discloses a fault diagnosis method for multiple parts of a rolling mill, and relates to the technical field of fault diagnosis. Acquiring state data of a plurality of parts of the rolling mill; converting the state data into an RGB image through a Grubby angle difference field; converting the RGB image into three images of R, G and B channels, and converting the image of each channel into a grey-scale map; performing dimension reduction processing on each grey-scale map, and obtaining an RGB image after dimension reduction according to the three grey-scale maps after dimension reduction processing; inputting the RGB image after dimension reduction into an improved ShuffleNetV2 network model to obtain a fault diagnosis result of the rolling mill; and the improved ShuffleNetV2 network model is obtained by adding a double attention mechanism into a basic ShuffleNetV2 network for training. According to the method, when fault diagnosis is carried out on the multiple parts of the rolling mill, the fault diagnosis efficiency can be improved while the fault diagnosis accuracy is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and particularly relates to a fault diagnosis method for multiple components of a rolling mill. Background Art

[0002] At present, a rolling mill is the core equipment for metal rolling production such as steel. As key transmission components, bearings and gears, once a fault occurs, it will cause the rolling mill to stop. Timely and accurate diagnosis and classification of faults can give early warnings and arrange maintenance, avoid unplanned shutdowns, maintain the smooth operation of the production line, and reduce the economic losses caused by production interruptions. For example, a large steel plant may lose several tons of steel production and corresponding processing profits due to a one-hour shutdown of the rolling mill.

[0003] The rolling mill operates under harsh working conditions such as high temperature, heavy load, and strong vibration. Bearings and gears are subjected to alternating loads, impact loads, and the influence of high temperature, and are prone to various fault forms such as wear, fatigue cracks, and fractures. The complex working conditions make the fault causes diverse and interrelated. Professional diagnosis and classification must be carried out to accurately find out the root causes of faults and take effective repair measures.

[0004] However, in the prior art, there is a problem of low efficiency in the method of fault diagnosis for multiple components of a rolling mill. Summary of the Invention

[0005] Based on this, it is necessary to provide a fault diagnosis method for multiple components of a rolling mill in view of the above technical problems. When diagnosing faults for multiple components of a rolling mill, this method can improve the fault diagnosis efficiency while ensuring the accuracy of fault diagnosis.

[0006] The present invention adopts the following technical solutions: The present invention provides a fault diagnosis method for multiple components of a rolling mill, including: Obtaining the state data of multiple components of the rolling mill; Converting the state data into an RGB image through the Gramian angular difference field; Converting the RGB image into three images of R, G, and B channels, and converting the image of each channel into a grayscale image; Performing dimensionality reduction processing on each grayscale image respectively, and obtaining a reduced-dimension RGB image according to the three reduced-dimension grayscale images; Inputting the reduced-dimension RGB image into an improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill; the improved ShuffleNetV2 network model is obtained by adding a dual attention mechanism to the basic ShuffleNetV2 network for training.

[0007] Optionally, performing dimensionality reduction processing on each grayscale image respectively includes: For any grayscale image, obtain the mean value of all data points in the grayscale image; According to the values of each data point and the mean value, determine the divergence matrix, and solve the eigenvalues and eigenvectors of the divergence matrix; The first K eigenvectors corresponding to the largest eigenvalues are determined as the new eigenvectors; According to the new eigenvectors and the original grayscale image, determine the grayscale image after dimensionality reduction.

[0008] Optionally, the improved ShuffleNetV2 network model includes an input module integrating an attention mechanism, multiple lightweight modules connected in series in sequence, and an output module integrating an attention mechanism; input the RGB image after dimensionality reduction into the improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill, including: Input the RGB image after dimensionality reduction into the improved ShuffleNetV2 network model, and extract the features in the RGB image after dimensionality reduction through the input module; Perform convolution operations on the features in the RGB image after dimensionality reduction through multiple lightweight modules connected in series in sequence to obtain convolution features; Perform feature prediction on the convolution features through the output module to obtain the fault diagnosis result of the rolling mill.

[0009] Optionally, the input module includes an input layer, a convolutional layer, an activation function, an attention mechanism layer, and a fully connected layer connected in series in sequence; extracting the features in the RGB image after dimensionality reduction through the input module includes: Receive the RGB image after dimensionality reduction through the input layer, and perform convolution operations on the RGB image after dimensionality reduction through the convolutional layer to extract local features; Increase the non-linear characteristics of the local features through the activation function to obtain non-linear features; Assign weights to different features in the non-linear features through the attention mechanism layer; Perform fully connected operations on the features output by the attention mechanism layer through the fully connected layer to obtain the features in the RGB image after dimensionality reduction.

[0010] Optionally, the attention mechanism layer includes a channel domain attention module and a spatial domain attention module; the channel domain attention module and the spatial domain attention module are designed in cascade.

[0011] Optionally, the output module includes a convolutional layer, an attention mechanism layer, and a fully connected layer connected in series in sequence; performing feature prediction on the convolution features through the output module to obtain the fault diagnosis result of the rolling mill includes: Perform convolution processing on the convolution features through the convolutional layer, and assign weights to the features after convolution through the attention mechanism layer; Map the features output by the attention mechanism layer to each fault category through a fully connected layer to obtain the fault diagnosis result of the rolling mill.

[0012] Optionally, the fault diagnosis result of the rolling mill includes the probability of each component of the rolling mill having each fault category; the method further includes: If the probability is greater than a preset fault threshold, determine the fault category corresponding to the probability as the fault category that the rolling mill has.

[0013] Optionally, the training process of the improved ShuffleNetV2 network model includes: Introduce a dual attention mechanism into the basic ShuffleNetV2 network to obtain an initial ShuffleNetV2 network; Obtain a training set, and pre-train the initial ShuffleNetV2 network on the training set through transfer learning to obtain an initial ShuffleNetV2 network model; the weight parameters in the initial ShuffleNetV2 network model are weight parameters with general feature extraction capabilities; Iteratively train the initial ShuffleNetV2 network model through the training set to obtain an improved ShuffleNetV2 network model.

[0014] The present invention provides a fault diagnosis device for multiple components of a rolling mill, including: An acquisition module for acquiring the state data of multiple components of the rolling mill; A conversion module for converting the state data into an RGB image through the Gram angular difference field, converting the RGB image into three images of the R, G, and B channels, and converting each channel image into a grayscale image; A dimensionality reduction module for respectively performing dimensionality reduction processing on each grayscale image, and obtaining a dimensionality-reduced RGB image according to the three dimensionality-reduced grayscale images; A classification module for inputting the dimensionality-reduced RGB image into the improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill; the improved ShuffleNetV2 network model is obtained by adding a dual attention mechanism to the basic ShuffleNetV2 network for training.

[0015] The present invention provides a computer-readable storage medium, the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned fault diagnosis method for multiple components of a rolling mill is implemented.

[0016] The present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above-mentioned fault diagnosis method for multiple components of a rolling mill is implemented.

[0017] The above at least one technical solution adopted by the present invention can achieve the following beneficial effects: In the present invention, first, the state data of multiple components of the rolling mill is converted into an RGB image through the Gram angular difference field. This conversion method can map one-dimensional state data into a two-dimensional space, retaining the time series information and features in the data, providing richer information for subsequent image processing and feature extraction, and helping to improve the accuracy of fault diagnosis. Moreover, when reducing the dimension of the RGB image, the image is converted into a three-channel image, and the dimensionality reduction process is performed on each grayscale image separately. In this way, while retaining the main features of the image, the data volume is reduced, the computational complexity is lowered, and thus the efficiency of fault diagnosis is improved. Further, ShuffleNetV2 itself is a lightweight network structure with high computational performance and good feature extraction ability. By adding a dual attention mechanism to the basic network, the model can automatically focus on key feature regions and channels when processing images, ignoring irrelevant information. Through the dual attention mechanism, the model can more accurately capture the features related to rolling mill faults in the image, enhance the feature representation ability, and thus improve the accuracy of fault diagnosis. At the same time, since the model can process data more targeted, reducing the calculation of irrelevant information, it helps to improve the efficiency of fault diagnosis. Description of the Drawings

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings: Figure 1 It is a schematic flow chart of a fault diagnosis method for multiple components of a rolling mill provided by the present invention; Figure 2 It is a schematic structural diagram of an improved ShuffleNetV2 network model; Figure 3 It is a schematic structural diagram of a lightweight module in the improved ShuffleNetV2 network model; Figure 4 It is a flow chart of a fault diagnosis method for multiple components of a rolling mill using the dimensionality reduction method (RPCA) and dual CBAM-ShuffleNetV2 provided by the present invention; Figure 5The RGB images after dimensionality reduction corresponding to a sampled bearing public dataset provided by the present invention. Among them, (a) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label H in the sampled bearing public dataset, (b) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label B7 in the sampled bearing public dataset, (c) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label IR7 in the sampled bearing public dataset, (d) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label in the sampled bearing public dataset, (e) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label B14 in the sampled bearing public dataset, (f) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label IR14 in the sampled bearing public dataset, (g) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label in the sampled bearing public dataset, (h) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label B21 in the sampled bearing public dataset, (i) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label IR21 in the sampled bearing public dataset, (j) is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label in the sampled bearing public dataset; Figure 6 The RGB images after dimensionality reduction corresponding to a mill bearing fault dataset provided by the present invention. Among them, (a) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label H in the mill bearing fault dataset, (b) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label B1 in the mill bearing fault dataset, (c) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label IR1 in the mill bearing fault dataset, (d) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label in the mill bearing fault dataset, (e) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label B2 in the mill bearing fault dataset, (f) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label IR2 in the mill bearing fault dataset, (g) is the RGB image after dimensionality reduction corresponding to the bearing data with the fault label in the mill bearing fault dataset; Figure 7 A data processing flow chart provided by the present invention; Figure 8Schematic diagram of the accuracy rate and loss value of a public bearing dataset provided by the present invention varying with the number of iterations at different rotational speeds. Among them, (a) shows the curve of the accuracy rate of fault diagnosis of the public bearing dataset varying with the number of iterations at rotational speeds of 1730, 1750, 1772, and 1792; (b) shows the curve of the loss value of the model of the public bearing dataset varying with the number of iterations at rotational speeds of 1730, 1750, 1772, and 1792. Figure 9 Classification confusion matrix of a public bearing dataset provided by the present invention at 4 rotational speeds. Among them, (a) shows the classification confusion matrix diagram of the public bearing dataset at a rotational speed of 1730; (b) shows the classification confusion matrix diagram of the public bearing dataset at a rotational speed of 1750; (c) shows the classification confusion matrix diagram of the public bearing dataset at a rotational speed of 1772; (d) shows the classification confusion matrix diagram of the public bearing dataset at a rotational speed of 1792. Figure 10 Schematic diagram of the accuracy rate and loss value of a rolling mill bearing fault dataset provided by the present invention at 3 rotational speeds. Among them, (a) shows the schematic diagram of the accuracy rate and loss value of the rolling mill bearing fault dataset at a rotational speed of 600; (b) shows the schematic diagram of the accuracy rate and loss value of the rolling mill bearing fault dataset at a rotational speed of 1200; (c) shows the schematic diagram of the accuracy rate and loss value of the rolling mill bearing fault dataset at a rotational speed of 1800. Figure 11 Confusion matrix of a rolling mill bearing fault dataset provided by the present invention at 3 rotational speeds. Among them, (a) shows the classification confusion matrix of the rolling mill bearing fault dataset at a rotational speed of 600; (b) shows the classification confusion matrix of the rolling mill bearing fault dataset at a rotational speed of 1200; (c) shows the classification confusion matrix of the rolling mill bearing fault dataset at a rotational speed of 1800. Figure 12 Cluster result diagram of a rolling mill bearing fault dataset provided by the present invention at 3 rotational speeds. Among them, (a) shows the data sample diagram of the rolling mill bearing fault dataset at a rotational speed of 600; (b) shows the cluster diagram of the rolling mill bearing fault dataset at a rotational speed of 600; (c) shows the data sample diagram of the rolling mill bearing fault dataset at a rotational speed of 1200; (d) shows the cluster diagram of the rolling mill bearing fault dataset at a rotational speed of 1200; (e) shows the data sample diagram of the rolling mill bearing fault dataset at a rotational speed of 1800; (f) shows the cluster diagram of the rolling mill bearing fault dataset at a rotational speed of 1800. Figure 13A reduced-dimension RGB image corresponding to a sampled public gear dataset provided by the present invention. Among them, figure (a) is the reduced-dimension RGB image corresponding to the gear data with a fault label of healthy in the sampled public gear dataset, figure (b) is the reduced-dimension RGB image corresponding to the gear data with a fault label of gear defect in the sampled public gear dataset, figure (c) is the reduced-dimension RGB image corresponding to the gear data with a fault label of tooth root wear in the sampled public gear dataset, figure (d) is the reduced-dimension RGB image corresponding to the gear data with a fault label of tooth surface wear in the sampled public gear dataset, and figure (e) is the reduced-dimension RGB image corresponding to the gear data with a fault label of broken tooth in the sampled public gear dataset; Figure 14 A reduced-dimension RGB image corresponding to a rolling mill gear fault dataset provided by the present invention. Among them, figure (a) is the reduced-dimension RGB image corresponding to the gear data with a fault label of healthy in the rolling mill gear fault dataset, figure (b) is the reduced-dimension RGB image corresponding to the gear data with a fault label of tooth root wear in the rolling mill gear fault dataset, figure (c) is the reduced-dimension RGB image corresponding to the gear data with a fault label of tooth surface wear in the rolling mill gear fault dataset, and figure (d) is the reduced-dimension RGB image corresponding to the gear data with a fault label of broken tooth in the rolling mill gear fault dataset; Figure 15 A curve showing the variation of the accuracy and loss value with the number of iterations of a sampled public gear dataset provided by the present invention under different vibration conditions. Among them, figure (a) is the curve showing the variation of the accuracy of fault diagnosis with the number of iterations of the sampled public gear dataset at vibration frequencies of 20 and 30, and figure (b) is the curve showing the variation of the loss value of the model with the number of iterations of the sampled public gear dataset at vibration frequencies of 20 and 30; Figure 16 A confusion matrix of a sampled public gear dataset provided by the present invention under 2 vibration conditions. Among them, figure (a) is the confusion matrix of the sampled public gear dataset at a vibration frequency of 20, and figure (b) is the confusion matrix of the sampled public gear dataset at a vibration frequency of 30; Figure 17 A curve showing the variation of the accuracy and loss value with the number of iterations of a rolling mill gear fault dataset provided by the present invention at 3 rotational speeds. Among them, figure (a) is the curve showing the variation of the accuracy and loss value with the number of iterations of the rolling mill gear fault dataset at a rotational speed of 600, figure (b) is the curve showing the variation of the accuracy and loss value with the number of iterations of the rolling mill gear fault dataset at a rotational speed of 1200, and figure (c) is the curve showing the variation of the accuracy and loss value with the number of iterations of the rolling mill gear fault dataset at a rotational speed of 1800; Figure 18A confusion matrix of a mill gear fault dataset provided by the present invention at three rotational speeds. Among them, (a) is the classification confusion matrix of the mill gear fault dataset at a rotational speed of 600, (b) is the classification confusion matrix of the mill gear fault dataset at a rotational speed of 1200, and (c) is the classification confusion matrix of the mill gear fault dataset at a rotational speed of 1800; Figure 19 A clustering result diagram of a mill gear fault dataset provided by the present invention at three rotational speeds. Among them, (a) is the data sample diagram of the mill gear fault dataset at a rotational speed of 600, (b) is the clustering diagram of the mill gear fault dataset at a rotational speed of 600, (c) is the data sample diagram of the mill gear fault dataset at a rotational speed of 1200, (d) is the clustering diagram of the mill gear fault dataset at a rotational speed of 1200, (e) is the data sample diagram of the mill gear fault dataset at a rotational speed of 1800, and (f) is the clustering diagram of the mill gear fault dataset at a rotational speed of 1800; Figure 20 A schematic diagram of a computer device for implementing a fault diagnosis method for multiple components of a rolling mill provided by the present invention. Detailed implementation manners

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0020] The present invention provides a fault diagnosis method for multiple components of a rolling mill, which can simultaneously realize the fault diagnosis of multiple components of the rolling mill.

[0021] The following will detail the technical solutions provided by each embodiment of the present invention in conjunction with the drawings.

[0022] Figure 1 A schematic diagram of the process of a fault diagnosis method for multiple components of a rolling mill in the present invention specifically includes the following steps: S101, obtain the state data of multiple components of the rolling mill.

[0023] Among them, multiple components of the rolling mill may include the bearings and gears of the rolling mill. The vibration signals of the bearings and gears of the rolling mill are mainly collected through sensors, and the vibration signals are used as state data.

[0024] S102, convert the state data into an RGB image through the Gramian angular difference field.

[0025] The Gramian Angular Difference Field (GADF) is a method for converting one-dimensional time series data (such as vibration signals) into two-dimensional images. It is based on trigonometric functions (cosine and sine) to encode the values in the time series and the relationships between them. The core idea is to map the values of the time series to the polar coordinate space and then construct an image by calculating the Gram matrix.

[0026] Specifically, calculate the maximum value in the vibration signal, then map each value in the vibration signal to the interval [0,1], and then convert it to an angle in polar coordinates (using the inverse cosine function); construct the Gram difference field matrix by calculating the cosine values of the differences between the angles. To obtain an RGB image, three-channel matrices need to be constructed. Here, simply copy the Gram difference field matrix three times and use them as the values of the R, G, and B channels respectively, and then combine them into an RGB image represented by a three-dimensional array.

[0027] S103, convert the RGB image into three images of the R, G, and B channels, and convert each channel image into a grayscale image; perform dimensionality reduction processing on each grayscale image respectively, and obtain the RGB image after dimensionality reduction according to the three grayscale images after dimensionality reduction processing.

[0028] The RGB image can be converted into three images of the R, G, and B channels using a preset conversion function, and each channel image is converted into a grayscale image.

[0029] Optionally, perform dimensionality reduction processing on each grayscale image respectively, including: for any grayscale image, obtain the mean value of all data points in the grayscale image; determine the divergence matrix according to the values of each data point and the mean value, and solve the eigenvalues and eigenvectors of the divergence matrix; determine the eigenvectors corresponding to the top K eigenvalues as the new eigenvectors; determine the grayscale image after dimensionality reduction according to the new eigenvectors and the original grayscale image.

[0030] Specifically, regard each grayscale image as an image matrix X, and the expression is: .

[0031] Among them, ; represents the column vector of the image matrix; represents the number of column vectors of the image matrix.

[0032] Obtain the mean value of the image matrix, the divergence matrix and the covariance matrix , as follows: (1); (2); (3).

[0033] Solving the eigenvalues and eigenvectors of the covariance matrix: It can be found that the divergence matrix is the covariance matrix multiplied by (total data volume - 1). Therefore, their eigenvalues and eigenvectors are the same. So, it is only necessary to solve the eigenvalues and eigenvectors of the divergence matrix. The characteristic equation is used to obtain the eigenvalues and the corresponding eigenvectors , where represents the number of eigenvectors, and the eigenvectors of the covariance matrix are represented as .

[0034] Arrange the eigenvectors of the covariance matrix in descending order of eigenvalues, and take the largest eigenvectors as row vectors to form a new eigenmatrix .

[0035] Convert the image data into the dimensional space constructed by the new eigenmatrix. That is, the principal component matrix of the new training image is , and the grayscale image after dimensionality reduction is obtained.

[0036] Recombine the three grayscale images after dimensionality reduction into a new image according to the original disassembly method. This image is the RGB image after dimensionality reduction.

[0037] The above dimensionality reduction method is a new two-dimensional picture data dimensionality reduction method. It reduces the dimensionality of RGB pictures, enhances the features of the data, and is more suitable for the recognition of convolutional neural networks. Especially in the application of lightweight networks, because lightweight networks process large data itself relatively slowly, which does not conform to the characteristics of lightweight networks such as few parameters, small computational amount, and short inference time. However, the dimensionality reduction method proposed in the present invention can greatly reduce the size of the data itself, making it more suitable for the application of lightweight networks.

[0038] S104, input the RGB image after dimensionality reduction into the improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill; the improved ShuffleNetV2 network model is obtained by adding a dual attention mechanism to the basic ShuffleNetV2 network for training.

[0039] As Figure 2 shown, Figure 2Schematic diagram of the improved ShuffleNetV2 network model; specifically, the improved ShuffleNetV2 network model includes an input module integrating an attention mechanism, a plurality of lightweight modules connected in series in sequence, and an output module integrating an attention mechanism; the dimension-reduced RGB image is input into the improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill, including: inputting the dimension-reduced RGB image into the improved ShuffleNetV2 network model, and extracting the features in the dimension-reduced RGB image through the input module; performing convolution operations on the features in the dimension-reduced RGB image sequentially through a plurality of lightweight modules connected in series in sequence to obtain convolution features; performing feature prediction on the convolution features through the output module to obtain the fault diagnosis result of the rolling mill.

[0040] It should be noted that Figure 2 The improved ShuffleNetV2 network model in [reference] is schematically shown with six modules. Among them, the first module includes an input module, and the second module, the third module, the fourth module, and the fifth module each include a lightweight module, and the sixth module includes an output module. Figure 2 Taking the lightweight module including 4 as an example, in practical applications, the number of lightweight modules in the improved ShuffleNetV2 network model is not limited.

[0041] Optionally, the input module includes an input layer, a convolutional layer, an activation function, an attention mechanism layer, and a fully connected layer connected in series in sequence; extracting the features in the dimension-reduced RGB image through the input module includes: receiving the dimension-reduced RGB image through the input layer, and performing a convolution operation on the dimension-reduced RGB image through the convolutional layer to extract local features; increasing the non-linear characteristics of the local features through the activation function to obtain non-linear features; assigning weights to different features in the non-linear features through the attention mechanism layer; performing a fully connected operation on the features output by the attention mechanism layer through the fully connected layer to obtain the features in the dimension-reduced RGB image.

[0042] Optionally, the attention mechanism layer includes a channel domain attention module and a spatial domain attention module; the channel domain attention module and the spatial domain attention module are designed in a cascaded manner. Among them, the channel domain attention module and the spatial domain attention module are Convolutional Block Attention Module (CBAM).

[0043] Optionally, the output module includes a convolutional layer, an attention mechanism layer, and a fully connected layer connected in series in sequence; the convolutional features are subjected to feature prediction through the output module to obtain the fault diagnosis result of the rolling mill, including: performing convolutional processing on the convolutional features through the convolutional layer, and assigning weights to the convolutional features through the attention mechanism layer; mapping the features output by the attention mechanism layer to each fault category through the fully connected layer to obtain the fault diagnosis result of the rolling mill.

[0044] Optionally, the fault diagnosis result of the rolling mill includes the probabilities of each component of the rolling mill having each fault category; this embodiment includes: if the probability is greater than a preset fault threshold, determining the fault category corresponding to the probability as the fault category that the rolling mill has a fault.

[0045] Optionally, the training process of the improved ShuffleNetV2 network model includes: introducing a dual attention mechanism into the basic ShuffleNetV2 network to obtain an initial ShuffleNetV2 network; obtaining a training set, and pre-training the initial ShuffleNetV2 network on the training set through transfer learning to obtain an initial ShuffleNetV2 network model; the weight parameters in the initial ShuffleNetV2 network model are weight parameters with general feature extraction capabilities; iteratively training the initial ShuffleNetV2 network model through the training set to obtain an improved ShuffleNetV2 network model.

[0046] Transfer learning can be simply defined as: by pre-training the initial ShuffleNetV2 network on the large-scale ImageNet training set, weight parameters with general feature extraction capabilities can be obtained. This model has already possessed a multi-level feature representation system through learning a large amount of data. For a small sample data set, it can still have a high recognition accuracy. Since the weight parameters have strong feature transfer capabilities, they can be directly introduced to extract features from other similar data sets; while improving the development efficiency of the network model, it can also improve the generalization performance and convergence efficiency of the model. Especially for the rolling mill simulation platform, the number of feature samples in the data of the bearings and gears taken is small, and it is necessary to make the model converge quickly to achieve fault diagnosis and classification of the rolling mill bearings and gears. Using the transfer learning strategy of similar data migration to make the model converge quickly and then improve the recognition accuracy of the model is an excellent way.

[0047] The design goal of CBAM is to solve the problem of insufficient representation ability of traditional convolutional neural networks in the process of feature extraction for multi-scale, morphological transformation, and direction-sensitive information.

[0048] CBAM effectively improves the performance bottlenecks of traditional convolutional neural networks in aspects such as the adaptability of multi-scale geometric features, the generalization of topological structures, and the robustness of direction perception through a dual-modal attention collaboration mechanism. Its technical implementation includes two core components: the channel-domain attention module realizes dynamic recalibration of feature channel weights and optimizes the cross-channel information integration efficiency; the spatial-domain attention module realizes adaptive allocation of spatial weights through learnable convolutional kernels and strengthens the representation intensity of local key regions. The two modules adopt a cascaded architecture design, which can be flexibly integrated into the feature extraction nodes at all levels of the deep network, and realize the adaptive enhancement of multi-level heterogeneous features through a gradual feature rectification mechanism, achieving refined modeling of complex spatial structures. Such a design significantly improves the discriminative expression ability of the network for heterogeneous features, meeting the optimization requirements of the mechanical field for model robustness and generalization performance.

[0049] The input layer is responsible for receiving the original image data. The convolutional layer performs convolution operations on the input data through convolutional kernels to extract local features. Subsequently, an activation function (such as ReLU) is introduced to increase non-linearity, thereby enhancing the expression ability of the model. The fully connected layer reduces the size of the feature map through downsampling operations, effectively reducing the computational complexity and improving the generalization ability of the model. It is achieved by selecting the maximum or average value within the pooling window. This helps to extract the most important features. The lightweight module usually consists of a stack of multiple convolutional and fully connected layers to gradually extract higher-level features. Deeper features can represent more complex patterns. Finally, the fully connected layer converts the extracted feature map into the final output of the network. Among them, the fully connected layer and the convolutional layer have great advantages in data feature extraction capabilities and are more suitable for processing two-dimensional images.

[0050] The improved ShuffleNetV2 network model adds a CBAM attention mechanism module to the first layer of convolution and the last layer of convolution. The improved network can better utilize its different depth-level structures to extract features of different dimensions from the input two-dimensional image data.

[0051] In a specific embodiment, as Figure 3 shown, Figure 3 FIG. is a schematic structural diagram of the lightweight module in the improved ShuffleNetV2 network model, specifically including: starting from the input of the lightweight module, there are two branches. One branch includes a convolutional layer, a batch normalization layer, an activation function, a depth convolutional layer, a batch normalization layer, a convolutional layer, a batch normalization layer, and an activation function connected in series in sequence. The other branch includes a depth convolutional layer, a batch normalization layer, a convolutional layer, a batch normalization layer, and an activation function connected in series in sequence. The outputs of the last activation functions of the two branches are both connected to the fully connected layer. The output of the fully connected layer is the input of the anonymous function, and the output of the anonymous function is the output of the lightweight module.

[0052] The improved ShuffleNetV2 network model in the present invention is a dual (Double) CBAM-ShuffleNetV2 network model. As Figure 4 shown, Figure 4 This is the flowchart of the fault diagnosis method for multiple components of the rolling mill using the dimensionality reduction method (RPCA) and the dual CBAM-ShuffleNetV2 provided by the present invention. Specifically, the main steps are as follows: (1) Data acquisition: Select the bearing public dataset, the gear public dataset, and obtain bearing vibration data on the roll bearing housing of the rolling mill test platform in the laboratory, and obtain gear vibration data on the outer connecting shaft of the driven wheel of the reduction box. 20 seconds of stable data is selected as the original dataset of the rolling mill bearings and gears.

[0053] (2) Data processing: Use the Gram angular difference field to convert the selected public datasets and the original datasets of bearings and gears obtained from the rolling mill platform into two-dimensional RGB image data. Then use RPCA to reduce the dimensionality of the RGB image data and highlight the features, and give the data fault labels for different faults and divide them into training and test datasets.

[0054] (3) Network training: Input the processed image data into the dual CABM-ShuffleNetV2 network. After data augmentation, then transfer the pre-trained module on the ImageNet dataset to the model training module, change the number of classification nodes in the last layer to the number of classification nodes in the input training set to adapt the network to the classification of the dataset. Finally, the weights of the highest accuracy rate in the model training will be saved and returned after each training.

[0055] (4) Fault diagnosis and classification: Input the weights of the highest accuracy rate and the test set into the network for fault classification. Draw a visual confusion matrix and a clustering diagram according to the classification results to further illustrate the classification effect. Among them, the public dataset is used to verify the feasibility of the proposed model framework, and the data of the rolling mill bearings and gears obtained through experiments is used for the application of the framework.

[0056] In one embodiment, in order to test the performance of the improved ShuffleNetV2 network model, verify its effectiveness and apply it to the fault diagnosis and classification of rolling mill bearings and gears, experiments are carried out on bearings and gears respectively on two datasets, namely the bearing public dataset of a certain university and the rolling mill bearing fault dataset, and the gear public dataset of another university and the rolling mill gear fault dataset. Two comparisons are also made. One is to compare with other deep learning-based methods to evaluate the feasibility and applicability of the model framework for the bearing and gear fault diagnosis methods. The other is to compare with the existing methods verified on the bearing database and gear dataset to evaluate the superiority of the model framework for the rolling mill bearing and gear fault diagnosis.

[0057] All network models were trained using Python 3.8 programming under the PyTorch framework, with an Intel Core i5-7300HQ CPU@2.5GHz and a GTX1050 (4G) under Windows11.

[0058] Case 1: Verify the feasibility and effectiveness of the rolling mill fault diagnosis model framework for bearing fault diagnosis (1) Data division To verify the effectiveness of the proposed framework for bearing diagnosis, the datasets used come from the bearing public dataset and the rolling mill bearing fault dataset of the research group. The bearing public dataset comes from the bearing data center of a certain university. The test platform it uses includes an electric motor, a sensor, a power meter, and electronic monitoring equipment. It mainly collects signals and detects the bearings supporting the electric motor. The test bearing is a deep groove ball bearing, and the fault of the bearing is an artificial fault, processed by single-point fault with electric spark. This dataset consists of healthy (H) and 3 artificial faults, namely inner ring fault (IR), rolling element fault (B), and outer ring fault (OR). Each fault state has three fault sizes, which are 0.007, 0.014, and 0.021 inches respectively. Four working conditions of 0HP, 1HP, 2HP, and 3HP are selected for the bearing operating state.

[0059] The present invention uses a sampling frequency of 12Khz, a motor load of 0.75KW, and working conditions at 4 different speeds; three different fault diameters corresponding to three types of inner ring, outer ring, and rolling element (at the 6 o'clock direction) are selected as fault data samples, plus a data sample of a healthy state, divided into ten categories in total. Each category generates 600 sample picture data through data processing, and is divided into a training set and a test set according to 8:2. The data division method in Table 1 is used to verify the feasibility of the proposed rolling mill fault diagnosis model framework for bearing classification.

[0060] Table 1 Bearing fault type samples of the bearing public dataset

[0061] Among them, as Figure 5 shown, Figure 5 is the RGB image after dimensionality reduction corresponding to the sampled bearing public dataset, Figure 5 Figure (a) in is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label H in the sampled bearing public dataset, Figure 5 Figure (b) in is the RGB image after dimensionality reduction corresponding to the bearing sample with the fault label B7 in the sampled bearing public dataset, Figure 5Figure (c) in it is the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of IR7 in the sampled bearing public dataset. Figure 5 Figure (d) in it is for the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of Figure 5 Figure (e) in it is the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of B14 in the sampled bearing public dataset. Figure 5 Figure (f) in it is the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of IR14 in the sampled bearing public dataset. Figure 5 Figure (g) in it is for the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of Figure 5 Figure (h) in it is the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of B21 in the sampled bearing public dataset. Figure 5 Figure (i) in it is the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of IR21 in the sampled bearing public dataset. Figure 5 Figure (j) in it is for the reduced - dimensional RGB image corresponding to the bearing sample with a fault label of

[0062] Among them, the rolling - mill bearing fault dataset comes from the rolling - mill vibration simulation experimental platform of a certain research group. The experimental platform used includes an electric motor, one horizontal and one vertical exciter, a planetary gearbox, a reduction gearbox, an operation console and electronic acquisition equipment. Among them, 2 acceleration sensors for collecting the vibration signals of the rolling - mill bearings are located at the vertical and horizontal positions of the upper working roll. 4 acceleration sensors for collecting the vibration signals of the rolling - mill gears are respectively located at the vertical and horizontal positions of the driving wheel and the driven wheel of the reduction gearbox. It mainly collects and detects the signals of the bearings externally connected to the rolling rolls and the gears of the reduction gearbox. The faults of the bearings and gears are artificial faults, and single - point fault machining is carried out by electric spark. In order to collect data close to that on the public dataset, the control - variable method is used. The rolling - mill bearing fault dataset also consists of healthy (H) and 3 artificial faults, namely inner - race fault (IR), rolling - element fault (B) and outer - race fault ( ).

[0063] Two different load - bearing capacities of 1HP and 2HP and three different rotational speeds of 600, 1200 and 1800 are selected for the bearing operating conditions. For each rotational speed, 6 fault states and 1 healthy state are sampled for the bearing fault categories, a total of 7 categories. 400 sample image data are generated for each category through data processing and divided into a training set and a test set according to 8:2. The data division method in Table 2 is used to verify the effectiveness of the proposed rolling - mill fault diagnosis model framework for classifying rolling - mill bearings.

[0064] Table 2 Division of Bearing Status Datasets of Various Types

[0065] As Figure 6 shown, Figure 6 is the reduced-dimension RGB map corresponding to the rolling mill bearing fault dataset. Figure 6 In (a) of , Figure 6 is the reduced-dimension RGB map corresponding to the bearing data with the fault label H in the rolling mill bearing fault dataset. Figure 6 In (b) of , Figure 6 is the reduced-dimension RGB map corresponding to the bearing data with the fault label B1 in the rolling mill bearing fault dataset. In (c) of , Figure 6 is the reduced-dimension RGB map corresponding to the bearing data with the fault label IR1 in the rolling mill bearing fault dataset. Figure 6 In (d) of , Figure 6 is the reduced-dimension RGB map corresponding to the bearing data with the fault label in the rolling mill bearing fault dataset.

[0066] (2) Bearing Data Processing Since the convolutional neural network has more advantages in processing image data, and the color images have more layers and can better display data features, in the present invention, GADF is used to convert the taken one-dimensional vibration signals into two-dimensional colors, and then RPCA is used to reduce the dimension of the two-dimensional color images, highlighting the data features while reducing noise, making it easier for the network to capture features, and indirectly improving the training speed of the network.

[0067] Due to the limitation of the data volume itself, the semi-overlapping sampling method is used when using GADF to capture more data features. The specific data processing flow is as Figure 7 shown.

[0068] (3) Experimental Comparative Analysis of Bearing Data As Figure 8 shown, Figure 8 is a schematic diagram showing the change of the accuracy rate and loss value of the bearing public dataset with the number of iterations at different rotational speeds. Among them, the four rotational speeds include: 1730, 1750, 1772, and 1792. Figure 8Figure (a) in it is a curve showing the variation of the accuracy of fault diagnosis of the bearing public dataset with the number of iterations at rotational speeds of 1730, 1750, 1772, and 1792. Figure 8 Figure (b) in it is a curve showing the variation of the loss value of the model with the number of iterations of the bearing public dataset at rotational speeds of 1730, 1750, 1772, and 1792. Figure 8 This fully demonstrates the feasibility of the model framework under variable rotational speeds.

[0069] To further illustrate the feasibility and classification effect of the dual CABM-ShuffleNetV2 for bearing classification, a classification confusion matrix of the bearing public dataset at 4 rotational speeds as shown in Figure 9 is plotted. Among them, Figure (a) is the classification confusion matrix diagram of the bearing public dataset at a rotational speed of 1730, Figure (b) is the classification confusion matrix diagram of the bearing public dataset at a rotational speed of 1750, Figure (c) is the classification confusion matrix diagram of the bearing public dataset at a rotational speed of 1772, and Figure (d) is the classification confusion matrix diagram of the bearing public dataset at a rotational speed of 1792. Each node on the horizontal and vertical coordinates of the classification confusion matrix diagram corresponds to a set fault type, and the elements on the diagonal represent the coincidence degree of the correct label and the predicted label. It can be seen that the number of misclassified ones is very small, indicating that the feasibility of the fault diagnosis of the bearing by this model framework is guaranteed.

[0070] To further verify the feasibility of the model for bearing fault diagnosis, a comparison is made with other convolutional neural networks of the same type, namely ResNeXt with a transfer learning module added, ResNeXt without a transfer learning module added, the residual network TL-ResNet with transfer learning but without grouped processing; the shallow networks GoogLeNet and AlexNet. To clearly highlight the feasibility of the model, several common evaluation criteria in the field of deep learning fault diagnosis are selected in the present invention, such as accuracy precision recall and score. The expressions of several indicators are as follows: ; ; ; ; where: and are respectively i the number of correctly predicted ones in and are respectively i the number of wrongly predicted ones in

[0071] The comparison of each model parameter is shown in Table 3.

[0072] Table 3 Parameters of different network models for bearings in a certain university

[0073] To further illustrate the superiority of the model, the present invention also compares some existing networks that have studied the bearing dataset as shown in Table 4. The existing networks include: UFE-ResNet50, L-CNN, IS-DATN, ADA, and GGRU-1DCNN. It can be clearly seen from the accuracy alone that the model of the method of the present invention has superiority in bearing diagnosis.

[0074] Table 4 Comparison of accuracy of different algorithms

[0075] (4) Application to the diagnosis of rolling mill bearings As Figure 10 shown, Figure 10 Figure shows the accuracy and loss values of the rolling mill bearing fault dataset at three speeds. Among them, Figure (a) shows the accuracy and loss values of the rolling mill bearing fault dataset at a speed of 600, Figure (b) shows the accuracy and loss values of the rolling mill bearing fault dataset at a speed of 1200, and Figure (c) shows the accuracy and loss values of the rolling mill bearing fault dataset at a speed of 1800; it shows that very good accuracy and small loss can be achieved at three different speeds, fully demonstrating the effectiveness of the model framework in diagnosing rolling mill bearings under variable speeds.

[0076] To further illustrate the effectiveness and classification effect of the dual CABM-ShuffleNetV2 in diagnosing rolling mill bearings, the present invention draws the classification confusion matrix and clustering result diagrams at three speeds as shown in Figure 11 , Figure 12 shown. Among them, Figure 11 Figure (a) in shows the classification confusion matrix of the rolling mill bearing fault dataset at a speed of 600, Figure 11 Figure (b) in shows the classification confusion matrix of the rolling mill bearing fault dataset at a speed of 1200, Figure 11 Figure (c) in shows the classification confusion matrix of the rolling mill bearing fault dataset at a speed of 1800, Figure 12 Figure (a) in shows the data sample diagram of the rolling mill bearing fault dataset at a speed of 600, Figure 12 Figure (c) in shows the data sample diagram of the rolling mill bearing fault dataset at a speed of 1200, Figure 12 Figure (e) in shows the data sample diagram of the rolling mill bearing fault dataset at a speed of 1800, Figure 12Figure (b) in it is the clustering diagram of the rolling mill bearing fault dataset at a rotational speed of 600, Figure 12 Figure (d) in it is the clustering diagram of the rolling mill bearing fault dataset at a rotational speed of 1200, Figure 12 Figure (f) in it is the clustering diagram of the rolling mill bearing fault dataset at a rotational speed of 1800. It can be seen from Figure 12 that the number of misclassified ones is very small, and the obvious clustering effect can also be seen from the clustering diagram, indicating that the proposed model framework is effective for the fault diagnosis of rolling mill bearings and the classification effect is very significant.

[0077] Case 2: Verify the feasibility and effectiveness of the rolling mill fault diagnosis model framework for gear classification (1) Data division To verify the effectiveness of the proposed framework for gear diagnosis, the datasets used come from the publicly available gear dataset and the rolling mill gear fault dataset. The publicly available gear dataset comes from the gear database of another university. The experimental platform used in this experiment is the Drivetrain Dynamic Simulator (DDS) for the gear experiment platform. This platform is composed of a motor, a motor controller, a planetary gearbox, a reduction gearbox, a brake, and a brake controller, and its main function is to collect signals and detect gears. The test gear is a standard part, and the gear faults are also artificial faults, processed by single-point faults with electric sparks; four types of faults, namely gear defect, tooth breakage, tooth root wear, and tooth surface wear, are set, plus the normal state, a total of five categories, as shown in Table 5. Two operating conditions, 0HP no load and 1HP load, are selected for the gear operation state.

[0078] The present invention selects 5 types of fault sample data under two operating conditions. Each category generates 600 sample picture data through data processing, and is divided into a training set and a test set according to 8:2 to verify the feasibility of the proposed rolling mill fault diagnosis model for gear classification using the data division method in Table 5.

[0079] Table 5 Gear fault type samples of the publicly available gear dataset

[0080] The rolling mill gear fault dataset sets three types of faults, namely tooth breakage, tooth root wear, and tooth surface wear, plus the normal state, a total of four categories, as shown in Table 6. The gear operation state selects three different rotational speeds of 600, 1200, and 1800, the same as those of the rolling mill bearings. Each category generates 400 sample picture data through data processing, and is divided into a training set and a test set according to 8:2 to verify the effectiveness of the proposed rolling mill fault diagnosis model for gear classification using the data division method in Table 6.

[0081] Table 6 Division of gear various state datasets

[0082] (2)Gear data processing The same method as that for bearing data processing is adopted, such as Figure 13 shown and Figure 14 shown, Figure 13 is the reduced-dimension RGB image corresponding to the sampled public gear dataset, Figure 14 is the reduced-dimension RGB image corresponding to the rolling mill gear fault dataset. Specifically, Figure 13 In (a) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of healthy in the sampled public gear dataset, Figure 13 In (b) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of gear defect in the sampled public gear dataset, Figure 13 In (c) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of tooth root wear in the sampled public gear dataset, Figure 13 In (d) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of tooth surface wear in the sampled public gear dataset, Figure 13 In (e) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of broken tooth in the sampled public gear dataset; Figure 14 In (a) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of healthy in the rolling mill gear fault dataset, Figure 14 In (b) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of tooth root wear in the rolling mill gear fault dataset, Figure 14 In (c) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of tooth surface wear in the rolling mill gear fault dataset, Figure 14 In (d) of , it is the reduced-dimension RGB image corresponding to the gear data with the fault label of broken tooth in the rolling mill gear fault dataset.

[0083] (3)Experimental comparison and analysis of gear data from another university As Figure 15 shown, Figure 15 is the curve of the accuracy rate and loss value varying with the number of iterations of the sampled public gear dataset under different vibration conditions. The vibration conditions include vibration frequencies of 20 and 30; Figure 15 In (a) of , it is the curve of the accuracy rate of fault diagnosis varying with the number of iterations of the sampled public gear dataset at vibration frequencies of 20 and 30, Figure 15 In (b) of , it is the curve of the loss value of the model varying with the number of iterations of the sampled public gear dataset at vibration frequencies of 20 and 30.

[0084] Very good accuracy can be achieved at two different rotational speeds, which fully demonstrates the feasibility of the model framework under variable rotational speeds.

[0085] To further illustrate the feasibility and classification effect of the dual CABM-ShuffleNetV2 for gear classification, the confusion matrices of the sampled gear public dataset under 2 vibration conditions are plotted. Among them, Figure (a) is the confusion matrix of the sampled gear public dataset at a vibration frequency of 20, and Figure (b) is the confusion matrix of the sampled gear public dataset at a vibration frequency of 30. Each node on the horizontal and vertical coordinates of the confusion matrix corresponds to a set fault type, and the elements on the diagonal represent the coincidence degree of the correct label and the predicted label. It can be seen that the number of misclassified cases is very small, indicating that the feasibility of the fault diagnosis of the bearing by this model framework is guaranteed. Figure 16

[0086] Similar to the comparison of the same type of network for bearings, the comparison of each model parameter is shown in Table 7.

[0087] Table 7 Each parameter of the sampled gear public dataset for different network models

[0088] To further illustrate the superiority of the model for gear fault diagnosis, the present invention is compared with the existing networks for the research on the gear public dataset of another university (SUFD). The existing networks include: SAE-DNN, GRU, BIGRU, LFGRU, VGG; the results are shown in Table 8. Compared with the gear classification networks used above, the proposed model has higher classification accuracy.

[0089] Table 8 Comparison of the accuracy of different algorithms

[0090] (4)Application to the diagnosis of rolling mill gears As Figure 17 shown, Figure 17 are the curves of the accuracy and loss value of the rolling mill gear fault dataset changing with the number of iterations at 3 rotational speeds. Figure 17 In Figure (a) of Figure 17 is the curve of the accuracy and loss value of the rolling mill gear fault dataset changing with the number of iterations at a rotational speed of 600. Figure 17 ​Figure (c) shows the curves of the accuracy and loss values of the rolling mill gear fault dataset varying with the number of iterations at a rotational speed of 1800. Very good accuracy and small loss can be achieved at three different rotational speeds, fully demonstrating the effectiveness of the model framework in diagnosing rolling mill gears under variable rotational speeds.

[0091] To further illustrate the effectiveness and classification effect of the dual CABM-ShuffleNetV2 in diagnosing rolling mill gears, the following are plotted as Figure 18 , Figure 19 shown, Figure 18 and Figure 19 respectively, which are the classification confusion matrix and clustering result diagram of the rolling mill gear fault dataset at three rotational speeds. Among them, Figure 18 Figure (a) in Figure 18 is the classification confusion matrix of the rolling mill gear fault dataset at a rotational speed of 600, Figure 18 Figure (b) in Figure 19 is the classification confusion matrix of the rolling mill gear fault dataset at a rotational speed of 1200, Figure 19 Figure (c) in Figure 19 is the classification confusion matrix of the rolling mill gear fault dataset at a rotational speed of 1800, Figure 19 Figure (a) in Figure 19 is the data sample diagram of the rolling mill gear fault dataset at a rotational speed of 600, Figure 19 Figure (c) in Figure 19 is the data sample diagram of the rolling mill gear fault dataset at a rotational speed of 1200,

[0092] The fault diagnosis method for multiple components of the rolling mill of the present invention has the following advantages: 1) Utilize the superiority of convolutional neural networks in recognizing image data, and use GADF to convert the original one-dimensional vibration signal into two-dimensional time-frequency image data.

[0093] 2) Denoise and enhance the features of the two-dimensional time-frequency image data through RPCA, and reduce the data itself. It not only retains more data information, enhances the features, but also reduces the data size, greatly reducing the parameters, calculation amount, inference time, etc. of the lightweight network.

[0094] 3) Use a dual CABM module to enhance the perception ability of the ShuffleNetV2 network model, and transfer learning is also inserted in the network model to accelerate the convergence speed of the model.

[0095] 4) For the two cases of bearings and gears respectively, publicly available datasets that have been verified are used to verify the feasibility of the model. The accuracy is guaranteed under different working conditions, and the superiority of the model is demonstrated by comparing with other models that have used the same set of data. Finally, it is applied on a mill simulation platform. According to the experimental results, it can be seen that the proposed model framework is effective for the fault diagnosis of multiple components of the mill.

[0096] When applying the fault diagnosis method for multiple components of the mill provided by the present invention, it is not necessary to Figure 1 execute according to the order of the steps shown. The specific execution order of each step can be determined as needed, and the present invention does not limit this.

[0097] The above is the fault diagnosis method for multiple components of the mill provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding fault diagnosis device for multiple components of the mill. The device includes: An acquisition module, configured to acquire the status data of multiple components of the mill; A conversion module, configured to convert the status data into an RGB image through the Gram angular difference field, convert the RGB image into three images of the R, G, and B channels, and convert the image of each channel into a grayscale image; A dimensionality reduction module, configured to perform dimensionality reduction processing on each grayscale image respectively, and obtain a reduced-dimensional RGB image according to the three reduced-dimensional grayscale images; A classification module, configured to input the reduced-dimensional RGB image into an improved ShuffleNetV2 network model to obtain a fault diagnosis result of the mill; the improved ShuffleNetV2 network model is obtained by adding a dual attention mechanism to the basic ShuffleNetV2 network for training.

[0098] For the specific limitations of the fault diagnosis device for multiple components of the mill, reference can be made to the limitations of the fault diagnosis method for multiple components of the mill in the above text, which will not be elaborated here. Each module in the above-mentioned fault diagnosis device for multiple components of the mill can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0099] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program can be used to execute the above-mentioned Figure 1A fault diagnosis method for multiple components of a rolling mill provided

[0100] The present invention also provides Figure 20 A schematic structural diagram of the computer device shown, as Figure 20 shown. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above Figure 1 A fault diagnosis method for multiple components of a rolling mill provided

[0101] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention can include at least one of non-volatile and volatile memories. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. The volatile memory can include random access memory (RAM) or an external cache. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0102] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded by the present invention.

Claims

1. A fault diagnosis method for multiple components of a rolling mill, characterized in that, Including: Obtaining the status data of multiple components of a rolling mill; Converting the status data into an RGB image through the Gram angular difference field; Converting the RGB image into three images of the R, G, and B channels, and converting the image of each channel into a grayscale image; Performing dimensionality reduction processing on each grayscale image respectively, and obtaining a dimensionality-reduced RGB image according to the three dimensionality-reduced grayscale images; Inputting the dimensionality-reduced RGB image into an improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill; the improved ShuffleNetV2 network model is obtained by adding a dual attention mechanism to the basic ShuffleNetV2 network for training.

2. The method according to claim 1, wherein The performing dimensionality reduction processing on each grayscale image respectively includes: For any grayscale image, obtaining the mean value of all data points in the grayscale image; Determining a divergence matrix according to the values of each data point and the mean value, and solving the eigenvalues and eigenvectors of the divergence matrix; Determine the eigenvectors corresponding to the largest first K several eigenvalues as the new eigenvectors; Determining the dimensionality-reduced grayscale image according to the new eigenvectors and the original grayscale image.

3. The method according to claim 1, characterized in that The improved ShuffleNetV2 network model includes an input module with a fusion attention mechanism, a plurality of lightweight modules connected in series in sequence, and an output module with a fusion attention mechanism; the inputting the dimensionality-reduced RGB image into the improved ShuffleNetV2 network model to obtain the fault diagnosis result of the rolling mill includes: Inputting the dimensionality-reduced RGB image into the improved ShuffleNetV2 network model, and extracting features in the dimensionality-reduced RGB image through the input module; Performing convolution operations on the features in the dimensionality-reduced RGB image in sequence through a plurality of lightweight modules connected in series to obtain convolution features; Performing feature prediction on the convolution features through the output module to obtain the fault diagnosis result of the rolling mill.

4. The method according to claim 3, wherein The input module includes an input layer, a convolutional layer, an activation function, an attention mechanism layer, and a fully connected layer connected in series in sequence; the extracting features in the dimensionality-reduced RGB image through the input module includes: Receiving the dimensionality-reduced RGB image through the input layer, and performing a convolution operation on the dimensionality-reduced RGB image through the convolutional layer to extract local features; Increasing the non-linear characteristics of the local features through the activation function to obtain non-linear features; Assigning weights to different features in the non-linear features through the attention mechanism layer; Performing a fully connected operation on the features output by the attention mechanism layer through the fully connected layer to obtain the features in the dimensionality-reduced RGB image.

5. The method according to claim 4, characterized in that, The attention mechanism layer includes a channel domain attention module and a spatial domain attention module; the channel domain attention module and the spatial domain attention module are designed in a cascade manner.

6. The method according to claim 3, wherein The output module includes a convolutional layer, an attention mechanism layer, and a fully connected layer connected in series in sequence; the performing feature prediction on the convolution features through the output module to obtain the fault diagnosis result of the rolling mill includes: Performing a convolution process on the convolution features through the convolutional layer, and assigning weights to the convolved features through the attention mechanism layer; Mapping the features output by the attention mechanism layer to each fault category through the fully connected layer to obtain the fault diagnosis result of the rolling mill.

7. The method according to claim 1, characterized in that, The fault diagnosis results of the rolling mill include the probabilities of various components of the rolling mill having various fault categories; the method further includes: If the probability is greater than a preset fault threshold, the fault category corresponding to the probability is determined as the fault category that occurs in the rolling mill.

8. The method according to claim 1, wherein The training process of the improved ShuffleNetV2 network model includes: Introducing a dual attention mechanism into the basic ShuffleNetV2 network to obtain an initial ShuffleNetV2 network; Obtaining a training set and pre-training the initial ShuffleNetV2 network on the training set through transfer learning to obtain an initial ShuffleNetV2 network model; the weight parameters in the initial ShuffleNetV2 network model are weight parameters with general feature extraction capabilities; Iteratively training the initial ShuffleNetV2 network model with the training set to obtain an improved ShuffleNetV2 network model.

Images