CN118035889A - Method, device, equipment, storage medium and product for fault classification of rotary equipment - Google Patents

Abstract

The application relates to a fault classification method, a fault classification device, a fault classification storage medium and a fault classification product for rotary equipment. The method comprises the following steps: extracting characteristics of fault information of the rotary equipment to be detected to obtain fault characteristics of the fault information; inputting fault characteristics into a target multi-scale multi-head self-attention force diagram convolutional neural network GCN model to obtain fault types of the rotating equipment to be tested; the target multi-scale multi-head self-attention force-seeking convolutional neural network GCN model is obtained by training an initial multi-scale multi-head self-attention GCN model based on a fusion graph feature sample, wherein the fusion graph feature sample is obtained by fusing target output information and a plurality of graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on a fault information sample, and the graph feature sample is determined by the fault information sample, an adjacent matrix and the initial multi-scale multi-head self-attention GCN model. By adopting the method, the fault detection precision can be improved.

Description

Fault classification method, device, equipment, storage medium and product for rotating equipment

Technical Field

The present application relates to the field of machine learning technology, and in particular to a method, apparatus, device, storage medium and product for fault classification of rotating equipment.

Background technique

Common rotating equipment such as motors and engines are prone to failure during operation. Therefore, it is necessary to perform fault detection on rotating equipment.

With the development of machine learning technology, learning models such as support vector machines and graph convolutional neural networks are increasingly used in fault detection of rotating equipment.

When training traditional graph convolutional neural networks, since a single graph network is used for detection, it is difficult to extract sufficient information from limited training samples. Therefore, the fault detection accuracy of the trained graph convolutional neural network model is low.

Summary of the invention

Based on this, it is necessary to provide a fault classification method, device, equipment, medium and product for rotating equipment that can improve the fault detection accuracy in response to the above technical problems.

In a first aspect, the present application provides a method for fault classification of rotating equipment. The method comprises:

Extract features of fault information of the rotating equipment to be tested to obtain fault features of the fault information;

The fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is trained by the initial multi-scale multi-head self-attention GCN model based on the fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

In one embodiment, the initial multi-scale multi-head self-attention GCN model includes multiple first sub-models and a multi-layer perceptron; the method further includes:

For each first sub-model, the fault information sample and the adjacency matrix corresponding to each first sub-model are input into the first sub-model to obtain a graph feature sample output by the first sub-model;

Inputting the fault information sample into the multilayer perceptron to obtain first output information; the target output information includes the first output information;

Obtaining a fused graph feature sample according to the first output information and each graph feature sample;

The initial multi-scale multi-head self-attention GCN model is trained according to the fusion graph feature samples to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In one embodiment, the first sub-model includes a first graph convolution layer and a second graph convolution layer, and the initial multi-scale multi-head self-attention GCN model includes a connection layer and a linear layer; and obtaining a fused graph feature sample according to the first output information and each graph feature sample includes:

For each first sub-model, input the fault information sample and the adjacency matrix corresponding to each first sub-model into the first graph convolution layer of the first sub-model to obtain a first intermediate graph feature;

Inputting the first intermediate graph features obtained by the first graph convolution layer of each first sub-model into the connection layer to obtain the second intermediate graph features;

Inputting the second intermediate graph feature into the linear layer to obtain second output information;

A fused graph feature sample is obtained according to the first output information, the second output information and each graph feature sample; the target output information includes the first output information and the second output information.

In one embodiment, a fused graph feature sample is obtained according to the first output information, the second output information and each graph feature sample; the target output information includes the first output information and the second output information, including:

Obtaining weighted values of the first output information, the second output information, and each graph feature sample according to the first output information, the second output information, each graph feature sample, a first weight corresponding to the first output information, a second weight corresponding to the second output information, and a third weight corresponding to each graph feature sample;

The fusion graph feature samples are obtained based on the weighted values.

In one embodiment, the number of fused graph feature samples is multiple; the initial multi-scale multi-head self-attention GCN model includes a multi-head self-attention fusion model and a fully connected layer; the initial multi-scale multi-head self-attention GCN model is trained according to the fused graph feature samples to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model, including:

Use the multi-head self-attention fusion model to connect the fusion graph feature samples to obtain the graph feature vector;

Use the fully connected layer to get the predicted fault category based on the graph feature vector;

According to the error between the predicted fault category and the actual fault category corresponding to the fault information sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In one embodiment, according to the error between the predicted fault category and the actual fault category corresponding to the fault information sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model, including:

Determine the loss value according to the error between the predicted fault category and the actual fault category corresponding to the fault information sample;

When the loss value is greater than the preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are adjusted according to the loss value to obtain the intermediate GCN model, so as to train the initial multi-scale multi-head self-attention GCN model until the loss value is less than the preset threshold, and the intermediate GCN model corresponding to the loss value less than the preset threshold is used as the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In a second aspect, the present application also provides a fault classification device for rotating equipment. The device comprises:

A feature extraction module is used to extract features from the fault information of the rotating equipment to be tested, and obtain the fault features of the fault information;

A fault classification module is used to input fault features into a target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is obtained by training the initial multi-scale multi-head self-attention GCN model based on fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods in the first aspect when executing the computer program.

In a fourth aspect, the present application further provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the steps of any one of the methods in the first aspect when executed by a processor.

The above-mentioned fault classification method, device, equipment, storage medium and product of rotating equipment fuse the target output information and multiple graph feature samples to obtain a fused graph feature sample, and then based on the fused graph feature sample, train the initial multi-scale multi-head self-attention GCN model to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model. Since the fused graph feature sample includes the target output information and multiple graph feature samples, the information contained is richer than the information contained in a single graph feature sample in traditional technology. Therefore, the prediction accuracy of the target multi-scale multi-head self-attention graph convolutional neural network GCN model obtained by training based on the fused graph feature sample with richer information is higher, thereby improving the accuracy of fault detection.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is an application environment diagram of a fault classification method for rotating equipment in one embodiment;

FIG2 is a schematic flow chart of a method for classifying faults of rotating equipment in one embodiment;

FIG3 is a schematic diagram of a patterning process in one embodiment;

FIG4 is a schematic diagram of a model corresponding to when K=2 and a model corresponding to when K=4 in one embodiment;

FIG5 is a schematic diagram of a process of training a target multi-scale multi-head self-attention graph convolutional neural network GCN model in one embodiment;

FIG6 is a schematic diagram of a process of training an initial multi-scale multi-head self-attention GCN model according to a fusion graph feature sample to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model in one embodiment;

FIG7 is a schematic flow chart of a method for classifying faults of rotating equipment in an exemplary embodiment;

FIG8 is a schematic diagram of a target multi-scale multi-head self-attention graph convolutional neural network GCN model in an exemplary embodiment;

FIG9 is a schematic diagram of a multi-head self-attention fusion model in an exemplary embodiment;

FIG10 is a schematic diagram of a fault classification device for a rotating device according to an embodiment;

FIG11 is a diagram showing the internal structure of a server in one embodiment;

FIG. 12 is a diagram showing the internal structure of a terminal in one embodiment.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

The fault classification method for rotating equipment provided in the embodiment of the present application can be applied in the application environment shown in FIG1. The computer device 102 extracts features from the fault information of the rotating equipment to be tested to obtain the fault features of the fault information; the fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is obtained by training the initial multi-scale multi-head self-attention GCN model based on the fusion graph feature samples, the fusion graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model. The computer device 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc.

In one embodiment, as shown in FIG. 2 , a method for fault classification of rotating equipment is provided, and the method is applied to the computer device 102 in FIG. 1 as an example for description, including the following steps:

Step 202: extracting features from the fault information of the rotating equipment to be tested, and obtaining fault features of the fault information.

The rotating device to be tested may be any one of a motor, an engine, and the like, and this embodiment does not limit this.

Optionally, first, a vibration sensor, a current sensor or other sensor installed in the rotating device to be tested is used to obtain a fault signal of the rotating device to be tested, and the fault signal of the rotating device to be tested is segmented to obtain fault information of the rotating device to be tested; then, feature extraction is performed on the fault information of the rotating device to be tested to obtain fault features of the fault information. The feature extraction method may be a fast Fourier transform method and a normalization method, or other methods, which are not limited in this embodiment.

Step 204, input the fault feature into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is based on the fusion graph feature samples, and the initial multi-scale multi-head self-attention GCN model is trained, the fusion graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

Optionally, since there is no explicit association between fault information samples, graph composition is required when using the initial multi-scale multi-head self-attention GCN model. A certain fault information sample can be regarded as a central node, and the KNN (K-nearestNeighbor) algorithm is used to calculate the pairwise distance between the central node and other fault information samples. The K fault information samples closest to the central node are regarded as the neighboring nodes associated with the central node, and the adjacency matrix A is used to describe the association between the fault information samples on the graph. The formula of the adjacency matrix A is shown in formula (1).

(1)

In formula (1), _xi is the central node, and _xj is the neighboring node associated with the central node.

Optionally, different K values may be taken to construct different graphs, and the number of constructed graphs may be an integer greater than or equal to 2, which is not limited in this embodiment. Assuming that the number of constructed graphs is 2, the values of K are 2 and 4 respectively, and the graph construction process is shown in FIG3. When K=2, the fault information sample _X1 is taken as the central node, the neighboring nodes associated with _X1 are _X2 and _X3 , and the adjacency matrix is A1. When K=4, the fault information sample _X5 is taken as the central node, the neighboring nodes associated with _X5 are _X1 , _X2 , _X3 , and _X4 , and the adjacency matrix is A2.

Figure 4 shows schematic diagrams of the model corresponding to K=2 and the model corresponding to K=4. The model corresponding to K=2 is represented as the first sub-model 1, and the first sub-model 1 includes two graph convolution layers, namely graph convolution layer 1-1 and graph convolution layer 1-2. The model corresponding to K=4 is represented as the first sub-model 2, and the first sub-model 2 includes two graph convolution layers, namely graph convolution layer 2-1 and graph convolution layer 2-2.

Input the fault information sample and the adjacency matrix A1 into the first sub-model 1 to obtain the graph feature sample 1, and input the fault information sample and the adjacency matrix A2 into the first sub-model 2 to obtain the graph feature sample 2. Input the fault information sample into the target network in the initial multi-scale multi-head self-attention GCN model to obtain the target output information. Among them, the target network in the initial multi-scale multi-head self-attention GCN model can be a multi-layer perceptron, or a sub-network composed of a connection layer and a linear layer, or a multi-layer perceptron and the sub-network, that is, the target output information can be the first output information output by the multi-layer perceptron, or the target output information can also be the second output information output by the sub-network, or the target output information can include the first output information and the second output information. This embodiment is not limited to this.

The target output information and multiple graph feature samples are fused to obtain fused graph feature samples. Based on the fused graph feature samples, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model. The fusion here can be weighted summation or other calculation methods, which is not limited in this embodiment.

The fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network (GCN) model to obtain the fault category of the rotating equipment to be tested.

In the above-mentioned fault classification method for rotating equipment, feature extraction is performed on the fault information of the rotating equipment to be tested to obtain the fault features of the fault information; the fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is obtained by training the initial multi-scale multi-head self-attention GCN model based on the fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model. Among them, the target output information and multiple graph feature samples are fused to obtain fused graph feature samples, and then based on the fused graph feature samples, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model. Since the fused graph feature samples include the target output information and multiple graph feature samples, the information contained is richer than the information contained in a single graph feature sample in traditional technology. Therefore, the target multi-scale multi-head self-attention graph convolutional neural network GCN model trained based on the fused graph feature samples with richer information has higher prediction accuracy, thereby improving the accuracy of fault detection.

In one embodiment, the initial multi-scale multi-head self-attention GCN model includes multiple first sub-models and multi-layer perceptrons. The process of training to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model is shown in FIG5, including:

Step 502: For each first sub-model, the fault information sample and the adjacency matrix corresponding to each first sub-model are input into the first sub-model to obtain the graph feature sample output by the first sub-model.

Optionally, assuming that the initial multi-scale multi-head self-attention GCN model includes two first sub-models, namely, first sub-model 1 and first sub-model 2, the fault information sample and the adjacency matrix corresponding to the first sub-model 1 are input into the first sub-model 1 to obtain the graph feature sample 1 output by the first sub-model. The fault information sample and the adjacency matrix corresponding to the first sub-model 2 are input into the first sub-model 2 to obtain the graph feature sample 2 output by the first sub-model.

Step 504: input the fault information sample into the multilayer perceptron to obtain first output information; the target output information includes the first output information.

Optionally, the fault information sample is input into a multilayer perceptron to obtain first output information, and the first output information is shown in formula (2).

(2)

In formula (2), is the feature corresponding to the fault information sample, /> It is the first output information.

Step 506: Obtain a fused graph feature sample according to the first output information and each graph feature sample.

Optionally, determine a first weight corresponding to the first output information and a third weight corresponding to each graph feature sample, add the product of the first weight and the first output information and the product of each graph feature sample and the third weight, and use the summed result as the fused graph feature sample.

Step 508, training the initial multi-scale multi-head self-attention GCN model according to the fusion graph feature samples to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

Optionally, the number of fusion graph feature samples may be one or more, which is not limited in this embodiment. If the number of fusion graph feature samples is one, the predicted fault category is obtained based on the fusion graph feature samples using the fully connected layer in the initial multi-scale multi-head self-attention GCN model, and the error between the predicted fault category and the actual fault category corresponding to the fault information sample is used to train the initial multi-scale multi-head self-attention GCN model to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In this embodiment, for each first sub-model, the fault information sample and the adjacency matrix corresponding to each first sub-model are input into the first sub-model to obtain the graph feature sample output by the first sub-model; the fault information sample is input into the multi-layer perceptron to obtain the first output information; the target output information includes the first output information; the fused graph feature sample is obtained according to the first output information and each graph feature sample; the initial multi-scale multi-head self-attention GCN model is trained according to the fused graph feature sample to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model. Among them, the first output information and multiple graph feature samples are fused to obtain the fused graph feature sample, and then based on the fused graph feature sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model. Since the fused graph feature sample includes the first output information and multiple graph feature samples, the information contained is richer than the information contained in the single graph feature sample in the traditional technology. Therefore, the accuracy of the target multi-scale multi-head self-attention graph convolutional neural network GCN model obtained by training based on the more information-rich fused graph feature sample is also higher.

In one embodiment, the first sub-model includes a first graph convolution layer and a second graph convolution layer, and the initial multi-scale multi-head self-attention GCN model includes a connection layer and a linear layer; obtaining a fused graph feature sample according to the first output information and each graph feature sample includes:

For each first sub-model, input the fault information sample and the adjacency matrix corresponding to each first sub-model into the first graph convolution layer of the first sub-model to obtain a first intermediate graph feature;

Inputting the first intermediate graph features obtained by the first graph convolution layer of each first sub-model into the connection layer to obtain the second intermediate graph features;

The second intermediate graph feature is input into the linear layer to obtain the second output information.

Optionally, assuming that the initial multi-scale multi-head self-attention GCN model includes two first sub-models, namely the first sub-model 1 and the first sub-model 2, the fault information sample and the adjacency matrix corresponding to the first sub-model 1 are input into the first graph convolution layer of the first sub-model 1 to obtain the first intermediate graph feature 1, the fault information sample and the adjacency matrix corresponding to the first sub-model 2 are input into the first graph convolution layer of the first sub-model 2 to obtain the first intermediate graph feature 2, the first intermediate graph feature 1 and the first intermediate graph feature 2 are input into the connection layer to obtain the second intermediate graph feature, the second intermediate graph feature is input into the linear layer to obtain the second output information, and the formula of the second output information is shown in formula (3).

(3)

In formula (3), The first intermediate figure feature 1, /> is the first intermediate image feature 2, W and b are trainable parameters.

A fused graph feature sample is obtained according to the first output information, the second output information and each graph feature sample; the target output information includes the first output information and the second output information.

Optionally, a weighted sum may be performed on the first output information, the second output information, and each graph feature sample, and the result of the weighted sum may be used as a fused graph feature sample.

In this embodiment, for each first sub-model, the fault information sample and the adjacency matrix corresponding to each first sub-model are input into the first graph convolution layer of the first sub-model to obtain the first intermediate graph feature; the first intermediate graph feature obtained by the first graph convolution layer of each first sub-model is input into the connection layer to obtain the second intermediate graph feature; the second intermediate graph feature is input into the linear layer to obtain the second output information; the fused graph feature sample is obtained according to the first output information, the second output information and each graph feature sample; the target output information includes the first output information and the second output information. Among them, the fused graph feature sample includes the first output information, the second output information and each graph feature sample, and the information contained is richer than the information contained in a single graph feature sample in the traditional technology.

In one embodiment, a fused graph feature sample is obtained according to the first output information, the second output information and each graph feature sample; the target output information includes the first output information and the second output information, including:

Obtaining weighted values of the first output information, the second output information, and each graph feature sample according to the first output information, the second output information, each graph feature sample, a first weight corresponding to the first output information, a second weight corresponding to the second output information, and a third weight corresponding to each graph feature sample;

The fusion graph feature samples are obtained based on the weighted values.

Optionally, formulas for calculating a first weight corresponding to the first output information, a second weight corresponding to the second output information, and a third weight corresponding to each graph feature sample are shown in formulas (4) and (5).

(4)

(5)

In formulas (4) and (5), assuming that the number of graph feature samples is 2, the value of i is 1, 2, 3, and 4, _Z1 represents graph feature sample 1, _Z2 represents graph feature sample 2, _Z3 represents the first output information, and _Z4 represents the second output information. and/> is the third weight corresponding to each graph feature sample,/> is the first weight corresponding to the first output information, /> is the second weight corresponding to the second output information. The sum of is 1.

According to the first output information, the second output information, each graph feature sample, the first weight corresponding to the first output information, the second weight corresponding to the second output information, and the third weight corresponding to each graph feature sample, the weighted value of the first output information, the second output information, and each graph feature sample is obtained. The calculation formula is shown in formula (6), and the weighted value is used as the fusion feature sample.

(6)

In formula (6), Represents the fusion graph feature sample.

In this embodiment, the weighted values of the first output information, the second output information, and each graph feature sample are obtained according to the first output information, the second output information, each graph feature sample, the first weight corresponding to the first output information, the second weight corresponding to the second output information, and the third weight corresponding to each graph feature sample; and the fused graph feature sample is obtained based on the weighted value. By assigning different weights to the first output information, the second output information, and each graph feature sample, the first output information, the second output information, and each graph feature sample can be adaptively fused, thereby enriching the information and avoiding the over-smoothing phenomenon caused by the deep network.

In one embodiment, the number of fused graph feature samples is multiple; the initial multi-scale multi-head self-attention GCN model includes a multi-head self-attention fusion model and a fully connected layer; the initial multi-scale multi-head self-attention GCN model is trained according to the fused graph feature samples to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model, and the process is shown in FIG6, including:

Step 602: Use a multi-head self-attention fusion model to connect each fused graph feature sample to obtain a graph feature vector.

Optionally, in order to make the representation ability of the fusion graph feature samples stronger, multiple sets of trainable parameters can be set , and then based on formulas (4), (5) and (6), multiple fusion graph feature samples are calculated./> ,in, The multi-head self-attention fusion model is used to connect multiple fusion graph feature samples, as shown in formula (7), to obtain the graph feature vector.

(7)

In formula (7), is the graph feature vector.

Step 604: Use the fully connected layer to obtain the predicted fault category based on the graph feature vector.

Optionally, the fully connected layer is used to perform linear transformation and Softmax function processing on the graph feature vector to obtain the predicted fault category, as shown in formula (8).

(8)

In formula (8), represents the graph feature vector, W and b represent trainable parameters, and y represents the predicted fault category.

Step 606, according to the error between the predicted fault category and the actual fault category corresponding to the fault information sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

Optionally, a loss function is used to describe the error between the predicted fault category and the actual fault category corresponding to the fault information sample. According to the loss value corresponding to the error, the initial multi-scale multi-head self-attention GCN model is trained until the loss value meets the preset condition, and the target multi-scale multi-head self-attention graph convolutional neural network GCN model is obtained. Among them, the loss function here can be cross entropy or other functions, which is not limited in this embodiment. The preset condition here can be that the loss value is less than the preset threshold, or that the loss value is equal to the preset threshold, which is not limited in this embodiment.

In this embodiment, the multi-head self-attention fusion model is used to connect each fusion graph feature sample to obtain a graph feature vector; the fully connected layer is used to obtain the predicted fault category based on the graph feature vector; according to the error between the predicted fault category and the real fault category corresponding to the fault information sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model. By introducing the multi-head self-attention fusion model, the representation ability of the fusion graph feature sample is made stronger, and the accuracy of the target multi-scale multi-head self-attention graph convolutional neural network GCN model is further improved.

In one embodiment, according to the error between the predicted fault category and the actual fault category corresponding to the fault information sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model, including:

The loss value is determined based on the error between the predicted fault category and the actual fault category corresponding to the fault information sample.

Optionally, cross entropy is used to describe the error between the predicted fault category and the actual fault category corresponding to the fault information sample, and the loss value is determined according to the cross entropy.

When the loss value is greater than the preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are adjusted according to the loss value to obtain the intermediate GCN model, so as to train the initial multi-scale multi-head self-attention GCN model until the loss value is less than the preset threshold, and the intermediate GCN model corresponding to the loss value less than the preset threshold is used as the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

Optionally, if the loss value is greater than a preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are adjusted to obtain an intermediate GCN model, and based on the fault information sample, the adjacency matrix corresponding to each first sub-model in the intermediate GCN model, and each first sub-model in the intermediate GCN model, the graph feature samples output by each first sub-model in the intermediate GCN model are determined, and based on the fault information sample and the intermediate GCN model, the target output information corresponding to the intermediate GCN model is determined, and the graph feature samples output by each first sub-model in the intermediate GCN model and the target output information corresponding to the intermediate GCN model are fused to obtain a fused feature sample corresponding to the intermediate GCN model, and the fused feature samples corresponding to the multiple intermediate GCN models are connected using the multi-head self-attention fusion model to obtain a graph feature vector corresponding to the intermediate GCN model, and the predicted fault category corresponding to the intermediate GCN model is obtained based on the graph feature vector corresponding to the intermediate GCN model using a fully connected layer, and a new loss value is determined according to the error between the predicted fault category corresponding to the intermediate GCN model and the true fault category corresponding to the fault information sample, and if the new loss value is less than the preset threshold, the intermediate GCN model is used as the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In this embodiment, the loss value is determined according to the error between the predicted fault category and the real fault category corresponding to the fault information sample; when the loss value is greater than the preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are adjusted according to the loss value to obtain an intermediate GCN model, so as to train the initial multi-scale multi-head self-attention GCN model until the loss value is less than the preset threshold, and the intermediate GCN model corresponding to the loss value less than the preset threshold is used as the target multi-scale multi-head self-attention graph convolutional neural network GCN model. By comparing the size relationship between the loss value and the preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are continuously adjusted to make the predicted value of the model closer to the true value, thereby improving the accuracy of the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In an exemplary embodiment, a method for fault classification of rotating equipment is provided, and the process is shown in FIG7 , including:

Step 701: for each first sub-model, input the fault information sample and the adjacency matrix corresponding to each first sub-model into the first sub-model to obtain the graph feature sample output by the first sub-model.

Step 702: Input the fault information sample into a multilayer perceptron to obtain first output information; the target output information includes the first output information.

Step 703: For each first sub-model, the fault information sample and the adjacency matrix corresponding to each first sub-model are input into the first graph convolution layer of the first sub-model to obtain the first intermediate graph feature.

Step 704: input the first intermediate graph features obtained by the first graph convolution layer of each first sub-model into the connection layer to obtain the second intermediate graph features.

Step 705: Input the second intermediate graph feature into the linear layer to obtain second output information.

Step 706, obtaining weighted values of the first output information, the second output information and each graph feature sample according to the first output information, the second output information, each graph feature sample, the first weight corresponding to the first output information, the second weight corresponding to the second output information and the third weight corresponding to each graph feature sample.

Step 707: Obtain fusion graph feature samples based on the weighted values.

Step 708: Use the multi-head self-attention fusion model to connect the fused graph feature samples to obtain a graph feature vector.

Step 709: Use the fully connected layer to obtain the predicted fault category based on the graph feature vector.

Step 710: determine a loss value according to an error between the predicted fault category and the actual fault category corresponding to the fault information sample.

Step 711, when the loss value is greater than the preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are adjusted according to the loss value to obtain an intermediate GCN model, so as to train the initial multi-scale multi-head self-attention GCN model until the loss value is less than the preset threshold, and the intermediate GCN model corresponding to the loss value less than the preset threshold is used as the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

Step 712: extract features from the fault information of the rotating equipment to be tested, and obtain fault features of the fault information.

Step 713, input the fault features into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested.

In the above-mentioned fault classification method for rotating equipment, the target output information and multiple graph feature samples are fused to obtain fused graph feature samples, and then based on the fused graph feature samples, the initial multi-scale multi-head self-attention GCN model is trained to obtain a target multi-scale multi-head self-attention graph convolutional neural network GCN model. Since the fused graph feature samples include the target output information and multiple graph feature samples, the information contained is richer than the information contained in a single graph feature sample in traditional technology. Therefore, the prediction accuracy of the target multi-scale multi-head self-attention graph convolutional neural network GCN model obtained by training based on the fused graph feature samples with richer information is higher, thereby improving the accuracy of fault detection.

In an exemplary embodiment, a schematic diagram of a target multi-scale multi-head self-attention graph convolutional neural network GCN model is provided, as shown in Figure 8. The fault information sample, adjacency matrix 1 and adjacency matrix 2 are taken as inputs, and the predicted fault category is obtained after being processed by each module in the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In an exemplary embodiment, a schematic diagram of a multi-head self-attention fusion model is provided, as shown in Figure 9. Graph feature sample 1, graph feature sample 2, first output information, and second output information are taken as input, and processed by a linear layer, a Tanh function, and a Softmax function to obtain a graph feature vector.

It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the present application also provides a rotating equipment fault classification device for implementing the above-mentioned rotating equipment fault classification method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiments of the fault classification device for one or more rotating equipment provided below can refer to the limitations of the rotating equipment fault classification method above, and will not be repeated here.

In one embodiment, as shown in FIG. 10 , a fault classification device 1000 for rotating equipment is provided, comprising: a feature extraction module 1020 and a fault classification module 1040, wherein:

The feature extraction module 1020 is used to extract features from the fault information of the rotating equipment to be tested, and obtain the fault features of the fault information;

The fault classification module 1040 is used to input the fault features into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is obtained by training the initial graph convolutional neural network GCN model based on the fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

In one embodiment, the fault classification device 1000 for rotating equipment further includes:

The first training module is used to input the fault information samples and the adjacency matrix corresponding to each first sub-model into the first sub-model for each first sub-model to obtain the graph feature samples output by the first sub-model; input the fault information samples into the multi-layer perceptron to obtain the first output information; the target output information includes the first output information; obtain the fused graph feature samples according to the first output information and each graph feature sample; train the initial multi-scale multi-head self-attention GCN model according to the fused graph feature samples to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In one embodiment, the fault classification device 1000 for rotating equipment further includes:

The second training module is used to input the fault information sample and the adjacency matrix corresponding to each first sub-model into the first graph convolution layer of the first sub-model for each first sub-model to obtain the first intermediate graph feature; input the first intermediate graph feature obtained by the first graph convolution layer of each first sub-model into the connection layer to obtain the second intermediate graph feature; input the second intermediate graph feature into the linear layer to obtain the second output information; obtain the fused graph feature sample according to the first output information, the second output information and each graph feature sample; the target output information includes the first output information and the second output information.

In one embodiment, the fault classification device 1000 for rotating equipment further includes:

The third training module is used to obtain the weighted values of the first output information, the second output information and each graph feature sample according to the first output information, the second output information, each graph feature sample, the first weight corresponding to the first output information, the second weight corresponding to the second output information, and the third weight corresponding to each graph feature sample; and obtain the fused graph feature sample based on the weighted value.

In one embodiment, the fault classification device 1000 for rotating equipment further includes:

The fourth training module is used to use the multi-head self-attention fusion model to connect each fused graph feature sample to obtain a graph feature vector; use the fully connected layer to obtain the predicted fault category based on the graph feature vector; according to the error between the predicted fault category and the actual fault category corresponding to the fault information sample, the initial multi-scale multi-head self-attention GCN model is trained to obtain the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

In one embodiment, the fault classification device 1000 for rotating equipment further includes:

The fifth training module is used to determine the loss value according to the error between the predicted fault category and the actual fault category corresponding to the fault information sample; when the loss value is greater than a preset threshold, the parameters of the initial multi-scale multi-head self-attention GCN model are adjusted according to the loss value to obtain an intermediate GCN model, so as to train the initial multi-scale multi-head self-attention GCN model until the loss value is less than the preset threshold, and the intermediate GCN model corresponding to the loss value less than the preset threshold is used as the target multi-scale multi-head self-attention graph convolutional neural network GCN model.

Each module in the above-mentioned fault classification device for rotating equipment can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in FIG11. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a fault classification method for rotating equipment is implemented.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG12. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a fault classification method for a rotating device is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, a trackball or a touch pad set on the housing of the computer device, or an external keyboard, touch pad or mouse, etc.

Those skilled in the art will understand that the structures shown in FIG. 11 and FIG. 12 are merely block diagrams of partial structures related to the scheme of the present application, and do not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figures, or combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

Extract features of fault information of the rotating equipment to be tested to obtain fault features of the fault information;

The fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is trained by the initial multi-scale multi-head self-attention GCN model based on the fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Extract features of fault information of the rotating equipment to be tested to obtain fault features of the fault information;

The fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is trained by the initial multi-scale multi-head self-attention GCN model based on the fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

Extract features of fault information of the rotating equipment to be tested to obtain fault features of the fault information;

The fault features are input into the target multi-scale multi-head self-attention graph convolutional neural network GCN model to obtain the fault category of the rotating equipment to be tested; the target multi-scale multi-head self-attention graph convolutional neural network GCN model is trained by the initial multi-scale multi-head self-attention GCN model based on the fused graph feature samples, the fused graph feature samples are obtained by fusing the target output information and multiple graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on the fault information samples, and the graph feature samples are determined based on the fault information samples, the adjacency matrix and the initial multi-scale multi-head self-attention GCN model.

In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and 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 embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

The technical features of the above embodiments may be arbitrarily combined. To make the description concise, 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, they should be considered to be within the scope of this specification.

The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims (10)

1. A method of fault classification for a rotating device, the method comprising:

extracting characteristics of fault information of the rotary equipment to be detected to obtain fault characteristics of the fault information;

Inputting fault characteristics into a target multi-scale multi-head self-attention force diagram convolutional neural network GCN model to obtain fault types of the rotating equipment to be tested; the target multi-scale multi-head self-attention force-seeking convolutional neural network GCN model is obtained by training an initial multi-scale multi-head self-attention GCN model based on a fusion graph feature sample, wherein the fusion graph feature sample is obtained by fusing target output information and a plurality of graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on a fault information sample, and the graph feature sample is determined by the fault information sample, an adjacent matrix and the initial multi-scale multi-head self-attention GCN model.

2. The method of claim 1, wherein the initial multi-scale multi-headed self-attention GCN model comprises a plurality of first sub-models and a multi-layer perceptron; the method further comprises the steps of:

Inputting fault information samples and adjacent matrixes corresponding to the first sub-models into the first sub-models aiming at the first sub-models to obtain graph feature samples output by the first sub-models;

inputting the fault information sample into a multi-layer sensor to obtain first output information; the target output information includes first output information;

Obtaining a fusion graph characteristic sample according to the first output information and each graph characteristic sample;

training the initial multi-scale multi-head self-attention GCN model according to the fusion map feature sample to obtain a target multi-scale multi-head self-attention force diagram convolutional neural network GCN model.

3. The method of claim 2, wherein the first sub-model comprises a first graph convolution layer and a second graph convolution layer, and the initial multi-scale multi-headed self-attention GCN model comprises a connection layer and a linear layer; obtaining a fusion graph feature sample according to the first output information and each graph feature sample, including:

Aiming at each first sub-model, inputting a fault information sample and an adjacent matrix corresponding to each first sub-model into a first graph convolution layer of the first sub-model to obtain a first intermediate graph characteristic;

Inputting first intermediate graph features obtained by the first graph convolution layers of the first sub-models into the connecting layer to obtain second intermediate graph features;

inputting the second intermediate graph characteristics into a linear layer to obtain second output information;

obtaining a fusion graph characteristic sample according to the first output information, the second output information and each graph characteristic sample; the target output information includes first output information and second output information.

4. A method according to claim 3, wherein a fused graph feature sample is obtained from the first output information, the second output information and each graph feature sample; the target output information includes first output information and second output information, including:

Obtaining the first output information, the second output information and the weighted value of each graph feature sample according to the first output information, the second output information, the first weight corresponding to the first output information, the second weight corresponding to the second output information and the third weight corresponding to each graph feature sample;

and obtaining a fusion map feature sample based on the weighted value.

5. The method according to any one of claims 2-4, wherein the number of fusion map feature samples is a plurality; the initial multi-scale multi-head self-attention GCN model comprises a multi-head self-attention fusion model and a full connection layer; training an initial multi-scale multi-head self-attention GCN model according to the fusion map feature sample to obtain a target multi-scale multi-head self-attention force diagram convolutional neural network GCN model, wherein the method comprises the following steps:

Connecting the fusion map feature samples by utilizing a multi-head self-attention fusion model to obtain map feature vectors;

obtaining a predicted fault category based on the graph feature vector by using the full connection layer;

And training the initial multi-scale multi-head self-attention GCN model according to the error between the predicted fault category and the real fault category corresponding to the fault information sample to obtain a target multi-scale multi-head self-attention force diagram convolutional neural network GCN model.

6. The method of claim 5, wherein training the initial multi-scale multi-headed self-attention GCN model to obtain the target multi-scale multi-headed self-attention seeking convolutional neural network GCN model based on errors between the predicted fault category and the true fault category corresponding to the fault information samples, comprises:

Determining a loss value according to an error between the predicted fault category and a real fault category corresponding to the fault information sample;

Under the condition that the loss value is larger than a preset threshold value, adjusting parameters of the initial multi-scale multi-head self-attention GCN model according to the loss value to obtain an intermediate GCN model, training the initial multi-scale multi-head self-attention GCN model until the loss value is smaller than the preset threshold value, and taking the intermediate GCN model corresponding to the loss value smaller than the preset threshold value as a target multi-scale multi-head self-attention power seeking convolutional neural network GCN model.

7. A fault classification device for a rotating apparatus, the device comprising:

the feature extraction module is used for extracting features of fault information of the rotary equipment to be detected to obtain fault features of the fault information;

The fault classification module is used for inputting fault characteristics into a target multi-scale multi-head self-attention force diagram convolutional neural network GCN model to obtain fault types of the rotating equipment to be detected; the target multi-scale multi-head self-attention force-seeking convolutional neural network GCN model is obtained by training an initial multi-scale multi-head self-attention GCN model based on a fusion graph feature sample, wherein the fusion graph feature sample is obtained by fusing target output information and a plurality of graph feature samples, the target output information is determined by the initial multi-scale multi-head self-attention GCN model based on a fault information sample, and the graph feature sample is determined by the fault information sample, an adjacent matrix and the initial multi-scale multi-head self-attention GCN model.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 6 when the computer program is executed.

9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, realizes the steps of the method of any one of claims 1 to 6.