Power machinery fault diagnosis method, device and equipment and storage medium

Abstract

The invention discloses a power machine fault diagnosis method, device and equipment and a storage medium, and relates to the technical field of fault diagnosis, and the method comprises the following steps: collecting a sound signal of a to-be-diagnosed power machine during operation, extracting a plurality of features of the sound signal, and converting the features into a plurality of feature maps; splicing the plurality of feature graphs along the width direction to obtain a spliced graph; inputting the spliced graph into a CNN-CBAM-based fault diagnosis model to obtain a model diagnosis result; and judging whether the power machine is a normal power machine or not according to a model diagnosis result. According to the method, various information features are extracted, and the fault identification accuracy of the power machine is greatly improved. And then the spliced graph is input to a CNN-CBAM-based fault diagnosis model, and fault identification is performed on the power machinery in combination with a convolutional neural network and attention mechanism method, so that the accuracy of fault identification is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method, device, equipment and storage medium for diagnosing power machinery faults. Background Art

[0002] Modern industrial power machinery has a complex structure. If a component fails, it will cause unnecessary economic losses and casualties. Therefore, it is necessary to perform fault diagnosis on power machinery.

[0003] Existing fault diagnosis technologies include methods based on vibration signals and temperature, but these methods all use contact measurement, which requires the deployment of sensors and makes signal acquisition inconvenient. There are also fault diagnosis methods based on infrared signals, but the cost of using this method for fault identification is high. Sound signal acquisition is simple and non-contact measurement is used, so fault diagnosis technology based on sound signals has received widespread attention from researchers at home and abroad.

[0004] At present, there are two mainstream methods used by domestic and foreign researchers for the study of abnormal detection of power machinery based on acoustic signals. One is to extract only a single acoustic signal feature, such as Fbank feature, MFCC feature or GFCC feature, and then optimize the network structure to achieve fault diagnosis of power machinery. This will lead to insufficient extraction of state information features, resulting in reduced classification accuracy and low accuracy of fault identification. The second is to classify the sound of power machinery by combining traditional feature fusion and convolutional neural network to achieve fault diagnosis, but this method only uses a single convolutional neural network method, resulting in reduced recognition accuracy. Summary of the invention

[0005] The present invention provides a power machinery fault diagnosis method, device, equipment and storage medium, which solves the problem of low fault identification accuracy caused by insufficient extraction of state information features and only using a single convolutional neural network in the existing power machinery abnormality detection method based on acoustic signals.

[0006] In a first aspect, the present invention provides a method for diagnosing a power machinery fault, comprising the following steps:

[0007] Acquiring acoustic signals of the power machinery to be diagnosed during operation, extracting multiple features of the acoustic signals and converting the multiple features into multiple feature graphs;

[0008] Splicing multiple feature maps along the width direction to obtain a spliced ​​map;

[0009] The spliced ​​graph is input into the fault diagnosis model based on CNN-CBAM to obtain the fault diagnosis results of the power machinery, including:

[0010] Receive the spliced ​​image based on two convolutional layers and extract features from the spliced ​​image;

[0011] Based on the CBAM layer, the extracted features are enhanced in channels and space;

[0012] Integrate the enhanced features based on the fully connected layer;

[0013] The integrated features are classified based on the Softmax layer to obtain the fault diagnosis results of the power machinery.

[0014] Preferably, the multiple features of the acoustic signal include filter group features, linear cepstral coefficient features, Mel cepstral coefficient features and filter group features.

[0015] Preferably, the step of extracting multiple features of the acoustic signal and converting the multiple features into multiple feature graphs comprises the following steps:

[0016] Perform fast Fourier transform on the acoustic signal to obtain a frequency domain signal;

[0017] Get the power spectrum of the frequency domain signal;

[0018] Filter the power spectrum to obtain the logarithmic energy;

[0019] Perform discrete cosine transform on the logarithmic energy to obtain coefficient features;

[0020] The coefficient features are visualized in the form of RGP images.

[0021] Preferably, before extracting the various features of the acoustic signal, the method further includes performing pre-emphasis, framing and windowing pre-processing on the acoustic signal.

[0022] Preferably, the convolution kernel size of the two convolution layers is 3×3, the number is 32 and 16 respectively, the Relu activation function is adopted, all zero padding is used and l2 regularization is added, the pooling kernel size is 2×2, after the first convolution layer, a batch normalization layer is added, and a Dropout layer is added after each pooling layer, and the value size is set to 0.4.

[0023] Preferably, before inputting the spliced ​​graph into the fault diagnosis model based on CNN-CBAM, the method further includes training the fault diagnosis model based on CNN-CBAM, wherein the training process includes:

[0024] Collect multiple sound signal samples when the power machinery is running, including normal power machinery sound signals and faulty power machinery sound signals, and label the corresponding sound signals, including normal sound labels and fault labels;

[0025] Divide the multiple acoustic signals into training set, validation set and test set in a ratio of 6:2:2;

[0026] Extract multiple features of each acoustic signal to obtain multiple feature maps, and splice the multiple feature maps along the width direction to obtain a spliced ​​map;

[0027] The fault diagnosis model is trained through multiple feature graphs of the training set and the validation set to obtain the initial weights of each layer of the fault diagnosis model;

[0028] Adjust the hyperparameters of the fault diagnosis model and observe the accuracy curve and loss curve of the model under different hyperparameter combinations.

[0029] Preferably, the fault diagnosis model is evaluated by a cross entropy loss function.

[0030] In a second aspect, the present invention provides a power machinery fault diagnosis device, comprising:

[0031] An acquisition module is used to acquire the acoustic signal of the power machinery to be diagnosed during operation, extract multiple features of the acoustic signal and convert the multiple features into multiple feature graphs;

[0032] A splicing module, used for splicing multiple feature maps along the width direction to obtain a spliced ​​map;

[0033] An input module is used to input the spliced ​​image into the fault diagnosis model based on CNN-CBAM to obtain the model diagnosis result;

[0034] A diagnosis module, used for judging whether the power machine is a normal power machine according to the model diagnosis result;

[0035] The input module comprises:

[0036] An extraction module, used for receiving the spliced ​​image based on two convolutional layers and performing feature extraction on the spliced ​​image;

[0037] Enhancement module, used to enhance the extracted features in channels and space based on the CBAM layer;

[0038] An integration module, used to integrate the enhanced features based on the fully connected layer;

[0039] The classification module is used to classify the integrated features based on the Softmax layer to obtain the fault diagnosis results of the power machinery.

[0040] In a third aspect, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned power machinery fault diagnosis method when executing the program.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program implements the above-mentioned power machinery fault diagnosis method when executed by a processor.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention first collects the acoustic signal of the power machinery to be diagnosed during operation, extracts multiple features of the acoustic signal and converts the multiple features into multiple feature graphs. The present invention does not adopt the traditional fault diagnosis based on one sound feature, but extracts multiple sound features, converts the multiple features into multiple feature graphs, and splices the multiple feature graphs into the same picture before inputting the model, converting the fault diagnosis work into image recognition work. The present invention extracts multiple information features, greatly improving the recognition accuracy of power machine faults. Then the spliced ​​graph is input into the fault diagnosis model based on CNN-CBAM, and the power machinery fault is identified by combining the convolutional neural network and the attention mechanism method, which effectively improves the accuracy of fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is a schematic diagram of the improved acoustic signal feature fusion process of the present invention;

[0046] Figure 2 It is a structural diagram of the CNN-CBAM fault diagnosis model of the present invention;

[0047] Figure 3 It is a flow chart of the fault diagnosis model training of the present invention;

[0048] Figure 4 The present invention is a fault diagnosis flow chart. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The present invention provides a method for diagnosing power machinery faults, specifically a method for diagnosing power machinery faults based on improved acoustic signal feature fusion and CNN-CBAM. Various sound signals are divided into training sets, validation sets and test sets, and multiple feature maps of the sound are spliced ​​in the width direction, labels corresponding to various faults are set respectively, and a CNN-CBAM model is constructed to classify the sound signals of the power machinery. The validation set and the training set are used during training, and the trained model is tested on the test set. The model parameters are adjusted according to the test results, and the batch size, learning rate and number of training rounds of the model are modified, so as to obtain the best fault diagnosis model. Reference Figure 4 , specifically including the following steps:

[0051] The first step is to collect the acoustic signal when the power machinery is running, extract various features of the acoustic signal and convert them into multiple feature maps, and splice the multiple feature maps along the width direction to obtain a spliced ​​map.

[0052] Reference Figure 1 , use the sound signal acquisition equipment to collect several sections of power machinery running sounds, these sounds include normal power machinery sounds and faulty power machinery sounds, every 0.5s as a sample, label the collected power machinery sounds, the labels include normal sound labels and fault labels, and then divide the collected power machinery sounds into training set, validation set and test set according to 6:2:2. Then use the sound feature extraction algorithm to extract the sound features, such as based on filter bank features (Fbank features), linear cepstral coefficient features (Lpcc features), Mel cepstral coefficient features (MFCC features) and Gammatone filter bank features (GFCC features), etc., take two or more of the features, the number of frames and dimensions of the extracted power machinery sounds need to be the same, then visualize the various features of the sound signal to obtain multiple feature maps. Multiple visualized images are tightly spliced ​​in the width direction to form a multi-feature spliced ​​image.

[0053] Take MFCC feature extraction as an example. It includes the following steps:

[0054] S1: Preprocessing of the collected sound signal. The preprocessing of the sound signal includes three steps: pre-emphasis, framing and windowing.

[0055] During the propagation of sound signals, the high-frequency part will be attenuated due to the influence of external environmental factors. Pre-emphasis is to increase the high-frequency component of the sound signal. Generally, a high-pass filter is used to pre-emphasize the sound signal. The pre-emphasis formula is as follows:

[0056] H(z)=1-μz -1

[0057] Wherein, μ represents the pre-emphasis coefficient, and its value is between 0.9375 and 0.97.

[0058] In order to better analyze the sound signal, it is necessary to frame the sound signal and divide a speech signal into several short-time signals. Generally, the overlapping segmentation method is used to frame the sound signal. The frame signal formula is as follows:

[0059] d=(lm) / n

[0060] m=cn

[0061] In the formula, l represents the length of the sound signal, c represents the frame length, d represents the number of frames obtained by framing, m represents the overlapping part of adjacent frames, and n represents the frame shift.

[0062] In order to increase the continuity of the left and right ends of each frame, it is generally necessary to perform a windowing operation on each signal frame in the preprocessing of the acoustic signal. Each frame can be multiplied by a window function. Commonly used window functions include Hamming window and rectangular window.

[0063] S2: Perform fast Fourier transform on the preprocessed acoustic signal.

[0064] After a series of preprocessing operations, the collected power machinery sound signal is obtained as a signal x(n) in the time domain. Then, the time domain signal is transformed by Fourier transform to obtain the frequency domain signal. The fast Fourier transform formula is as follows:

[0065]

[0066] S3: Find the power spectrum.

[0067] The power spectrum is shown below:

[0068] E(i)=|X(i)| 2

[0069] S4: Filtering is performed through a filter group.

[0070] The above power spectrum X(i) is passed through an m-dimensional filter bank. Assume that the power spectrum X(i) is passed through an m-dimensional Mel filter bank H m (i), the logarithmic energy of the Mel spectrum can be obtained as follows:

[0071]

[0072] S5: Calculate the logarithmic DCT.

[0073] The logarithmic energy obtained by the filter group is then subjected to discrete cosine transform to obtain the coefficient feature C(n). Since adjacent filters in the filter have overlapping parts, there is a certain correlation between the sound feature coefficients. The discrete cosine transform can reduce the correlation. The formula is as follows:

[0074]

[0075] S6: Multi-feature image stitching.

[0076] The above method for solving sound features is used to solve a variety of sound features. Assuming that the shapes of the obtained features are m×n, the features are presented in the form of RGB images using the visualization function, and the multiple feature images are closely spliced ​​in the width direction to remove the blank parts.

[0077] Traditional feature fusion methods mainly include splicing fusion, frequency domain mixing, time-frequency mixing, and embedded mixing. Traditional splicing fusion is performed by superimposing fusion on the convolution channel, while the present invention adopts the feature map splicing method to perform feature fusion, which is to splice the sound feature map before inputting it into the model, and tightly splice the image in the width direction. Then, one image contains multiple sound features, which can improve the recognition accuracy of power machine faults.

[0078] Step 2: Input the spliced ​​graph into the fault diagnosis model based on CNN-CBAM to obtain the fault diagnosis results of the power machinery.

[0079] CNN network has a good classification effect on the voiceprint signal of power machinery. In the convolutional neural network, the convolution layer and the pooling layer can effectively extract the features of the image, and the fully connected layer can effectively map the extracted features to each category. However, this traditional convolutional neural network model does not take into account the association between different feature maps, which will reduce the accuracy of the feature class, and then reduce the accuracy of machine fault diagnosis. The attention mechanism is a mechanism that imitates human visual attention. It can adjust the weights of each image feature. Through the attention mechanism, the network model can pay more attention to more important features. The attention mechanism includes channel attention mechanism, self-attention mechanism, pixel attention mechanism and spatial attention mechanism.

[0080] CBAM is a hybrid attention mechanism module proposed by Sanghyua et al. It is plug-and-play and consists of a channel attention mechanism and a spatial attention mechanism. CBAM includes a channel attention mechanism and a spatial attention mechanism. The channel attention mechanism (CAM) can focus on the relationship between the features of each channel and can focus on the more important parts of the input image. The workflow of the channel attention mechanism is: first, use global maximum pooling and global average pooling to extract channel dimension features from the input feature map, then send the results to the multi-layer perceptron (MLP) for learning, and finally merge the learning results, and then after the mapping process of the Sigmoid function, the channel attention weight matrix can be obtained. The formula is as follows:

[0081] M C (F)=σ(MLP(AvgPool(F))+MLP(MaxPool(F)))

[0082] Where F is the input feature, σ is the Sigmoid activation function, MLP is the shared fully connected layer, AvgPool is the average pooling, and MaxPool is the maximum pooling.

[0083] The spatial attention mechanism (Spatial Attention Mechanism, SAM) can focus on the input image containing rich information. The workflow of the spatial attention mechanism is: first, use global maximum pooling and global average pooling to extract spatial dimension features from the input feature map, then splice the results according to channels, and finally, perform convolution on the spliced ​​results, and use the Sigmoid activation function to obtain the spatial attention weight matrix. The formula is as follows:

[0084] M s =σ(f 7×7 ([AvgPool(F); MaxPool(F)]))

[0085] Where F is the input feature, σ is the Sigmoid activation function, and f 7×7 is a 7×7 convolution kernel, AvgPool is average pooling, and MaxPool is maximum pooling.

[0086] Reference Figure 2Therefore, in order to improve the recognition rate of the model, the present invention adds the CBAM module to the convolutional neural network to construct a CNN-CBAM model. Specifically, it includes two layers of convolutional pooling layers. The convolution kernel size of the two convolutional layers is 3×3, and the number is 32 and 16 respectively. The Relu activation function is used, and all zero padding is used. In addition, l2 regularization is added to prevent overfitting. The pooling kernel size is 2×2. After the first convolutional layer, a batch normalization layer is added, and a Dropout layer is added after each pooling layer. The value size is set to 0.4. The CBAM layer is added after the two convolutional pooling layers to enhance the network's channel and spatial feature representation capabilities, and then the output is sent to the Softmax layer.

[0087] The spliced ​​image obtained in the previous step is continuously passed through two layers of convolutional pooling layers for preliminary feature extraction, and then the output of the two layers of convolutional pooling layers is used as the input of the CBAM module to update the weight of the attention mechanism, and the output is pooled. Finally, it is transformed by the Flatten layer and flattened and integrated by the fully connected layer, and then sent to the Softmax layer to complete the classification of normal and faulty power machinery sounds. The training set and validation set can be used to perform preliminary training on the model, obtain the initial weights of each layer of the model, and then adjust the hyperparameters to observe the accuracy curve and loss curve of the model under different hyperparameter combinations.

[0088] To ensure that the CNN-CBAM model can better classify power machinery sounds, it is necessary to select the network's hyperparameters, such as learning rate, optimization parameters, batch size, and number of training rounds. The control variable method is used to select each parameter to achieve the best model performance.

[0089] The model of the present invention adopts the Adam optimization algorithm, which can realize adaptive adjustment of the historical gradient square and the actual parameter learning rate. The learning rate is one of the important parameters in deep learning. The size of the learning rate has an important influence on the training results. When the learning rate is set too small, the training process is slow and takes too long. When the learning rate is too large, the model may not converge or converge too slowly. Batch size is also an important factor affecting the training results. The appropriate batch size can optimize the training speed and performance of the model.

[0090] The fault diagnosis model training process of the present invention is as follows: Figure 3 First, input the training set into the network, initialize the weights of the network model, set the parameters of the network model, and then train the model. The criterion for the model to reach a better solution is: after multiple rounds of training, the model finally converges, and compares the accuracy curve and loss curve under each set of hyperparameters, so that the fault diagnosis model recognition effect is optimal.

[0091] The loss function is an important part of deep learning and can evaluate the quality of the model. The larger the loss function, the poorer the training effect of the model and the larger the error. The smaller the loss function, the higher the degree of fit of the model and the better the trained deep learning recognition effect. The fault diagnosis model of the present invention uses the cross entropy loss function to evaluate the model. The cross entropy function loss formula is as follows:

[0092]

[0093] In the formula, m is the number of samples, c is the number of categories of training data, 1{y (i)} represents the indicator function.

[0094] The third step: collect the sound signal of the power machinery to be diagnosed during operation, pre-process the collected sound data, perform fast Fourier transform, calculate the power spectrum, pass through the filter group, calculate the logarithmic DCT, and visualize the features and splice the feature image, and then input the obtained multiple feature splicing images into the fault diagnosis model to output the operating status of the power machinery, so that normal power machinery or faulty power machinery can be monitored in real time. The process of power machinery fault identification is as follows: Figure 3 shown.

[0095] This method extracts multiple acoustic signal features, splices them into multi-feature images, and inputs them into a deep learning model. It aims to transform the problem of power machinery sound signal recognition into an image recognition problem, combines convolutional neural networks and attention mechanisms, uses CNN models to improve the ability of feature extraction, and further combines CBAM modules to dynamically adjust weights according to feature images, which combines the advantages of a single feature while avoiding the disadvantages of insufficient representation.

[0096] Based on the same concept, the present invention also provides a power machinery fault diagnosis device, including a collection module, a splicing module, an input module and a diagnosis module.

[0097] The acquisition module is used to collect the acoustic signals of the power machinery to be diagnosed during operation, extract multiple features of the acoustic signals and convert the multiple features into multiple feature graphs.

[0098] The stitching module is used to stitch multiple feature maps along the width direction to obtain a stitching map.

[0099] The input module is used to input the splicing graph into the fault diagnosis model based on CNN-CBAM to obtain the model diagnosis results.

[0100] The diagnosis module is used to determine whether the power machine is a normal power machine according to the model diagnosis result.

[0101] The input module includes an extraction module, an enhancement module, an integration module and a classification module.

[0102] The extraction module is used to receive the spliced ​​image based on two convolutional layers and perform feature extraction on the spliced ​​image.

[0103] The enhancement module is used to enhance the extracted features in channels and space based on the CBAM layer.

[0104] The integration module is used to integrate the enhanced features based on the fully connected layer.

[0105] The classification module is used to classify the integrated features based on the Softmax layer to obtain the fault diagnosis results of the power machinery.

[0106] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned power machinery fault diagnosis method when executing the program.

[0107] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned power machinery fault diagnosis method is implemented.

[0108] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0109] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for diagnosing a power machinery fault, characterized in that: The following steps are involved: Acquiring acoustic signals of the power machinery to be diagnosed during operation, extracting multiple features of the acoustic signals and converting the multiple features into multiple feature graphs; Splicing multiple feature maps along the width direction to obtain a spliced ​​map; The spliced ​​image is input into the fault diagnosis model based on CNN-CBAM to obtain the model diagnosis result; Judging whether the power machine is a normal power machine according to the model diagnosis result; The splicing graph is input into the fault diagnosis model based on CNN-CBAM to obtain the model diagnosis result, including: Receive the spliced ​​image based on two convolutional layers and extract features from the spliced ​​image; Based on the CBAM layer, the extracted features are enhanced in channels and space; Integrate the enhanced features based on the fully connected layer; The integrated features are classified based on the Softmax layer to obtain normal power machinery sound signals and faulty power machinery sound signals.

2. A power machinery fault diagnosis method as claimed in claim 1, characterized in that: The various features of the acoustic signal include filter bank features, linear cepstral coefficient features, Mel cepstral coefficient features and filter bank features.

3. A power machinery fault diagnosis method as claimed in claim 2, characterized in that: The method of extracting multiple features of the acoustic signal and converting the multiple features into multiple feature graphs comprises the following steps: Perform fast Fourier transform on the acoustic signal to obtain a frequency domain signal; Get the power spectrum of the frequency domain signal; Filter the power spectrum to obtain the logarithmic energy; Perform discrete cosine transform on the logarithmic energy to obtain coefficient features; The coefficient features are visualized in the form of RGP images.

4. A power machinery fault diagnosis method as claimed in claim 1, characterized in that: Before extracting various features of the acoustic signal, the acoustic signal is also pre-processed by pre-emphasis, framing and windowing.

5. A power machinery fault diagnosis method as claimed in claim 1, characterized in that: The convolution kernel size of the two convolutional layers is 3×3, the number is 32 and 16 respectively, the Relu activation function is used, all zero padding is used and l2 regularization is added, the pooling kernel size is 2×2, and a batch normalization layer is added after the first convolutional layer, and a Dropout layer is added after each pooling layer, and the value size is set to 0.

4.

6. A power machinery fault diagnosis method as claimed in claim 1, characterized in that: Before inputting the spliced ​​image into the fault diagnosis model based on CNN-CBAM, the fault diagnosis model based on CNN-CBAM is trained. The training process includes: Collect multiple sound signal samples when the power machinery is running, including normal power machinery sound signals and faulty power machinery sound signals, and label the corresponding sound signals, including normal sound labels and fault labels; Divide the multiple acoustic signals into training set, validation set and test set in a ratio of 6:2:2; Extract multiple features of each acoustic signal to obtain multiple feature maps, and splice the multiple feature maps along the width direction to obtain a spliced ​​map; The fault diagnosis model is trained through multiple feature graphs of the training set and the validation set to obtain the initial weights of each layer of the fault diagnosis model; Adjust the hyperparameters of the fault diagnosis model and observe the accuracy curve and loss curve of the model under different hyperparameter combinations.

7. A power machinery fault diagnosis method as claimed in claim 1, characterized in that: The fault diagnosis model is evaluated by the cross entropy loss function.

8. A power machinery fault diagnosis device, characterized in that: include: An acquisition module is used to acquire the acoustic signal of the power machinery to be diagnosed during operation, extract multiple features of the acoustic signal and convert the multiple features into multiple feature graphs; A splicing module, used for splicing multiple feature maps along the width direction to obtain a spliced ​​map; An input module is used to input the spliced ​​image into the fault diagnosis model based on CNN-CBAM to obtain the model diagnosis result; A diagnosis module, used for judging whether the power machine is a normal power machine according to the model diagnosis result; The input module comprises: An extraction module, used for receiving the spliced ​​image based on two convolutional layers and performing feature extraction on the spliced ​​image; Enhancement module, used to enhance the extracted features in channels and space based on the CBAM layer; An integration module, used to integrate the enhanced features based on the fully connected layer; The classification module is used to classify the integrated features based on the Softmax layer to obtain the fault diagnosis results of the power machinery.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for diagnosing power machinery faults as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the power machinery fault diagnosis method described in any one of claims 1 to 7 is implemented.

Images