Ship propulsion motor fault diagnosis method and device and readable storage medium

Abstract

The invention provides a ship propulsion motor fault diagnosis method and device and a readable storage medium, the ship propulsion motor adopts a permanent magnet synchronous motor, and the method comprises the steps of collecting normal operation data of the permanent magnet synchronous motor under different working conditions and abnormal data of turn-to-turn short circuit and excitation loss faults; overlapping and sampling the collected three-phase current signals to form a fixed-length sample, and converting the three-phase current signals of the sample into phase comparison current points in a gray image state to form a sample data set; a ship propulsion motor fault diagnosis neural network model is built, the model comprises an input layer, a feature extraction layer, a full connection layer and an output layer, an improved residual network module is arranged in the feature extraction layer, a channel attention mechanism and a pre-activation residual module are fused, and the feature learning ability of key fault features of the ship propulsion motor is enhanced; and based on the trained ship propulsion motor fault diagnosis neural network model, performing early turn-to-turn short circuit and excitation loss fault diagnosis of the permanent magnet synchronous motor.

Description

Technical Field

[0001] The present invention relates to the technical field of ship propulsion motor fault diagnosis, and particularly relates to a ship propulsion motor fault diagnosis method, device and readable storage medium. Background Art

[0002] In recent years, due to its high efficiency, high power density and low maintenance cost, the permanent magnet synchronous motor (PMSM) has gradually become the core component in the ship propulsion system. However, during the long-term operation of the motor, various faults such as inter-turn short circuit and demagnetization may occur. These faults not only affect the normal operation of the motor but may also lead to the paralysis of the ship propulsion system, which is directly related to the normal operation and navigation safety of the ship. Therefore, how to diagnose motor faults timely and accurately is crucial for preventing accidents, reducing downtime and lowering maintenance costs.

[0003] Currently, in the field of motor fault diagnosis, time-domain and frequency-domain feature analysis techniques are mainly used to analyze the monitoring signals of faulty equipment. Common methods include Fourier transform, empirical mode decomposition, wavelet transform, etc. Although these methods have significant diagnostic advantages in theory, there are still certain limitations in the actual process. Especially under variable speed and variable load conditions, the collected signals are extremely vulnerable to noise interference, which also brings great difficulties to fault identification.

[0004] Converting signals into image information is an effective way to solve the problem of noise interference. By converting the motor state characteristic signals into images, the signals originally affected by noise can be intuitively presented through visual changes such as rotation, scaling, blurring, and shifting of the images. The patent document with the application number CN202311664830 proposes a PMSM fault diagnosis method based on image color features. This method first obtains the original vibration signal of the motor and performs empirical mode decomposition. Then, the decomposed signal is converted into a symmetric dot matrix diagram of three RGB channels by using the Symmetrized Dot Pattern (SDP) technology. Finally, a DenseNet network model improved by integrating the scSE module is used to learn the shape and color features of the image to achieve fault diagnosis. Although the SDP technology performs well in processing vibration signals, such signals are often difficult to avoid noise interference in the actual operating environment of the motor. In contrast, current signals not only have stronger anti-interference ability but are also easier to collect, which makes current signals have great potential in motor fault diagnosis. However, how to effectively convert current signals into image information for easy analysis and identification of fault features is still a technical challenge.

[0005] In addition, the current PMSM fault diagnosis methods based on current signals have the disadvantages of low diagnostic sensitivity and low diagnostic efficiency. Deep learning has shown great potential in this regard. In particular, convolutional neural networks have become the most widely used effective method for feature extraction due to their characteristics such as sparse connections, weight sharing, and local features. However, traditional network models often have certain defects, such as insufficient feature extraction ability and poor generalization ability of the model. Therefore, how to improve the sensitivity and efficiency of diagnosing PMSM faults using current signals is also a current research hotspot. Summary of the Invention

[0006] In order to overcome the defects of the above-mentioned existing technologies, the present invention provides a fault diagnosis method, device, and readable storage medium for a ship propulsion motor. The method is a fault diagnosis method for a ship propulsion motor based on a phase-contrastive current dot pattern (PCCDP) and an improved residual network (IResNet), which can realize the conversion of three-phase current signals into images. On the basis of the pre-activated residual module, the improved residual network integrates a squeeze-and-excitation module to enhance the ability to capture key fault features and improve the efficiency of PMSM fault diagnosis.

[0007] To solve the above technical problems, the technical solution of the present invention is a fault diagnosis method for a ship propulsion motor. The ship propulsion motor uses a permanent magnet synchronous motor and includes the following processes: Collect the normal operation data of the permanent magnet synchronous motor under different working conditions, as well as the abnormal data of turn-to-turn short circuit and demagnetization faults. Perform overlapping sampling on the collected three-phase current signals to form fixed-length samples, and convert the three-phase current signals of each sample into phase-contrastive current dots in the grayscale image state to form a sample data set. Build a fault diagnosis neural network model for the ship propulsion motor. The network model structure includes an input layer, a feature extraction layer, a fully connected layer, and an output layer. Among them, an improved residual network module is set in the feature extraction layer. The improved residual network module integrates a channel attention mechanism and a pre-activated residual module to enhance the feature learning ability of key fault features of the ship propulsion motor. Based on the trained fault diagnosis neural network model for the ship propulsion motor, perform early turn-to-turn short circuit and demagnetization fault diagnosis on the permanent magnet synchronous motor.

[0008] Moreover, when collecting data of a permanent magnet synchronous motor under different working conditions, the working condition parameters are defined by the combination of torque and speed. The speed is set at two levels of 750 rpm and 1000 rpm, and the torque corresponds to four levels of 25%, 50%, 75% and 100% of the rated torque; the inter-turn short circuit fault is set at two fault levels of 5% and 10%; the field loss fault is based on 25% and 50% of the magnetic flux.

[0009] Moreover, an overlapping sampling of three-phase current data is performed by using an equal-interval sliding window data enhancement method. Each sample consists of 1024 data points, and the shift interval is 384; and each sample is converted into a corresponding phase-contrast current point image.

[0010] Moreover, the input layer performs preliminary feature extraction on the phase-contrast current point image input to the ship propulsion motor fault diagnosis neural network model, and generates a set of feature maps.

[0011] Moreover, the feature extraction layer includes 2 groups of improved residual layers, which further perform deep feature extraction on a set of feature maps obtained after preprocessing by the input layer.

[0012] Moreover, in the improved residual network module, the input feature map first passes through a batch normalization layer and enters the activation residual module of the channel attention mechanism. The feature map obtained by the channel attention mechanism and the original input feature map are multiplied by weight coefficients channel by channel, and the feature map with channel attention is output; then, through the activation function, it passes through the convolutional layer, batch normalization layer, activation function, and convolutional layer in sequence. The obtained feature map is connected with the original input feature map by residual connection, and finally, through a batch normalization layer and an activation function, the final output of the improved residual network module is obtained.

[0013] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the ship propulsion motor fault diagnosis method described above is implemented.

[0014] On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the ship propulsion motor fault diagnosis method described above is implemented.

[0015] On the other hand, the present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the ship propulsion motor fault diagnosis method described above is implemented.

[0016] The above solution realizes the fault diagnosis of permanent magnet synchronous motors. By introducing the phase-contrast current dot plot, the present invention provides a novel solution for the visualization of three-phase currents. At the same time, by adopting an improved residual network, it can adaptively extract key features in the dataset, significantly improving the accuracy and reliability of permanent magnet synchronous motor fault diagnosis. Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention designs a visualization method for three-phase currents, namely the phase-contrast current dot plot. This algorithm improves the traditional SDP algorithm. While realizing the visualization of current signals, it can also intuitively display current information. And this signal processing method can reduce noise interference, which is beneficial for the network to extract the characteristic signals of PMSM.

[0017] (2) The present invention designs a new residual network structure based on the pre-activated residual network module and introduces an attention mechanism. Through this mechanism, it can automatically adjust the importance of channels according to data features, enabling the model to focus on key feature channels, thereby obtaining better performance in various scenarios, ensuring faster fault diagnosis of PMSM and more comprehensive and accurate extraction of fault features.

[0018] (3) The present invention does not require manual feature analysis and extraction of the original data, reducing the cumbersome process of feature extraction in permanent magnet synchronous motor fault diagnosis. It can automatically extract the fault-related features from the original three-phase stator winding current signals through this network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to make the technical solutions of the present invention clearer, the drawings included in the document will be briefly described below. It should be clear that these drawings only represent several embodiments of the present invention. For those skilled in the relevant art, without additional creative work, they can completely derive other possible drawings based on these drawings.

[0020] Figure 1 It is a flowchart of fault diagnosis for a ship propulsion motor according to an embodiment of the present invention.

[0021] Figure 2 It is a phase-contrast current dot plot according to an embodiment of the present invention.

[0022] Figure 3 It is a structural diagram of a neural network model for fault diagnosis of a ship propulsion motor according to an embodiment of the present invention.

[0023] Figure 4 It is a diagram of an improved residual network module according to an embodiment of the present invention.

[0024] Figure 5 It is a curve graph of fault diagnosis accuracy according to an embodiment of the present invention.

[0025] Figure 6 This is the fault diagnosis confusion matrix diagram of the embodiment of the present invention.

[0026] Figure 7 This is the fault diagnosis t-SNE diagram of the embodiment of the present invention.

[0027] Figure 8 This is the schematic structural diagram of the electronic device provided by the embodiment of the present invention. Specific embodiments

[0028] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] A fault diagnosis method for a ship propulsion motor provided by the present invention has a diagnostic flow block diagram as Figure 1 shown. The figure includes four parts: data acquisition, sample generation, diagnostic model, and diagnostic process. Among them, data acquisition focuses on in-depth analysis of the faults of the PMSM and captures key operating data through precision sensors; sample generation is responsible for converting the collected data into grayscale images to construct a data set for model training and testing, ensuring the quality and applicability of the data; the diagnostic model uses an improved residual network, and this advanced neural network structure can more effectively learn data features and improve the accuracy of fault diagnosis; the diagnostic process is to carefully plan and execute the training and verification process of the model, and through iterative optimization and parameter adjustment, ensure the high performance and reliability of the model in practical applications; The fault diagnosis method for the ship propulsion motor in the embodiment specifically includes the following steps: S1. Data acquisition: Build a permanent magnet synchronous motor experimental platform, and use the data acquisition system to collect the normal operating data of the motor under different working conditions, as well as the abnormal data of inter-turn short circuit and demagnetization faults.

[0030] In the embodiment, the inter-turn short-circuit fault is set in phase B, and different severities of inter-turn short-circuit faults are realized by short-circuiting the lead points of different turns of the phase B coil; the loss-of-magnetism fault is experimentally collected by replacing motors with different fault degrees; the experiment collects three-phase current signals through a data acquisition system; to ensure the comprehensiveness of data collection under different working conditions, the working condition parameters are defined by the combination of torque and speed. The speed is set to two levels of 750 rpm and 1000 rpm, and the torque corresponds to four levels of 25%, 50%, 75%, and 100% of the rated torque; the inter-turn short-circuit fault is set to two fault degrees of 5% and 10%; the loss-of-magnetism fault is based on 25% and 50% of the magnetic flux; the sampling frequency is 10 KHz; the fault data sets used for training and testing are shown in Table 1;

[0031] S2. Sample generation: The collected three-phase current signals are oversampled to form fixed-length samples, and the three-phase current signals of each sample are converted into phase-contrast current points in the grayscale image state, and a sample data set is formed.

[0032] In the embodiment, the collected three-phase current signals are oversampled to form fixed-length samples, and each sample is converted into a PCCDP image in the grayscale image state, as Figure 2 shown: The data enhancement method of an equally spaced sliding window is used to oversample the three-phase current data. Each sample consists of 1024 data points, and the shift interval is 384.

[0033] The PCCDP technology is an improved current visualization method inspired by the SDP technology. The traditional SDP technology was originally an algorithm for processing time series that converts one-dimensional sequences into graphs and is commonly used for the visualization of vibration signals. However, the current signal is different from the vibration signal. It not only has three-dimensional time series information but also has strong periodicity. Therefore, how to effectively convert the current signal into image information for easy analysis and identification of fault characteristics is still a technical challenge, and the PCCDP technology just solves this problem. The present invention proposes to use the PCCDP technology to realize the fault diagnosis of ship propulsion motors.

[0034] The implementation process of the PCCDP technology is as follows: The polar coordinate plane is divided into six 60° sectors, and two-phase current signals are mapped in each sector; the three-phase current information is mapped onto two convex lobes and two concave lobes, and each phase current data point corresponds to a point on the lobe; therefore, in the polar coordinate system, each point maps four points with the same radius but different angles; for each sampling point in each phase current, its radius is expressed as:

[0035] In the formula, represents the th sampling point of each phase current signal, and represent the minimum and maximum values of the current at present. The angles of the convex pleated leaf and the concave pleated leaf are calculated with variables; The angle of the convex pleated leaf is expressed as:

[0036] In the formula, represents the angular domain gain factor, represents the time delay coefficient, represents the initial angle of plotting; The angle of the concave pleated leaf is expressed as:

[0037] As can be seen from the above, when drawing the PCCDP image, it is also necessary to set and two key parameters to ensure that the image can clearly display the current information and provide effective support for subsequent fault diagnosis research. First of all, the setting of the angular domain gain factor is crucial. In order to prevent signal overlap and considering the possible errors in the drawing process, it is necessary to set the range of within the range of [5, 28]. And is generally set within the range of [1, 10]. Secondly, in order to maximize the difference between the fault state images for subsequent diagnosis research, the hunter-prey optimization algorithm is used to optimize these two parameters, and the sum of the Euclidean distances of the pixel points in the PCCDP image of each fault state is used as the objective function of this algorithm. In the embodiment, the optimal parameters of the PCCDP image are obtained according to this principle: is 28, is 8.

[0038] S3. Build a fault diagnosis neural network model for ship propulsion motors: Based on the improved residual network module, a PMSM fault diagnosis model is established and built under the TensorFlow framework; the structure of this network model is mainly divided into four parts: input layer, feature extraction layer, fully connected layer, output layer; among them, the feature extraction layer uses an improved residual network module that integrates the SE attention mechanism (SE represents channels) and the pre-activation residual module, aiming to enhance the feature learning ability and adaptively learn the importance of each channel, so that the network can focus more on the identification of key features.

[0039] The present invention proposes to use an improved residual network module (hereinafter referred to as the improved residual network module) as the basic module for feature extraction of the network model. This module is based on the pre-activated residual block and integrates the SE channel attention mechanism, which enhances the network's ability to identify important fault features. At the same time, since the pre-activated residual block does not apply regularization even on the path of the most direct identity mapping of information propagation, overfitting is likely to occur. Therefore, a batch normalization layer needs to be added after the addition operation to regularize the signal output. Then, activation is performed through the ReLU layer to avoid overfitting and improve the model's feature learning ability. In the embodiment, a ship propulsion motor fault diagnosis neural network model as shown in Figure 3 is built. The network model structure is mainly divided into four parts: an input layer, a feature extraction layer, a fully connected layer, and an output layer. Among them, the activation function is uniformly ReLU, and it specifically includes the following sub-steps: S301, input layer: Input the PCCDP image into the network model, and preliminarily extract the features of the image, and generate a set of feature maps after the operation of this layer; as shown in the structure from Conv1 to BN in Figure 3 , there are a convolutional layer Conv1 and a batch normalization layer BN in sequence. Among them, Conv1 is 16 1×1 convolutional kernels.

[0040] S302, feature extraction layer: Use 2 groups of improved residual layers to further perform deep feature extraction on a set of feature maps after preprocessing by the input layer; as shown in the structure from Conv2 to IResNet_Block2 in Figure 3 , there are a convolutional layer Conv2, an activation function ReLU, an improved residual network module IResNet_Block1, a convolutional layer Conv3, an activation function ReLU, and an improved residual network module IResNet_Block2 in sequence. That is, the first group of improved residual layers includes a convolutional layer Conv2, an activation function ReLU, and an improved residual network module IResNet_Block1, and the second group of improved residual layers includes a convolutional layer Conv3, an activation function ReLU, and an improved residual network module IResNet_Block2. Among them, Conv2 is 32 1×1 convolutional kernels, Conv3 is 64 1×1 convolutional kernels, the convolutional layer in IResNet_Block1 is 32 convolutional kernels, and the convolutional layer in IResNet_Block2 is 64 convolutional kernels.

[0041] The network structure of the improved residual network module in the feature extraction layer is as shown in Figure 4As shown in the figure, in each improved residual network module IResNet_Block, the input feature map first passes through a batch normalization layer BN to stabilize the training process and improve the model performance. Subsequently, the data flows to the activation residual module containing the SE attention mechanism. In this module, the input feature map generates a 1×1×C feature map through global average pooling layer GAP. Then these feature maps are processed by two fully connected layers. The first fully connected layer FC1 uses the ReLU activation function, and the second fully connected layer FC2 uses the Sigmoid activation function to calculate the attention weights for each channel. Finally, the 1×1×C feature map obtained by the channel attention mechanism and the original input feature map are multiplied by the weight coefficients channel by channel to output the feature map with channel attention. Secondly, the output feature map of the SE attention mechanism module is activated by the ReLU activation function to increase non-linearity, and then sequentially passes through the convolutional layer Conv4, batch normalization layer BN, activation function ReLU, and convolutional layer Conv5 to further extract and process features. Then, the feature map processed by the convolutional layer Conv5 is connected to the original input feature map residually, and finally passes through a batch normalization layer BN and an activation function ReLU to obtain the final output of the improved residual network module. The convolutional layers in this module all use 5×5 convolutional kernels.

[0042] S303, Fully Connected Layer: Input the extracted feature map into the fully connected layer, and there is an activation function ReLU following each fully connected layer.

[0043] In the embodiment, the extracted feature map is input into the fully connected layers FC3, FC4, and FC5, and there is an activation function ReLU following each fully connected layer; as Figure 3 In the structure shown from FC3 to ReLU, there are sequentially a fully connected layer FC3, an activation function ReLU, a fully connected layer FC4, an activation function ReLU, a fully connected layer FC5, and an activation function ReLU. Among them, FC3 outputs 1024 data points, FC4 outputs 512 data points, and FC5 outputs 256 data points; S304, Output Layer: The output result will output the probability of each category through a SoftMax layer.

[0044] The main parameter settings of the neural network model structure in the embodiment are shown in Table 2.

[0045]

[0046] S4. The preprocessed dataset is divided into a training set and a test set in a ratio of 7:3. The training set is then used for the training of the neural network. During the training process, the Adam method combined with the cross-entropy loss function is used to train and adjust the parameters of the network model to achieve the best diagnostic effect. The training parameters are set as shown in Table 3.

[0047]

[0048] Among them, Batch size represents the number of data samples used to update the model parameters in each iteration during the training process; Epoch represents the number of times the entire training dataset is used to train the model; Learning rate represents a hyperparameter used to control the update step size of the model parameters in the optimization algorithm; Beta_1, Beta_2, and Epsilon are three hyperparameters used in the Adam optimization algorithm: Beta_1 is used to control the exponential decay rate of the first-order moment estimate, Beta_2 is used to control the exponential decay rate of the second-order moment estimate; Epsilon is a small constant used to prevent division by zero.

[0049] S5. The test set data is input into the trained neural network model for inspection. The application effect of the network model in the PMSM fault diagnosis of the ship propulsion system is evaluated from three aspects: accuracy, confusion matrix, and t-SNE clustering effect, which verifies that the improved residual network model has significant application potential in PMSM fault diagnosis. The present invention selects the CNN neural network for a comparative experiment with this method, and the experimental results are as Figure 5 shown. It can be seen from the figure that after 10 iterations, the training accuracy of IResNet is significantly higher than that of CNN, and it has started to converge after the 20th iteration, with a faster convergence speed. It tends to be stable after the 50th iteration, while the CNN model has not converged after 100 iterations. Figure 6 It is the confusion matrix of the fault classification model on the test set. Figure 6 Part (a) in it is the confusion matrix of the CNN model on the test set, Figure 6 and part (b) in it is the confusion matrix of the IResNet model on the test set. For the HDE fault, both the CNN model and the IResNet model can make correct diagnoses. For the AR fault state, the diagnostic results of the IResNet model are the same as those of the traditional CNN model. For the HL state, the number of misdiagnoses of the IResNet model is one more than that of the traditional CNN model. But for the LDE fault state, the number of misdiagnoses of the IResNet model is less than that of the traditional CNN model.

[0050] To more intuitively demonstrate the feature extraction and classification capabilities of the method proposed in the present invention, the t-SNE visualization method is used to reduce the dimension of the output of the last layer of the model, and the results are as follows Figure 7 shown. Figure 7 Part (a) in it is the t-SNE visualization graph of the prediction results of the CNN model on the test set, Figure 7 and part (b) in it is the t-SNE visualization graph of the prediction results of the IResNet model on the test set. It can be seen from the figure that the points of the same category in the IResNet model are clustered in an area, far from the points of other categories, with obvious boundaries, while there are many overlapping areas for two types of faults in the CNN. Finally, the accuracy rate of the improved residual network model in identifying faults reaches 98.5% In summary, the diagnostic method proposed in the present invention can not only visualize the three-phase current, but also the proposed improved residual neural network model can automatically extract the features of the three-phase current signal to accurately identify the inter-turn short circuit and demagnetization faults of the PMSM. The present invention can provide strong technical support for the stable operation of the PMSM in the ship propulsion system, and has important practical application value and scientific research significance.

[0051] In specific implementation, those skilled in the art can use software technology to realize the automatic operation of the above process. Correspondingly, if a ship propulsion motor fault diagnosis solution is provided, including a computer or a server, and the above process is executed on the computer or the server for ship propulsion motor fault diagnosis, it should also be within the protection scope of the present invention.

[0052] In specific implementation, the above process can be realized by computer software technology for automatic operation.

[0053] In another embodiment, there is also provided an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned ship propulsion motor fault diagnosis method.

[0054] In another embodiment, there is also provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the above-mentioned ship propulsion motor fault diagnosis method.

[0055] In another embodiment, there is also provided a computer program product, including a computer program, characterized in that the computer program realizes the above-mentioned ship propulsion motor fault diagnosis method when executed by a processor.

[0056] Figure 8 An example of the schematic physical structure of an electronic device is shown in Figure 8As shown in the figure, the electronic device may include: a processor, a communications interface, a memory, and a communication bus. Among them, the processor, the communications interface, and the memory complete communication with each other through the communication bus. The processor may call the logical instructions in the memory to execute the method for temporal reconstruction of the effective angular momentum of the Earth's fluid.

[0057] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts. Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments. The above is only the specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field of the present invention can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all of them should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for diagnosing faults of a ship propulsion motor, wherein the ship propulsion motor adopts a permanent magnet synchronous motor, and wherein: This includes the following processing: Collect normal operation data of permanent magnet synchronous motors under different working conditions, as well as abnormal data of turn-to-turn short circuit and demagnetization faults; The collected three-phase current signals are overlapped and sampled to form fixed-length samples, and the three-phase current signals of each sample are converted into phase contrast current points in a grayscale image state to form a sample data set; A neural network model for ship propulsion motor fault diagnosis is constructed. The network model structure includes an input layer, a feature extraction layer, a fully connected layer, and an output layer. An improved residual network module is set in the feature extraction layer. The improved residual network module integrates the channel attention mechanism and the pre-activation residual module to enhance the feature learning ability of the key fault characteristics of the ship propulsion motor. Based on the trained ship propulsion motor fault diagnosis neural network model, early turn-to-turn short circuit and demagnetization fault diagnosis of permanent magnet synchronous motor is carried out.

2. The method for diagnosing a fault of a ship propulsion motor according to claim 1, characterized in that: When collecting data on different operating conditions of permanent magnet synchronous motors, the operating parameters are defined by a combination of torque and speed. The speed is set to 750rpm and 1000rpm, and the torque corresponds to four levels of 25%, 50%, 75% and 100% of the rated torque; the turn-to-turn short-circuit fault is set to two fault levels of 5% and 10%; the demagnetization fault is based on 25% and 50% of the magnetic flux.

3. The ship propulsion motor fault diagnosis method according to claim 1 is characterized in that: The three-phase current data are overlapped and sampled using an equally spaced sliding window data enhancement method. Each sample consists of 1024 data points with a shift interval of 384. Each sample is converted into a corresponding phase contrast current point image.

4. The ship propulsion motor fault diagnosis method according to claim 1 is characterized in that: The input layer performs preliminary feature extraction on the phase contrast current point image input to the ship propulsion motor fault diagnosis neural network model, and generates a set of feature mapping diagrams.

5. The ship propulsion motor fault diagnosis method according to claim 1, characterized in that: The feature extraction layer includes two groups of improved residual layers, which further perform deep feature extraction on a group of feature maps obtained after preprocessing by the input layer.

6. The ship propulsion motor fault diagnosis method according to claim 1, characterized in that: In the improved residual network module, the input feature map first passes through the batch normalization layer and enters the activated residual module of the channel attention mechanism. The feature map obtained by the channel attention mechanism and the feature map of the original input are multiplied by the weight coefficient channel by channel, and the feature map with channel attention is output; then, through the activation function, and then through the convolution layer, batch normalization layer, activation function, and convolution layer in turn, the obtained feature map is residually connected with the feature map of the original input, and finally through a batch normalization layer and activation function, the final output of the improved residual network module is obtained.

7. An electronic device comprising 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 ship propulsion motor fault diagnosis method according to any one of claims 1 to 6 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the ship propulsion motor fault diagnosis method according to any one of claims 1 to 6 is implemented.

9. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the ship propulsion motor fault diagnosis method according to any one of claims 1 to 6 is implemented.

Images