A method for identifying rotor bar breakage faults in synchronous motors based on stator current.

Abstract

This invention provides a method for identifying rotor bar breakage faults in synchronous motors based on stator current. The method includes: acquiring stator current sample data of the synchronous motor during a rotor bar breakage fault, wherein the stator current sample data is obtained through current sensor acquisition and generative adversarial network (GAN); performing a d-q transform on the stator current sample data, followed by Fourier decomposition to obtain feature data; constructing a fault diagnosis model based on CNN and Transformer based on the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault category, wherein the hyperparameters of the fault diagnosis model are optimized using an optimization algorithm; and using the fault diagnosis model to diagnose the rotor bar breakage fault category. This invention improves the accuracy, efficiency, and generalization ability of synchronous motor rotor bar breakage fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of motor fault diagnosis technology, and in particular to a method for identifying the fault status of broken bars in the rotor of a synchronous motor based on stator current. Background Technology

[0002] As a core device for power transmission and energy conversion in modern industry, the operating status of electric motors directly affects production efficiency and safety. However, electric motors often fail due to various factors during operation, such as broken rotor bars, rotor eccentricity, and inter-turn short circuits. These failures ultimately lead to equipment damage and significant economic losses. Broken rotor bars account for approximately 10% of all failures. When this occurs, current cannot flow smoothly through the broken bar and instead concentrates on the remaining conductors, causing localized overheating and ultimately resulting in motor shutdown, impacting industrial production efficiency. Therefore, rapid and accurate identification of broken rotor bars in synchronous motors is crucial.

[0003] Current research methods for motor fault diagnosis can be broadly categorized into three types. The first involves establishing mathematical models of the motor under different operating conditions and analyzing the changes in motor operating parameters to determine its operational status. This method requires specialized knowledge from the personnel. The second method analyzes the impact of signals generated during motor operation on the motor's state. However, when motors face complex and changing environments and states, key characteristic quantities such as current and torque exhibit dynamic changes, leading to drawbacks such as low diagnostic efficiency, low accuracy, and difficulty in real-time monitoring. In recent years, artificial intelligence has brought new solutions to various industries, and the adoption of AI algorithms has become a research hotspot. Numerous methods combining AI with fault diagnosis have been proposed. Commonly used data preprocessing methods include Hilbert series transforms and empirical mode decomposition (EMD). Hilbert series transforms involve complex integration and convolution operations, and the computational complexity increases significantly when the signal length is long. When the signal frequencies are similar, EMD suffers from mode aliasing problems. Most existing literature relies on small samples, which may not fully represent the distribution and characteristics of the overall data, limiting their generalization ability on new data and leading to poor performance in practical applications. Furthermore, regardless of whether the data comes from real-world scenarios or laboratory collection, there is a data imbalance problem, which severely impacts the accuracy of diagnostic models. Numerous models have been proposed, such as Support Vector Machines (SVM) and Long Short-Term Memory (LSTM). When data is imbalanced, SVM classification results can be biased, and its performance largely depends on feature engineering. When processing sequential data, LSTM consumes significant computational resources due to its recurrent structure. For longer sequences, the vanishing gradient problem may prevent the model from learning long-range information.

[0004] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0005] (1) Existing technologies use Hilbert series decomposition algorithms, which involve a large number of convolution operations or Fourier transform operations, resulting in high computational complexity. Especially in situations where large amounts of data are processed or real-time requirements are high, they cannot meet the needs of rapid diagnosis.

[0006] (2) Existing technologies are affected by imbalances in the data collection process, and their performance is overly dependent on manual feature engineering. Summary of the Invention

[0007] To address the aforementioned problems in existing technologies, this invention provides a method for identifying rotor bar breakage fault states in synchronous motors based on stator current, thereby improving the accuracy, efficiency, and generalization ability of synchronous motor rotor bar breakage fault diagnosis. Its advantages include simple calculation, high accuracy, low computational resource consumption, and practical feasibility.

[0008] This invention provides a method for identifying rotor bar breakage fault status in a synchronous motor based on stator current, comprising:

[0009] The stator current sample data of the first synchronous motor under rotor bar breakage fault is obtained. The stator current sample data is obtained by collecting current sensor data and generating adversarial network.

[0010] After performing dq transform on the stator current sample data, Fourier decomposition is then performed to obtain the feature data.

[0011] Based on the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault category, a fault diagnosis model based on CNN and Transformer is constructed, and the hyperparameters of the fault diagnosis model are optimized using an optimization algorithm.

[0012] After performing dq transformation on the current stator current data of the second synchronous motor, Fourier decomposition is then performed to obtain feature data. The feature data of the current stator current data is then input into the fault diagnosis model to obtain the rotor bar breakage fault category output by the fault diagnosis model.

[0013] According to the present invention, a method for identifying the rotor bar breakage fault state of a synchronous motor based on stator current is provided, which acquires stator current sample data of a first synchronous motor when a rotor bar breakage fault occurs, including:

[0014] The stator current sample data of the first synchronous motor were obtained when the rotor bar failure category was 1 broken bar, 2 broken bars, and 3 broken bars, respectively.

[0015] According to the present invention, a method for identifying the rotor bar breakage state of a synchronous motor based on stator current is provided. The method involves performing a dq transform on the stator current sample data, followed by Fourier decomposition to obtain feature data, including:

[0016] The stator current sample data i1 for each frequency component is phase-shifted twice, and the stator current sample data after the two phase shifts are i2 and i3, respectively;

[0017] For I = [i1, i2, i3] T Perform a dq transformation to obtain the d-axis component i. d and q-axis component i q ;

[0018] Extracting the d-axis component i using Fourier decomposition d DC component i da and q-axis component i q DC component i qa ;

[0019] Calculate the DC component i da and i qa The sum of squares is used to obtain the amplitude of each frequency component as the characteristic data.

[0020] According to the present invention, a method for identifying the rotor bar breakage fault state of a synchronous motor based on stator current is provided. The first synchronous motor generates current harmonic components when a rotor bar breakage fault occurs. The formula for the stator current sample data is as follows:

[0021]

[0022] Where n is a positive integer; A1, ω1, and θ1 are the fundamental amplitude, angular frequency, and initial phase angle of the stator current sample data, respectively; A2, ..., A n These are the amplitudes of each harmonic component of the stator current sample data; ω2, ..., ω n These are the angular frequencies of the harmonic components of the stator current sample data; θ2, ..., θ n It is the initial phase angle of each harmonic component of the stator current sample data.

[0023] According to the present invention, a method for identifying the rotor bar breakage state of a synchronous motor based on stator current is provided. The method involves performing a dq transform on the stator current sample data, followed by Fourier decomposition to obtain feature data, including:

[0024] Shifting the stator current sample data i1 of frequency component ω2 120° to the left in the time domain yields current i2, and shifting it 240° to the left yields current i3. i2 and i3 are expressed as:

[0025]

[0026] Where θ′ and θ″ are the phase differences between the fundamental components of i2 and i3 and i1, respectively; θ′ n and θ″ n These are the phase differences of each harmonic component in i2 and i3 relative to i1;

[0027] For I = [i1, i2, i3] T The formula for performing the dq transform is as follows, P 3 / 2 Here is the dq transformation matrix:

[0028]

[0029] Extracting the d-axis component i using Fourier decomposition d DC component i da and q-axis component i q DC component i qa The formula is as follows:

[0030]

[0031] The DC component i is calculated using the following formula. da and i qa The sum of squares is used to obtain the amplitude of the frequency component ω2 as the characteristic data:

[0032]

[0033] According to the present invention, a method for identifying the fault state of a synchronous motor rotor bar broken based on stator current is provided, wherein the generative adversarial network includes a generator and a discriminator;

[0034] The generator is used to generate new data based on the input random noise;

[0035] The discriminator is used to determine the likelihood that the new data belongs to real data;

[0036] The generator structure includes a fully connected layer, an activation function LeakyReLU, a first deconvolution layer, a second deconvolution layer, and a third deconvolution layer connected in sequence.

[0037] The discriminator consists of a first convolutional layer, a second convolutional layer, a Dropout layer, and a fully connected layer connected in sequence.

[0038] The present invention provides a method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current. The fault diagnosis model based on CNN and Transformer includes a CNN, a Transformer encoder and a classifier connected in sequence.

[0039] The CNN is used to extract local features of the feature data of the current stator current data;

[0040] The Transformer encoder is used to capture the global dependencies of the feature data of the current stator current data;

[0041] The classifier is used to output the rotor bar breakage fault category of the second synchronous motor.

[0042] According to the present invention, a method for identifying rotor bar breakage fault states of a synchronous motor based on stator current is provided. Based on the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault category, a fault diagnosis model based on CNN and Transformer is constructed. The hyperparameters of the fault diagnosis model are optimized using an optimization algorithm, including:

[0043] The feature data of the stator current sample data and the corresponding actual rotor bar breakage fault categories are used as samples to divide the data into a training set and a validation set.

[0044] Under the current hyperparameters of the fault diagnosis model, the fault diagnosis model is trained using the training set, and the trained fault diagnosis model is evaluated on the validation set to obtain the fitness value corresponding to the current hyperparameters.

[0045] If the fitness value corresponding to the current hyperparameter is better than the global optimal fitness value, then the global optimal fitness value is updated to the fitness value corresponding to the current hyperparameter, and the global optimal parameter corresponding to the global optimal fitness value is updated to the current hyperparameter.

[0046] After updating the current hyperparameters of the fault diagnosis model using an optimization algorithm, the fault diagnosis model continues to be trained, evaluated, and updated with globally optimal parameters until the maximum number of iterations is reached, and the globally optimal parameters are output.

[0047] The model trained on the fault diagnosis model using the training set under the globally optimal parameters will be used as the final fault diagnosis model.

[0048] According to the present invention, a method for identifying the rotor bar breakage fault state of a synchronous motor based on stator current is provided. The hyperparameters of the fault diagnosis model include the number of CNN layers, attention dimension, number of Transformer encoder layers, and number of multi-head attention heads.

[0049] According to the present invention, a method for identifying rotor bar breakage fault states of a synchronous motor based on stator current is provided, wherein the trained fault diagnosis model is evaluated on the validation set to obtain the fitness value corresponding to the current hyperparameter, including:

[0050] The cross-entropy loss of the validation set is used as the fitness value corresponding to the current hyperparameter.

[0051] This invention provides a method for identifying synchronous motor rotor bar breakage fault states based on stator current. By collecting the stator current during rotor bar breakage faults, performing a dq transform on the stator current, and then performing Fourier decomposition on the transform result to obtain feature data, this method can more directly and clearly reflect the current characteristics under motor operating conditions. The calculation process is relatively simple, improving diagnostic efficiency while ensuring accuracy. Furthermore, it utilizes GANs to generate synthetic data highly similar to real motor fault data, expanding the dataset size and balancing data categories. GAN-based data augmentation effectively alleviates the problem of data imbalance and generates high-quality data that meets model requirements. By deeply learning the intrinsic features of the data, the model's performance is improved. Generalization ability; As an end-to-end model, CNN-Transformer has the complementary advantages of local feature extraction and global dependency capture when processing one-dimensional time series data. It can adapt to various complex data patterns, reduce the reliance on manual feature engineering, and its parallel structure has higher efficiency and accuracy when processing large-scale datasets. The hyperparameters of the CNN-Transformer network model are globally optimized using optimization algorithms to find the optimal combination of model parameters, thereby improving the diagnostic performance and efficiency of the model. The augmented data generated by GAN is used together with the original data to train the CNN-Transformer model, which helps to enhance the model's sensitivity to fault features and improve accuracy. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0053] Figure 1 This is one of the flowcharts of the synchronous motor rotor bar breakage fault state identification method based on stator current provided by the present invention;

[0054] Figure 2 This is the second flowchart of the synchronous motor rotor bar breakage fault state identification method based on stator current provided by the present invention;

[0055] Figure 3 This is a schematic diagram of the fault diagnosis model based on CNN and Transformer in the synchronous motor rotor bar breakage fault state identification method based on stator current provided by the present invention.

[0056] Figure 4 This is a schematic diagram of the iterative optimal parameters in the synchronous motor rotor bar breakage fault state identification method based on stator current provided by the present invention;

[0057] Figure 5 This is a schematic diagram illustrating the effect of CNN-Transformer model training in the synchronous motor rotor bar breakage fault state identification method based on stator current provided by the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0059] Existing technologies generally use Hilbert series transforms to extract fault feature frequency components and then use SVM models for fault diagnosis. Hilbert series decomposition algorithms involve a large number of convolution or Fourier transform operations, resulting in high computational complexity. When the signal length is long, the computational complexity increases significantly, making it unsuitable for real-time diagnosis. Furthermore, SVM performance relies too heavily on manual feature engineering, and SVM classification results can be biased when the data is imbalanced.

[0060] The following is combined Figure 1 The present invention describes a method for identifying rotor bar breakage fault states in a synchronous motor based on stator current, comprising:

[0061] Step 101: Obtain stator current sample data of the first synchronous motor when the rotor bar is broken. The stator current sample data is obtained by collecting current sensors and generating adversarial networks.

[0062] Step 102: After performing dq transformation on the stator current sample data, Fourier decomposition is then performed to obtain feature data;

[0063] Step 103: Based on the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault category, construct a fault diagnosis model based on CNN and Transformer. The hyperparameters of the fault diagnosis model are optimized using an optimization algorithm.

[0064] Step 104: After performing dq transformation on the current stator current data of the second synchronous motor, Fourier decomposition is performed to obtain feature data. The feature data of the current stator current data is input into the fault diagnosis model to obtain the rotor bar breakage fault category output by the fault diagnosis model.

[0065] In the data acquisition and processing steps, the stator current of the synchronous motor under rotor bar breakage fault conditions is collected. The stator current undergoes a dq transform, and the transform result is then subjected to Fourier decomposition to obtain characteristic data. Hall effect current sensors can be used to collect the stator current under rotor bar breakage fault conditions. The dq transform is effective in electrical signal applications. As a linear transform, it does not alter key information in the original current signal and can more directly and clearly reflect the current characteristics under motor operating conditions. Furthermore, its calculation is simple, mainly involving algebraic and trigonometric function calculations, making it easy to implement and beneficial for real-time diagnosis.

[0066] In the data augmentation step based on GAN (Generative Adversarial Networks), GANs are used to augment data and balance data categories.

[0067] A fault diagnosis model based on a hybrid CNN and Transformer model is constructed. This model automatically extracts data features in an end-to-end manner, overcoming shortcomings such as weak generalization ability, poor robustness, and difficult maintenance. Furthermore, optimization algorithms are used to optimize hyperparameters such as the number of convolutional layers, channels, and attention heads. This achieves accurate classification of motor faults, offering advantages such as high accuracy, fast diagnosis speed, and strong robustness. The first and second synchronous motors can be the same motor or different motors; this embodiment does not impose such limitations. A complete flowchart is shown below. Figure 2 As shown.

[0068] This embodiment collects stator current under rotor bar breakage fault conditions in motors, performs dq transform on the stator current, and then performs Fourier decomposition on the transform result to obtain feature data. This can more directly and clearly reflect the current characteristics under motor operating conditions, while the calculation process is relatively simple, improving diagnostic efficiency while ensuring diagnostic accuracy. GAN is used to generate synthetic data highly similar to real motor fault data, expanding the dataset size and balancing data categories. GAN-based data augmentation can effectively alleviate the problem of data imbalance and generate high-quality data that meets model requirements. By deeply learning the intrinsic features of the data, the model's generalization ability is improved. CNN-Trans As an end-to-end model, GAN offers complementary advantages in processing one-dimensional time-series data through local feature extraction and global dependency capture. It can adapt to various complex data patterns, reduce reliance on manual feature engineering, and its parallel structure provides higher efficiency and accuracy when processing large-scale datasets. Optimization algorithms are used to globally optimize the hyperparameters of the CNN-Transformer network model, finding the optimal combination of model parameters to improve the model's diagnostic performance and efficiency. The augmented data generated by GAN is used together with the original data to train the CNN-Transformer model, which helps to enhance the model's sensitivity to fault features and improve accuracy.

[0069] Based on the above embodiments, this embodiment obtains stator current sample data of the first synchronous motor during rotor bar breakage faults, including:

[0070] The stator current sample data of the first synchronous motor were obtained when the rotor bar failure category was 1 broken bar, 2 broken bars, and 3 broken bars, respectively.

[0071] The stator current of the motor under three operating conditions—one broken bar, two broken bars, and three broken bars—is collected by a current sensor. After performing dq transformation on the stator current, Fourier decomposition is performed on the transformation result to extract the DC component to highlight the fault characteristics. Finally, the amplitude of the frequency component is obtained by summing the squares of the DC components as feature data.

[0072] Based on the above embodiments, this embodiment performs dq transform on the stator current sample data and then Fourier decomposition to obtain feature data, including:

[0073] The stator current sample data i1 for each frequency component is phase-shifted twice, and the stator current sample data after the two phase shifts are i2 and i3, respectively;

[0074] For I = [i1, i2, i3] T Perform a dq transformation to obtain the d-axis component i. d and q-axis component i q ;

[0075] Extracting the d-axis component i using Fourier decomposition d DC component i da and q-axis component i q DC component i qa ;

[0076] Calculate the DC component i da and i qa The sum of squares is used to obtain the amplitude of each frequency component as the characteristic data.

[0077] Based on the above embodiments, in this embodiment, the first synchronous motor generates current harmonic components when a rotor bar breakage fault occurs. The formula for the stator current sample data is as follows:

[0078]

[0079] Where n is a positive integer; A1, ω1, and θ1 are the fundamental amplitude, angular frequency, and initial phase angle of the stator current sample data, respectively; A2, ..., A n These are the amplitudes of each harmonic component of the stator current sample data; ω2, ..., ω n These are the angular frequencies of the harmonic components of the stator current sample data; θ2, ..., θn It is the initial phase angle of each harmonic component of the stator current sample data.

[0080] Based on the above embodiments, this embodiment performs dq transform on the stator current sample data and then Fourier decomposition to obtain feature data, including:

[0081] Shifting the stator current sample data i1 of frequency component ω2 120° to the left in the time domain yields current i2, and shifting it 240° to the left yields current i3. i2 and i3 are expressed as:

[0082]

[0083] Where θ′ and θ″ are the phase differences between the fundamental components of i2 and i3 and i1, respectively; θ′ n and θ″ n These are the phase differences of each harmonic component in i2 and i3 relative to i1;

[0084] For I = [i1, i2, i3] T The formula for performing the dq transform is as follows, P 3 / 2 Here is the dq transformation matrix:

[0085]

[0086] Extracting the d-axis component i using Fourier decomposition d DC component i da and q-axis component i q DC component i qa The formula is as follows:

[0087]

[0088] The DC component i is calculated using the following formula. da and i qa The sum of squares is used to obtain the amplitude of the frequency component ω2 as the characteristic data:

[0089]

[0090] The sum of squares of the DC components obtained by Fourier decomposition of the dq transform results in the amplitude of the frequency ω2 component, which is used as the fault identification feature spectrum data. The dq transform can convert arbitrary frequency components of the stator current i1 into DC components to highlight fault characteristics.

[0091] The stator current under rotor bar breakage fault conditions is collected using a Hall current sensor; the stator current is phase-shifted to obtain I; and I is transformed by dq to obtain i in formula (3). d and i qThe DC component i in formula (4) is obtained through Fourier decomposition. da and i qa The amplitude of this frequency component is obtained through formula (5);

[0092] The characteristic frequency amplitudes from 0 to 200 Hz were selected as the feature dataset and normalized. In this embodiment, rotor bar breakage faults are classified according to the number of broken rotor bars. One broken bar is classified as an early minor fault and marked as 0; two broken bars are classified as an early moderate fault and marked as 1; three broken bars are classified as an early severe fault and marked as 2. Table 1 shows a partial sample of the dataset, f n (n = 1, 2, ..., 200) represents the characteristic frequency.

[0093] Table 1 Partial Data Sample

[0094]

[0095] Based on the above embodiments, the generative adversarial network (GAN) in this embodiment includes a generator and a discriminator;

[0096] The generator is used to generate new data based on the input random noise;

[0097] The discriminator is used to determine the likelihood that the new data belongs to real data;

[0098] The generator structure includes a fully connected layer, an activation function LeakyReLU, a first deconvolution layer, a second deconvolution layer, and a third deconvolution layer connected in sequence.

[0099] The discriminator consists of a first convolutional layer, a second convolutional layer, a Dropout layer, and a fully connected layer connected in sequence.

[0100] A GAN model is constructed, consisting of a generator and a discriminator. The generator generates new data similar to the original data. Its input is random noise, and its output is the generated new data. The discriminator distinguishes the generated data from the original data. Its input is both the original data and the generated new data, and its output is either 0 or 1, indicating the probability that the data belongs to the real data.

[0101] The generator model can be structured as follows: the first layer is a fully connected layer with LeakyReLU activation function; the second layer is a deconvolution layer with a stride of 1; the third and fourth layers are also deconvolution layers with a stride of 2. The discriminator model can be structured as follows: first, two consecutive convolutional layers with 5×5 kernels and a stride of 2; the third layer introduces a Dropout layer; and finally, a fully connected layer.

[0102] Determine the training strategy and train the GAN model. In this case, the cross-entropy function is chosen as the loss function, the Adam algorithm is used as the optimizer, the initial learning rate is 0.0001, and the number of training epochs is 500.

[0103] After training, the best-performing generator model is invoked, and random noise is input into the generator to obtain new data samples. For example, if there are 2000 samples of each of the three types of data collected, a total of 10500 samples are obtained through this step.

[0104] This embodiment utilizes Generative Adversarial Network (GAN) technology to generate synthetic data that is highly similar to real motor fault data, thereby expanding the scale of the dataset and fully leveraging the model's learning capabilities.

[0105] Based on the above embodiments, the fault diagnosis model based on CNN and Transformer in this embodiment includes a CNN, a Transformer encoder and a classifier connected in sequence;

[0106] The CNN is used to efficiently extract local features from the feature data of the current stator current data;

[0107] The Transformer encoder is used to capture the global dependencies of the feature data of the current stator current data;

[0108] The classifier is used to output the rotor bar breakage fault category of the second synchronous motor, so as to achieve accurate identification and diagnosis of motor fault characteristics.

[0109] The fault diagnosis model structure based on CNN and Transformer is as follows: Figure 3 As shown, the model consists of three parts. The first part is a convolutional neural network. Each convolutional kernel in the convolutional layer is 3×3 in size and padded with 1. Finally, a max pooling layer with a convolutional kernel of 2×2 and a stride of 2 is used to perform downsampling operations to reduce the amount of computation.

[0110] The second part consists of the Transformer encoder structure. The self-attention mechanism considers all positions in the sequence at the same time when processing sequential data, which improves computational efficiency. Its encoder layer consists of two sub-layers: a multi-head self-attention mechanism and a feedforward neural network.

[0111] The third part is the classifier. It uses an average pooling layer to reduce the data dimensionality, and then uses a fully connected layer to map the features to the classification results to obtain the final prediction output. In this case, the dataset has three categories, and the output parameter of the fully connected layer is set to 3.

[0112] Based on the above embodiments, this embodiment constructs a fault diagnosis model based on CNN and Transformer according to the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault category. The hyperparameters of the fault diagnosis model are optimized using optimization algorithms, including:

[0113] The stator current sample data feature data and the corresponding actual rotor bar breakage fault categories are used as samples to divide the data into training set and validation set, and the optimization algorithm parameters are initialized.

[0114] Under the current hyperparameters of the fault diagnosis model, the fault diagnosis model is trained using the training set, and the trained fault diagnosis model is evaluated on the validation set to obtain the fitness value corresponding to the current hyperparameters.

[0115] If the fitness value corresponding to the current hyperparameter is better than the global optimal fitness value, then the global optimal fitness value is updated to the fitness value corresponding to the current hyperparameter, and the global optimal parameter corresponding to the global optimal fitness value is updated to the current hyperparameter.

[0116] After updating the current hyperparameters of the fault diagnosis model using an optimization algorithm, the fault diagnosis model is then trained, evaluated, and its global optimal parameters are updated.

[0117] Determine if the iteration has reached the termination condition. If the condition is met, stop the iteration and output the globally optimal parameters. If the number of iterations has not exceeded the maximum number of iterations, continue to the next step and increment the number of iterations by 1.

[0118] The model trained on the fault diagnosis model using the training set under the globally optimal parameters will be used as the final fault diagnosis model.

[0119] The optimal parameters of the fault diagnosis model are determined based on an optimization search algorithm. Algorithms such as Sparrow Search are used to globally optimize the model's hyperparameters, including attention dimension, number of network layers, and number of attention heads, to find the optimal combination of model parameters.

[0120] Based on the optimal parameter combination configuration model obtained from the optimization algorithm, the augmented data generated by GAN is used together with the original data to train the fault diagnosis model, thereby enhancing the model's sensitivity to fault features and improving the diagnostic accuracy.

[0121] The steps to derive the optimal parameter combination configuration model based on the optimization algorithm include:

[0122] Determine the fitness function and the model hyperparameters that need to be optimized. The cross-entropy loss of the validation set can be selected as the objective function. The optimized model parameters include the number of CNN layers, the number of attention dimensions, the number of encoder layers, and the number of multi-head attention heads.

[0123] Data partitioning and initialization of sparrow optimization algorithm parameters. The obtained dataset is divided into a training set and a validation set according to the given ratio. The training set is used to train the model, and the validation set is used to evaluate the model performance under the combination of optimization algorithm parameters. Initialize the sparrow positions and number, maximum number of iterations, number of training rounds, and upper and lower bounds. Each sparrow position represents a model parameter configuration, and the position dimension of an individual sparrow is the parameter dimension of the model optimization. In this case, the dataset is partitioned in an 8:2 ratio, and the values ​​of the number of sparrows, maximum number of iterations, and number of training rounds are set to 10, 15, and 15, respectively. Upper bound lb = [1, 32, 1, 2]; lower bound ub = [3, 128, 3, 3].

[0124] Calculate the fitness of the current sparrow. Input the current position of each sparrow (i.e., the parameter configuration of the network model) into the network model, train and calculate the error of the model under this configuration, which is used as the fitness value. In each iteration, record the sparrow with the best fitness in the sparrow group as the current global optimal solution;

[0125] Update the current globally optimal parameters. Compare the current sparrow's fitness with the globally optimal fitness. If the current sparrow's fitness is better, update the globally optimal solution and its corresponding parameters.

[0126] The sparrow's position is updated. The sparrow algorithm has three roles: discoverer, follower, and watcher. The discoverer's new position is determined by the following formula;

[0127]

[0128] in, Let represent the position of the i-th sparrow in the j-th dimension during the t-th iteration, α be a random number within (0,1], ST represent the safety value, Q be a normally distributed random number, L be a 1×d unit vector, R² represent the warning value, and iter max It represents the maximum number of iterations.

[0129] The new position of the followers is calculated using the following formula:

[0130]

[0131] in, Let X represent the worst individual in the t-th iteration. P Indicates the current optimal parameters, A + =A T (AA T ) -1 ;

[0132] The new position of the vigilant is calculated using the following formula:

[0133]

[0134] Where β is a random number that follows a normal distribution with a mean of 0 and a variance of 1. Let f represent the worst individual in the t-th iteration, where K is a random number in the range [-1, 1]. i f represents the current fitness value of the sparrow. g and f w These represent the current best fitness value and the current worst fitness value, respectively.

[0135] Evaluation and updating of the fitness function. After the sparrow's position is updated, the fitness of each sparrow's new position is re-evaluated. By training the network model, the loss function value is calculated and updated as the new fitness value.

[0136] Determine if the iteration has reached the termination condition. If the condition is met, stop the iteration and output the optimal network model parameters. If the number of iterations has not exceeded the maximum number of iterations, continue to the next step.

[0137] Update the sparrow's location information and iteration count. After each iteration, the optimal solution and fitness of the sparrow population are updated; the iteration count is incremented by 1, and the process proceeds to the step of calculating the fitness of the current sparrow. Figure 4 This is a schematic diagram illustrating the optimal parameters for the algorithm iteration in this case.

[0138] In this case, the fitness value converged to the minimum in the 9th iteration, at which point the fitness value was 0.1294. The corresponding hyperparameter combination was: 3 CNN layers, 78 attention dimensions, 2 encoder layers, and 2 multi-head attention heads.

[0139] The optimal parameter combination configuration model derived from the sparrow search algorithm is as follows: the CNN module has three convolutional layers, the encoder layer has 2 layers, the attention heads are set to 2, and the attention dimension is set to 78. The augmented data generated by the GAN is used together with the original data to train the fault diagnosis model. The cross-entropy loss function is chosen, as shown in the following formula:

[0140]

[0141] Where m is the number of samples, p(x) i Let q(x) be the true probability distribution of the sample. i ) represents the probability distribution predicted by the model.

[0142] Based on the above embodiments, the hyperparameters of the fault diagnosis model in this embodiment include the number of CNN layers, attention dimension, number of Transformer encoder layers, and number of multi-head attention heads.

[0143] Based on the above embodiments, this embodiment evaluates the trained fault diagnosis model on the validation set to obtain the fitness value corresponding to the current hyperparameters, including:

[0144] The cross-entropy loss of the validation set is used as the fitness value corresponding to the current hyperparameter.

[0145] The training effect of the CNN and Transformer hybrid model based on the sparrow search algorithm constructed in this embodiment is shown in the figure below. Figure 5 As shown in the table. Experimental verification confirms that this method has high accuracy, reaching 99.2%, and a short diagnosis time, with an average recognition speed of 0.35s per sample, proving that the method is feasible. Table 2 shows the model recognition results for some samples.

[0146] Table 2 Partial Samples

[0147]

[0148]

[0149] The expected benefits and commercial value of the technical solution of this invention after transformation are as follows:

[0150] 1. Improved Efficiency and Real-Time Performance of Synchronous Motor Fault Diagnosis: This method extracts fault features based on the stator current dq transform. Compared to Hilbert series transforms, wavelet transforms, and EMD (Empirical Mode Decomposition), the dq transform, as a linear transform, does not alter the key information in the original current signal, and can more directly and clearly reflect the current characteristics under motor operating conditions. Furthermore, the calculation process of the dq transform is relatively simple. This computational simplicity shortens the time required for fault diagnosis, enabling this method to improve diagnostic efficiency while ensuring accuracy, thus meeting the needs of industrial applications with high real-time requirements.

[0151] 2. Enhanced robustness of the motor fault diagnosis model: By employing GANs for data augmentation, the model can learn features from a large amount of data, helping to reduce overfitting to specific fault data. Furthermore, factors such as noise and sensor errors during motor operation can lead to instability in fault data. GANs generate fault samples with noise and variation, enabling the model to learn how to extract useful fault information from these disturbances during training. In practical applications, even with noisy or low-quality inputs, the model maintains high diagnostic accuracy.

[0152] 3. Market Competitive Advantage: Companies adopting synchronous motor rotor bar breakage fault identification methods based on stator current characteristic spectrum will gain a significant technological advantage in the market. Timely and accurate fault diagnosis not only improves the quality and reliability of motor products but also enables companies to lead their competitors in product upgrades and after-sales service, attracting more customers and expanding market share.

[0153] The technical solution of this invention fills a gap in the domestic and international industry: While some synchronous motor fault diagnosis methods exist in existing literature, most are limited to traditional signal processing techniques or based on small-sample learning, which to some extent restricts the accuracy and generalization ability of the diagnosis. This invention combines multiple techniques such as stator current dq transformation, GAN data augmentation, and deep learning to form a comprehensive fault diagnosis system. This method not only provides fast diagnosis and high fault identification accuracy, but also exhibits strong system robustness, effectively addressing complex and varied motor fault conditions. Furthermore, this method is applicable to different motors and various fault types causing current harmonics. For each specific motor, this method can automatically construct the optimal model structure that best matches its own characteristics.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current, characterized in that, include: The stator current sample data of the first synchronous motor under rotor bar breakage fault is obtained. The stator current sample data is obtained by collecting current sensor data and generating adversarial network. After performing dq transform on the stator current sample data, Fourier decomposition is then performed to obtain the feature data. Based on the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault category, a fault diagnosis model based on CNN and Transformer is constructed, and the hyperparameters of the fault diagnosis model are optimized using an optimization algorithm. After performing dq transformation on the current stator current data of the second synchronous motor, Fourier decomposition is then performed to obtain feature data. The feature data of the current stator current data is then input into the fault diagnosis model to obtain the rotor bar breakage fault category output by the fault diagnosis model. After performing a dq transform on the stator current sample data, Fourier decomposition is then performed to obtain feature data, including: The stator current sample data i1 for each frequency component is phase-shifted twice, and the stator current sample data after the two phase shifts are i2 and i3, respectively; right =[ , , ] T Perform a dq transformation to obtain the d-axis component i. d and q-axis component i q ; Extracting the d-axis component i using Fourier decomposition d DC component i da and q-axis component i q DC component i qa ; Calculate the DC component i da and i qa The sum of squares is used to obtain the amplitude of each frequency component as the characteristic data.

2. The method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current according to claim 1, characterized in that, Obtain stator current sample data of the first synchronous motor during rotor bar breakage fault, including: The stator current sample data of the first synchronous motor were obtained when the rotor bar failure category was 1 broken bar, 2 broken bars, and 3 broken bars, respectively.

3. The method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current according to claim 1, characterized in that, The first synchronous motor generates current harmonic components when a rotor bar breakage fault occurs. The formula for the stator current sample data is as follows: ； Where n is a positive integer; A1, ω1, and θ1 are the fundamental amplitude, angular frequency, and initial phase angle of the stator current sample data, respectively; A2, ..., A n These are the amplitudes of each harmonic component of the stator current sample data; ω2, ..., ω n These are the angular frequencies of the harmonic components of the stator current sample data; θ2, ..., θ n It is the initial phase angle of each harmonic component of the stator current sample data.

4. The method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current according to claim 3, characterized in that, After performing a dq transform on the stator current sample data, Fourier decomposition is then performed to obtain feature data, including: Shifting the stator current sample data i1 of frequency component ω2 120° to the left in the time domain yields current i2, and shifting it 240° to the left yields current i3. i2 and i3 are expressed as: ； in, and These are the phase differences between the fundamental components of i2 and i3 and i1, respectively. and These are the phase differences of each harmonic component in i2 and i3 relative to i1; right =[ , , ] T The formula for performing the dq transform is as follows, P 3 / 2 Here is the dq transformation matrix: ； Extracting the d-axis component i using Fourier decomposition d DC component i da and q-axis component i q DC component i qa The formula is as follows: ； The DC component i is calculated using the following formula. da and i qa The sum of squares is used to obtain the amplitude of the frequency component ω2 as the characteristic data: 。 5. The method for identifying the fault status of a synchronous motor rotor bar breakage based on stator current according to any one of claims 1-4, characterized in that, The generative adversarial network includes a generator and a discriminator; The generator is used to generate new data based on the input random noise; The discriminator is used to determine the likelihood that the new data belongs to real data; The generator structure includes a fully connected layer, an activation function LeakyReLU, a first deconvolution layer, a second deconvolution layer, and a third deconvolution layer connected in sequence. The discriminator consists of a first convolutional layer, a second convolutional layer, a Dropout layer, and a fully connected layer connected in sequence.

6. The method for identifying the fault status of a synchronous motor rotor bar breakage based on stator current according to any one of claims 1-4, characterized in that, The fault diagnosis model based on CNN and Transformer consists of a CNN, a Transformer encoder, and a classifier connected in sequence. The CNN is used to extract local features of the feature data of the current stator current data; The Transformer encoder is used to capture the global dependencies of the feature data of the current stator current data; The classifier is used to output the rotor bar breakage fault category of the second synchronous motor.

7. The method for identifying the fault status of a synchronous motor rotor bar breakage based on stator current according to any one of claims 1-4, characterized in that, Based on the feature data of the stator current sample data and the corresponding actual rotor bar breakage fault categories, a fault diagnosis model based on CNN and Transformer is constructed. The hyperparameters of the fault diagnosis model are optimized using optimization algorithms, including: The feature data of the stator current sample data and the corresponding actual rotor bar breakage fault categories are used as samples to divide the data into a training set and a validation set. Under the current hyperparameters of the fault diagnosis model, the fault diagnosis model is trained using the training set, and the trained fault diagnosis model is evaluated on the validation set to obtain the fitness value corresponding to the current hyperparameters. If the fitness value corresponding to the current hyperparameter is better than the global optimal fitness value, then the global optimal fitness value is updated to the fitness value corresponding to the current hyperparameter, and the global optimal parameter corresponding to the global optimal fitness value is updated to the current hyperparameter. After updating the current hyperparameters of the fault diagnosis model using an optimization algorithm, the fault diagnosis model continues to be trained, evaluated, and updated with globally optimal parameters until the maximum number of iterations is reached, and the globally optimal parameters are output. The model trained on the fault diagnosis model using the training set under the globally optimal parameters will be used as the final fault diagnosis model.

8. The method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current according to claim 7, characterized in that, The hyperparameters of the fault diagnosis model include the number of CNN layers, attention dimension, number of Transformer encoder layers, and number of multi-head attention heads.

9. The method for identifying the fault state of a synchronous motor rotor bar breakage based on stator current according to claim 7, characterized in that, The trained fault diagnosis model is evaluated on the validation set to obtain the fitness value corresponding to the current hyperparameters, including: The cross-entropy loss of the validation set is used as the fitness value corresponding to the current hyperparameter.

Images