Induction motor stator turn-to-turn short circuit fault diagnosis method and device combined with recursive graph

Abstract

The present invention discloses a method and device for diagnosing stator turn-to-turn short-circuit faults in an induction motor, using a recursion graph. The method uses real-time acquisition of the three-phase stator current of a three-phase induction motor to generate a real-time recursion graph. Using the real-time recursion graph as input, a motor turn-to-turn short-circuit CNN diagnostic model is used to diagnose the motor's turn-to-turn short-circuit condition. The CNN diagnostic model is constructed by training a convolutional neural network based on a sample set of recursion graphs under various operating conditions, including turn-to-turn short-circuit ratios, loads, and power supply imbalances. This technical solution eliminates the effects of load fluctuations and power supply voltage imbalance during turn-to-turn short-circuit diagnosis, improving diagnostic accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor fault diagnosis, and more specifically relates to a method and device for diagnosing stator turn-to-turn short circuit faults of an induction motor combined with a recursive graph. Background Art

[0002] Induction motors, with their simple structure, durability, and low cost, are widely used in various industrial scenarios. Surveys show that approximately 30% to 40% of induction motor faults originate from stator winding interturn shorts. The currents induced by interturn shorts generate heat and lead to magnetic field imbalance in the motor, which in turn causes local hot spots and damages the motor's insulation system. Induction motor interturn shorts typically originate from minor shorts, but identifying minor interturn shorts within the 5% threshold remains a difficult problem. Detecting and identifying minor interturn shorts in induction motors at an early stage can prevent them from becoming catastrophic, significantly enhancing the safe operation of the motor itself and industrial processes. Signal analysis-based diagnosis of motor stator interturn shorts has been extensively researched, with Motor Current Signal Analysis (MCSA) being a widely used method. This method uses various signal processing methods, such as negative-sequence current, negative-sequence apparent impedance, and Park transform, to extract fault signatures for diagnosis. In recent years, there has been a surge in research on AI-based turn-to-turn short-circuit fault diagnosis. This approach, often combined with signal analysis, can effectively improve diagnostic accuracy. There are also reports on the use of multi-signal fusion analysis to diagnose stator winding turn-to-turn short-circuit faults. This multi-signal fusion approach can effectively mitigate the problems of missed and misdiagnosed faults typically associated with single-signal fault signatures, significantly improving diagnostic accuracy.

[0003] The current research methods are easily affected by load fluctuations and power supply voltage imbalance when diagnosing inter-turn short circuits, and the accuracy of judging minor inter-turn short circuits is not very high. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and device for diagnosing stator inter-turn short circuit faults of an induction motor in combination with a recursive diagram, which can eliminate the influence of load fluctuation and power supply voltage imbalance during inter-turn short circuit diagnosis and improve the accuracy of diagnosis.

[0005] In order to achieve the above object, the solution of the present invention is:

[0006] A method for diagnosing stator turn-to-turn short-circuit faults in induction motors based on recursive graphs.

[0007] Obtain the three-phase stator current of the three-phase induction motor in real time and obtain a real-time recurrence diagram;

[0008] Taking the real-time recursive graph as input, using the motor turn-to-turn short circuit CNN diagnosis model to diagnose the motor turn-to-turn short circuit state;

[0009] Among them, the motor inter-turn short-circuit CNN diagnostic model is constructed based on a multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations, and is trained using a convolutional neural network.

[0010] Obtain the three-phase stator current of the three-phase induction motor in real time and obtain the real-time recurrence diagram, including,

[0011] The three-phase stator current of the three-phase induction motor is obtained from the sensor in real time, and Park's transform and extended Park's transform are used to obtain two-dimensional dq current and one-dimensional Park's vector respectively.

[0012] The obtained one-dimensional Park's vector is processed to obtain an embedding space vector set, and a recursion matrix is ​​calculated. A real-time recursion graph is obtained based on the recursion matrix.

[0013] The obtained one-dimensional Park's vector is processed to obtain an embedding space vector set, and a recursion matrix is ​​calculated. Based on the recursion matrix, a real-time recursion graph is obtained, including:

[0014] Determine the embedding dimension m and the time delay τ according to the one-dimensional Park's vector, and construct an embedding space vector constituting a vector set;

[0015] A recursion matrix RR is calculated for the vector set, and the obtained recursion matrix is ​​plotted in a two-dimensional graph to obtain a real-time recursion graph.

[0016] The construction method of the multi-operating condition recursive graph sample set under the above-mentioned different turn-to-turn short-circuit ratios, different loads, and power supply imbalance combinations is as follows:

[0017] Simulate the operating conditions of a three-phase induction motor under different turn-to-turn short-circuit ratios, loads, and supply voltages, and collect the corresponding three-phase stator current data;

[0018] Preprocessing the three-phase stator current data to transform the three-phase stator current into a two-dimensional dq current and a one-dimensional Park's vector;

[0019] The recursion graph is used to convert the one-dimensional Park's vector time series signal into a two-dimensional image, and the two-dimensional images corresponding to different operating conditions constitute the multi-operating condition recursion graph sample set.

[0020] Simulate the operating conditions of the three-phase induction motor under different inter-turn short-circuit ratios, loads and supply voltages, and collect the corresponding three-phase stator current data, including simulating the operating conditions of the induction motor under normal operating conditions and collecting stator current data under different loads; simulating the operating conditions of the induction motor during an inter-turn short-circuit fault with a short-circuit ratio of less than 5%, and collecting stator current data under different loads; simulating the operating conditions of the induction motor when it is connected to an unbalanced voltage, and collecting stator current data under different loads.

[0021] A CNN diagnostic model for motor turn-to-turn short circuit is constructed using convolutional neural network training, including the use of optimization methods to optimize the hyperparameters of the CNN diagnostic model for motor turn-to-turn short circuit; an optimization problem is established with the learning rate, L-2 regularization factor and momentum as optimization variables and the classification objective function as the optimization goal, and the BO (Bayesian Optimization) algorithm is used to optimize the hyperparameters.

[0022] The framework of the above convolutional neural network includes 4 convolutional layers, 3 pooling layers, 2 fully connected layers, 1 Dropout layer and a Softmax layer. In the convolutional layer, the convolution kernel size is set to 4×4, and the number of convolution kernels is set to 64, 32, 16 and 8 respectively. The stride of each CL is set to 1, and the ReLu activation function is used in the convolutional layer; in the pooling layer, the pooling area is set to 2×2, and the stride is set to 2; in the Dropout layer, the Dropout Rate is set to 0.5.

[0023] A device for diagnosing stator turn-to-turn short circuit faults of an induction motor combined with a recursive graph, comprising:

[0024] A real-time recurrence graph acquisition module is configured to obtain a real-time recurrence graph by acquiring three-phase stator currents of the three-phase induction motor in real time; and

[0025] a turn-to-turn short circuit state diagnosis module configured to take the real-time recursive graph as input and utilize a motor turn-to-turn short circuit CNN diagnosis model to diagnose the motor turn-to-turn short circuit state;

[0026] Among them, the motor inter-turn short-circuit CNN diagnostic model is constructed based on a multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations, and is trained using a convolutional neural network.

[0027] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the steps of the above-mentioned method for diagnosing stator turn-to-turn short circuit faults of an induction motor combined with a recursion graph are implemented.

[0028] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing stator inter-turn short circuit faults of an induction motor combined with a recursive diagram are implemented.

[0029] After adopting the above scheme, the present invention has the following advantages:

[0030] (1) When constructing the diagnostic model, the present invention uses learning samples that cover most of the motor's operating conditions and takes into account load fluctuations and power supply imbalance interference, making the established CNN (convolutional neural network) diagnostic model suitable for online diagnosis of early-stage minor turn-to-turn short circuits in three-phase induction motors;

[0031] (2) The present invention transforms the three-phase stator current of the motor into two-dimensional dq current, one-dimensional Park's vector and two-dimensional recursion graph in sequence through data preprocessing, and uses the texture features of the recursion graph to enhance the characterization capability of the slight inter-turn short circuit of the motor;

[0032] (3) The present invention constructs a CNN learning model to extract the two-dimensional recursive graph features of three-phase motors in multiple working conditions, and performs classification model learning and training, which can improve the online diagnosis performance of slight short circuits between turns of induction motors. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is the overall workflow diagram of the present invention;

[0034] Figure 2 This is an example of the stator current of the three-phase induction motor under typical operating conditions targeted by the present invention;

[0035] Figure 3 This is an example PV curve diagram under typical operating conditions of a three-phase induction motor;

[0036] Figure 4 This is an example of a recurrence diagram for a three-phase induction motor under typical operating conditions;

[0037] Figure 5 It is the CNN model structure proposed by the present invention;

[0038] Figure 6 This is a flow chart of the present invention for diagnosing inter-turn short circuit of a three-phase induction motor. DETAILED DESCRIPTION

[0039] The technical solutions and beneficial effects of the present invention will be described in detail below with reference to the accompanying drawings.

[0040] The present invention provides a method and device for diagnosing stator turn-to-turn short circuit faults of an induction motor in combination with a recursive diagram. The basic route is as follows: Figure 1First, the three-phase stator current is obtained from the three-phase induction motor, and Park's transformation is used to convert the three-phase current into I in the dq coordinate system. d , I q , and then the two-dimensional current variable I d , I q A one-dimensional Park's vector is synthesized; the recurrence plot (RP) method is then used to convert the one-dimensional time series of Park's vectors into a two-dimensional image. Finally, the 2D image is processed using a CNN deep learning algorithm to ultimately obtain a CNN diagnostic model for motor turn-to-turn short circuits. This patent's identification of stator turn-to-turn short circuits in induction motors is divided into two phases: offline and online. In the offline phase, three-phase stator current data of a certain model of three-phase induction motor is experimentally obtained under normal, mild turn-to-turn short circuit, and unbalanced supply voltage conditions for subsequent processing. The experimental operating conditions of the motor should cover as many possible operating conditions as possible during field operation, and a large dataset should be obtained to establish the motor turn-to-turn short circuit diagnostic model. Once the offline modeling is complete, it can be put into online operation. However, it should be noted that this model is only applicable to turn-to-turn short circuit diagnosis of motors of the same model as the experimental motor; when applied to other models, there is a risk of misjudgment. During online turn-to-turn short circuit diagnosis, the motor's three-phase current is measured in real time, and a RP two-dimensional image is obtained through preprocessing and input into the CNN diagnostic model. The model can then diagnose the motor's turn-to-turn short circuit condition.

[0041] The present invention provides a method for diagnosing an induction motor stator turn-to-turn short circuit fault in combination with a recursive diagram, comprising the following steps:

[0042] Step 1: obtaining the three-phase stator current of the three-phase induction motor in real time to obtain a real-time recursive graph;

[0043] Step 2: using the real-time recursive graph as input, using the motor turn-to-turn short circuit CNN diagnostic model to diagnose the motor turn-to-turn short circuit state, including whether the turn-to-turn short circuit fault occurs or not;

[0044] The motor inter-turn short-circuit CNN diagnostic model is constructed based on a multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations, and is trained using a convolutional neural network.

[0045] In step 1, a real-time recursive graph is obtained based on the three-phase stator current of the three-phase induction motor obtained in real time by the following method:

[0046] First, the three-phase stator current of the three-phase induction motor is obtained from the sensor in real time, and the two-dimensional dq current and one-dimensional Park's vector are obtained by using Park's transform and extended Park's transform respectively.

[0047] The obtained Park's vectors are then processed to obtain an embedded space vector set, and a recursive matrix is ​​calculated. Based on the recursive matrix, a real-time recursive graph is obtained. In this step, the embedding dimension m and time delay τ are first determined based on the one-dimensional Park's vectors, and a vector set of embedded space vectors is constructed. The recursive matrix RR is then calculated for the obtained embedded space vector set. The obtained recursive matrix is ​​then plotted in a two-dimensional graph to obtain a real-time recursive graph. In particular, the embedding dimension m can be determined using the false neighbor method, and the time delay τ can be determined using mutual information. Binarization is not performed during the recursive matrix calculation to obtain a texture image for subsequent processing, thereby avoiding information loss caused by binarization.

[0048] Among them, the construction method of the multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations is as follows:

[0049] First, the operating conditions of a three-phase induction motor under different turn-to-turn short-circuit ratios, loads, and supply voltages are simulated, and corresponding three-phase stator current data is collected. The three-phase stator current data includes, but is not limited to, stator current data under different loads during normal operation of the induction motor, stator current data under different loads when the induction motor experiences a turn-to-turn short-circuit fault with a short-circuit ratio of less than 5%, and stator current data under different loads when the induction motor is connected to an unbalanced voltage. In particular, under each operating condition, the stator current must be collected after the three-phase induction motor enters a quasi-steady state. To ensure data accuracy, sampling may be repeated multiple times, for example, more than 15 times, under each operating condition.

[0050] Then, preprocessing the three-phase stator current data, thereby transforming the three-phase stator current into a two-dimensional dq current and a one-dimensional Park's vector;

[0051] A recursive graph is used to convert the one-dimensional time series signal of the Park's vector obtained above into a two-dimensional image. The recursive graph texture features are used to enhance the characterization capability of slight inter-turn short circuits in the motor, so as to realize the diagnosis of slight inter-turn short circuits under interference. The two-dimensional images corresponding to different operating conditions constitute the multi-condition recursive graph sample set.

[0052] In step 2, the motor turn-to-turn short-circuit CNN diagnostic model uses an end-to-end CNN framework to distinguish recurrence graphs under different operating conditions. During CNN model training, the number of training generations and the number of mini-batch samples are set to 50 and 60, respectively, and the Adam optimizer is used for model learning. An optimization method is used to optimize the hyperparameters of the CNN turn-to-turn short-circuit diagnostic model. An optimization problem is established with the learning rate, L-2 regularization factor, and momentum as optimization variables and a classification objective function as the optimization goal. The Bayesian Optimization (BO) algorithm is used to optimize the hyperparameters.

[0053] The CNN framework includes 4 convolutional layers (CLs), 3 pooling layers (PLs), 2 fully connected layers (FCL), 1 dropout layer, and a softmax layer. In the CLs, the convolution kernel size is set to 4×4, and the number of convolution kernels is set to 64, 32, 16, and 8 respectively. The stride dimension of each CL is set to 1. The Rectified Linear Unit (ReLu) activation function is used in the convolution layer. In the pooling layer, the pooling area is set to 2×2 and the stride is set to 2. A dropout layer is added after the last pooling layer to deal with the problem of model overfitting, and the dropout rate is set to 0.5. The output is finally transmitted to two cascaded fully connected layers for turn-to-turn short circuit classification. The number of hidden neurons in the last fully connected layer is C to distinguish C different classes.

[0054] The present invention also provides an induction motor stator turn-to-turn short circuit fault diagnosis device combined with a recursive graph, comprising:

[0055] A real-time recurrence graph acquisition module is configured to obtain a real-time recurrence graph by acquiring three-phase stator currents of the three-phase induction motor in real time; and

[0056] a turn-to-turn short circuit state diagnosis module configured to take the real-time recursive graph as input and utilize a motor turn-to-turn short circuit CNN diagnosis model to diagnose the motor turn-to-turn short circuit state;

[0057] Among them, the motor inter-turn short-circuit CNN diagnostic model is constructed based on a multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations, and is trained using a convolutional neural network.

[0058] The real-time recursive graph acquisition module includes:

[0059] a first conversion module configured to obtain three-phase stator currents of a three-phase induction motor from a sensor in real time, and sequentially use Park's transform and extended Park's transform to obtain two-dimensional dq currents and one-dimensional Park's vectors; and

[0060] The second conversion module is configured to process the obtained one-dimensional Park's vector to obtain an embedding space vector set, calculate a recursion matrix, and obtain a real-time recursion graph based on the recursion matrix.

[0061] The second conversion module includes:

[0062] A vector set construction module is configured to determine the embedding dimension m and the time delay τ according to the one-dimensional Park's vector, and construct an embedding space vector constituting a vector set; and

[0063] The real-time recurrence graph calculation module is configured to calculate the recurrence matrix RR for the vector set, and plot the obtained recurrence matrix in a two-dimensional graph to obtain a real-time recurrence graph.

[0064] The method for constructing the multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations is as follows:

[0065] Simulate the operating conditions of a three-phase induction motor under different turn-to-turn short-circuit ratios, loads, and supply voltages, and collect the corresponding three-phase stator current data;

[0066] Preprocessing the three-phase stator current data, thereby sequentially applying Park's transformation and extended Park's transformation to the three-phase stator current to obtain a two-dimensional dq current and a one-dimensional Park's vector;

[0067] The recursion graph is used to convert the one-dimensional Park's vector time series signal into a two-dimensional image, and the two-dimensional images corresponding to different operating conditions constitute the multi-operating condition recursion graph sample set.

[0068] Simulate the operating conditions of the three-phase induction motor under different inter-turn short-circuit ratios, loads and supply voltages, and collect the corresponding three-phase stator current data, including simulating the operating conditions of the induction motor under normal operating conditions and collecting stator current data under different loads; simulating the operating conditions of the induction motor during an inter-turn short-circuit fault with a short-circuit ratio of less than 5%, and collecting stator current data under different loads; simulating the operating conditions of the induction motor when it is connected to an unbalanced voltage, and collecting stator current data under different loads.

[0069] A CNN diagnostic model for motor turn-to-turn short circuit is constructed using convolutional neural network training, including the use of optimization methods to optimize the hyperparameters of the CNN diagnostic model for motor turn-to-turn short circuit; an optimization problem is established with the learning rate, L-2 regularization factor and momentum as optimization variables and the classification objective function as the optimization goal, and the BO algorithm is used to solve the hyperparameter optimization.

[0070] The framework of the convolutional neural network includes 4 convolutional layers, 3 pooling layers, 2 fully connected layers, 1 Dropout layer and a Softmax layer. In the convolutional layer, the convolution kernel size is set to 4×4, the number of convolution kernels is set to 64, 32, 16 and 8 respectively, and the step size of each CL is set to 1. The ReLu activation function is used in the convolutional layer; in the pooling layer, the pooling area is set to 2×2, the stride is set to 2; in the Dropout layer, the Dropout Rate is set to 0.5.

[0071] The following is a specific embodiment to illustrate the technical solution and technical effects of the present invention in detail, including the following steps:

[0072] Step 1: Acquiring three-phase induction motor operating data: This embodiment utilizes deep learning theory to diagnose motor turn-to-turn shorts, necessitating the acquisition of comprehensive offline learning samples to establish a turn-to-turn short diagnosis model. This invention addresses the problem of diagnosing minor motor turn-to-turn shorts, taking into account interference such as motor load and supply voltage imbalance. Therefore, the acquired data samples must encompass operating conditions such as normal, minor turn-to-turn shorts, variable load, and supply voltage imbalance.

[0073] In the above-mentioned method for diagnosing stator turn-to-turn short circuit of an induction motor combined with a recursive graph, in step 1, the specific operation method for constructing the learning sample is as follows:

[0074] Step 1.1: Build an experimental platform for a specific model of three-phase induction motor. The experimental platform needs to be able to simulate mild inter-turn short-circuit conditions, variable load conditions, and power supply imbalance conditions. The experimental platform also needs to be equipped with three-phase stator current detection, processing, and storage equipment, and the sampling frequency of the three-phase stator current is set to 20kHz. On the experimental platform, the motor inter-turn short-circuit simulation is achieved by short-circuiting the terminal connected to specific turns of the motor stator coil, and the power supply imbalance is achieved by adjusting the three-phase autotransformer. The experimental platform is not an innovative feature of the present invention and will not be described in detail here.

[0075] Step 1.2: Keep the three-phase motor running normally. Change the motor load through the experimental platform and collect and store the three-phase stator current of the motor under different loads. Repeat the three-phase stator current collection 15 times under each load. Data collection and storage must be performed only after the motor is running smoothly.

[0076] In step 1.3, interturn short-circuit faults of varying severity are simulated by selectively shorting the terminals leading from the stator coils. This invention focuses on diagnosing minor interturn shorts, so the number of simulated short-circuit turns in the experiment is controlled to less than 5% of the total number of coil turns. The three-phase currents under each interturn short-circuit fault are repeatedly collected 15 times, and current collection and storage are also required after the motor enters a quasi-steady state. The motor load is also adjusted for each interturn short-circuit fault. This allows for multiple combinations of interturn short-circuit ratios and load variations, resulting in a comprehensive learning sample set. These experimental operating condition combinations essentially cover the motor's operating conditions in actual industrial scenarios.

[0077] In step 1.4, the autotransformer on the motor test platform is adjusted to connect an unbalanced voltage to the three-phase motor. The motor load is then adjusted to obtain various operating conditions. When the motor enters a quasi-steady state, current data is collected and stored.

[0078] Step 2: Preprocessing of the three-phase current of the motor. In step 1, the three-phase current data under multiple working conditions were obtained by conducting multiple groups of experiments on the experimental platform. Figure 2 The three-phase stator current data of the motor under several typical working conditions are: Figure 2 (a) is the three-phase stator current under healthy no-load conditions, Figure 2 (b) is the three-phase stator current under 2.5% turn-to-turn short circuit and 60% load, Figure 2 (c) is the three-phase stator current under voltage unbalance condition; Figure 2 It can be seen that the current curves under different working conditions are different, which provides a basis for the diagnosis of inter-turn short circuits. However, the large number of variables makes the impact of inter-turn short circuits, load changes, and power supply imbalance on the three-phase current unclear, so further processing of the three-phase current data is necessary.

[0079] In the above-mentioned method for diagnosing stator turn-to-turn short circuit of an induction motor combined with a recursive graph, in step 2, the specific operation method of data preprocessing is as follows:

[0080] Step 2.1, Park's transformation. Park's transformation converts the motor variables in the three-phase coordinate system (ABC coordinate system) to the rotating coordinate system (dq coordinate system):

[0081]

[0082] Where, I d , I q are the d-axis and q-axis currents, I a , I b and I c are the three-phase stator currents respectively. Park's transformation transforms the three-phase currents into mutually perpendicular I d , I q , which simplifies the calculation and implementation process of three-phase motors.

[0083] Step 2.2, expand Park's vector construction. Step 2.1 obtains I by Park's transformation d , I q However, it is also cumbersome to analyze two currents at the same time to diagnose inter-turn short circuit. In this embodiment, the extended Park's transformation method is used to transform I d and I q Fused together to form the extended Park's Vector (PV):

[0084]

[0085] Figure 3 Shown are PV signals under several typical working conditions. Figure 3(a) is the PV signal under healthy no-load conditions, Figure 3 (b) is the PV signal under 2.5% turn-to-turn short circuit and 60% load. Figure 3 (c) is the PV signal under voltage unbalance condition; Figure 3 It can be seen that the non-stationary characteristics of the PV signal are strongly correlated with turn-to-turn short circuits, load changes, and voltage imbalance. The self-learning framework of this embodiment does not directly learn the PV signal to perform turn-to-turn short circuit diagnosis, but instead innovatively converts the PV signal into a two-dimensional recursive graph for processing.

[0086] Step 3: Construct a two-dimensional recurrence plot. Recurrence plots (RP) are an important method for analyzing the periodicity, chaos, and nonstationarity of time series. They can reveal the internal structure of time series and provide prior knowledge about similarity, information content, and predictability. The PV signal obtained above using equation (2) is a one-dimensional time series. This method converts it into a two-dimensional image and, through image processing, obtains more implicit features related to turn-to-turn short circuits.

[0087] In the above-mentioned method for diagnosing stator turn-to-turn short circuit of an induction motor in combination with a recursion graph, in step 3, the specific operation method of constructing the two-dimensional recursion graph is as follows:

[0088] Step 3.1, Embedding Space vector construction, suppose the time series of a PV is represented by x1,x2,…,x N , N is the number of samples of the PV time series, and let m be the embedding dimension and τ be the time delay, then the embedding space vector can be expressed as:

[0089] y k =(x k ,x k+τ ,x k+2τ ,…,x k+(m-1)τ ) (3)

[0090] In the formula, m and τ are determined by the false neighbor method and mutual information respectively. Through the above processing, we can obtain the vector set {y1,y2,y3,…,y M}, the number of vectors M is expressed as:

[0091] M=N-(m-1)τ (4)

[0092] Step 3.2: Recurrence Matrix Construction. The above obtained embedding space vector set consisting of M vectors, and the recurrence matrix is ​​calculated as follows:

[0093] RRi,j (μ)=H(μ-||y i -y j ||) (5)

[0095] Where H(*) represents the Heaviside function, ||·|| represents the Euclidean norm, and μ represents the distance threshold. The recursive matrix elements obtained in Equation (5) are binary numbers, which lose a lot of information. In this embodiment, binarization is not performed during the recursive matrix calculation to obtain a texture image for subsequent processing, avoiding information loss caused by binarization.

[0096] Step 3.3, recursion graph construction. Plot the recursion matrix obtained above in a two-dimensional graph to obtain the recursion graph. Figure 4 The following are the recursive diagrams obtained for a three-phase motor under several working conditions: Figure 4 (a) is the recurrence diagram under healthy no-load condition, Figure 4 (b) is the recurrence diagram under 2.5% turn-to-turn short circuit and 60% load. Figure 4 (c) is the recursive diagram under voltage unbalance condition; Figure 4 It can be seen that the recursive graph textures obtained under different working conditions of the three-phase motor have different characteristics, which provides a basis for the diagnosis of motor inter-turn short circuit. The inter-turn short circuit diagnosis model can be established by learning the recursive graphs under various working conditions.

[0097] Step 4: Construct a CNN-based turn-to-turn short-circuit diagnostic model. Through the above steps, we obtain recurrence graphs for various operating conditions, including different turn-to-turn short-circuit ratios, loads, and power supply imbalances. Each recurrence graph contains its own type label. This embodiment uses a CNN to learn from the recurrence graphs and construct a turn-to-turn short-circuit diagnostic model for load fluctuations and power supply imbalances.

[0098] In the above-mentioned method for diagnosing stator turn-to-turn short circuits in an induction motor combined with a recursive graph, in step 4, the specific operation method for constructing a CNN-based turn-to-turn short circuit diagnosis model is as follows:

[0099] Step 4.1, CNN diagnostic model framework. This embodiment constructs an end-to-end CNN framework to distinguish recursive graphs under different working conditions. The CNN framework includes 4 convolutional layers (CLs), 3 pooling layers (PLs), 2 fully connected layers (FCL), 1 Dropout layer, and a Softmax layer. In the CLs, the convolution kernel size is set to 4×4, and the number of convolution kernels is set to 64, 32, 16, and 8 respectively. The stride dimension of each CL is set to 1. The Rectified Linear Unit (ReLu) activation function is used in the convolution layer. In the pooling layer, the pooling area is set to 2×2, and the stride is set to 2. A Dropout layer is added after the last pooling layer to deal with the problem of model overfitting, and the Dropout Rate is set to 0.5. The output is finally transmitted to two cascaded fully connected layers for turn-to-turn short circuit classification. The number of hidden neurons in the last fully connected layer is C to distinguish C different classes. The CNN model framework of this patent is as follows Figure 4 As shown, where C is set to 14.

[0100] Step 4.2, CNN training. In this embodiment, when training the CNN model, the number of training epochs and the number of minibatch samples are set to 50 and 60 respectively, and the Adam optimizer is used for model learning.

[0101] Step 4.3, CNN model hyperparameter optimization. In CNN model learning and training, the learning rate, L-2 regularizer, and momentum parameters have a significant impact on model learning and training. Using random parameters or trial calculation methods cannot obtain the optimal hyperparameters. This embodiment uses an optimization method to optimize the hyperparameters of the CNN inter-turn short circuit diagnosis model. The optimization problem is expressed as:

[0102]

[0103] In the formula, h1-h3 represent the three hyperparameters of learning rate, L-2 regularization factor and momentum respectively, and cls_obf represents the classification objective function. By solving the optimization problem, the optimal hyperparameter combination of the CNN inter-turn short circuit diagnosis model can be obtained, and the best classification performance can be achieved under this combination. The optimization problem (6) includes the solution process of the CNNet(h) model, which cannot be solved by conventional optimization methods. This patent uses the Bayesian Optimization (BO) algorithm to solve the optimization problem (6). The BO algorithm is applicable to generalized objective functions. It only needs to specify the input and output, and does not need to know the internal structure and mathematical properties of the objective function. BO updates the posterior distribution of the objective function by continuously adding sample points until the posterior distribution basically fits the true distribution. That is, the BO algorithm takes into account the parameter information of the previous time when optimizing the solution, and can better adjust the current parameters. Hyperparameter optimization is not the innovation of the present invention, so the execution sub-steps of the BO algorithm will not be repeated here. Through steps 4.1-4.3, a CNN diagnostic model for inter-turn short circuit under load fluctuation and power supply imbalance is constructed, thus completing the offline modeling process of this patent.

[0104] This embodiment conducts experiments on a certain type of three-phase induction motor and establishes a CNN diagnostic model for slight inter-turn short circuit. The model also takes into account the influence of motor load fluctuation and power supply imbalance. The obtained diagnostic model covers most of the motor's operating conditions and can be applied online to the inter-turn short circuit diagnosis of the same model motor. The process of offline model construction and online diagnosis in this embodiment is as follows: Figure 6 shown.

[0105] Step 5: Online detection and preprocessing of the three-phase motor stator current. In the field application of the same model three-phase induction motor, the three-phase stator current is obtained from the sensor in real time, and the Park's transformation is performed using formula (1) to obtain I d , I q , and then process it according to formula (2) to obtain the PV time series;

[0106] Step 6: Real-time recurrence graph construction. Using the PV time series obtained in step 5, process it according to formula (3) to obtain the embedded space vector and construct a vector set. Then calculate the recurrence matrix corresponding to the embedded space vector set according to formula (5). It should be noted that no binarization is performed when calculating the recurrence matrix. Finally, the recurrence matrix obtained above is plotted in a two-dimensional graph to obtain a real-time recurrence graph.

[0107] Step 7: On-site turn-to-turn short-circuit fault identification. The real-time recursive graph obtained in Step 6 is fed into the trained CNN diagnostic model for motor turn-to-turn short-circuit. The model then diagnoses the motor's turn-to-turn short-circuit condition. Because the diagnostic model construction considers the effects of interference factors such as motor load fluctuations and power supply imbalance, the diagnostic model can eliminate these interferences when used online in real time.

[0108] The turn-to-turn short-circuit diagnostic function proposed in the present invention can only be enabled after the motor enters a quasi-steady state, and adopts a periodic call mode. On the other hand, the present invention is mainly aimed at early-stage minor turn-to-turn short circuits, and the real-time requirements are not high. The diagnostic cycle can be set to the minute level. After the offline learning and training of the CNN turn-to-turn short-circuit diagnostic model is completed, the amount of calculation required for on-site real-time call is relatively small. It can be implemented using a Raspberry Pi or other computing platforms, and the time required for calculation will be much less than the diagnostic cycle. It can be seen that the real-time performance of this model can be guaranteed.

[0109] In this embodiment, an apparatus for diagnosing stator turn-to-turn short circuits in an induction motor using the above-described recursive diagram includes: a three-phase induction motor experimental platform with adjustable turn-to-turn short circuit ratio, load, and supply voltage; a three-phase induction motor stator current sensor; a signal conditioning device for signal amplification and shaping; a signal acquisition unit; and a calculation unit.

[0110] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Persons skilled in the art may make various modifications, additions, or substitutions to the described specific embodiments without departing from the spirit of the present invention or exceeding the scope of the appended claims.

[0111] Although this invention frequently uses terms such as three-phase induction motor, stator current measurement signal, Prak's transform, recurrence graph, and CNN (convolutional neural network), the use of other terms is not excluded. These terms are used solely to more conveniently describe and explain the essence of this invention; interpreting them as any additional limitations would be contrary to the spirit of this invention.

[0112] An embodiment of the present invention further provides another computer device, comprising a processor and a memory configured to store a computer program that can be run on the processor; wherein, when the processor is configured to run the computer program, the method steps in the aforementioned embodiment are executed.

[0113] In practical applications, the processor includes a field-programmable gate array (FPGA), and the processor may be a central processing unit (CPU) or a digital signal processor (DSP). It is understood that for different devices, the electronic components used to implement the above-mentioned processor functions may also be other, and the embodiments of the present invention are not specifically limited thereto.

[0114] The above-mentioned memory can be a volatile memory (volatile memory), such as a random-access memory (RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (ROM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provides instructions and data to the processor.

[0115] In an exemplary embodiment, the present invention further provides a computer-readable storage medium for storing a computer program.

[0116] Optionally, the computer-readable storage medium can be applied to any one of the methods in the embodiments of the present invention, and the computer program enables the computer to execute the corresponding processes implemented by the processor in each method in the embodiments of the present invention. For the sake of brevity, they are not repeated here.

[0117] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0118] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0119] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0122] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional 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.

[0123] 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 such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for diagnosing stator turn-to-turn short circuit faults in an induction motor using a recursive graph, characterized by: Obtain the three-phase stator current of the three-phase induction motor in real time and obtain a real-time recurrence diagram; Taking the real-time recursive graph as input, using the motor turn-to-turn short circuit CNN diagnosis model to diagnose the motor turn-to-turn short circuit state; The motor turn-to-turn short-circuit CNN diagnostic model is constructed based on a multi-operating condition recursive graph sample set under different turn-to-turn short-circuit ratios, different loads, and power supply imbalance combinations, and is trained using a convolutional neural network. Among them, the three-phase stator current of the three-phase induction motor is obtained in real time, and a real-time recursive diagram is obtained, including: The three-phase stator current of the three-phase induction motor is obtained from the sensor in real time, and Park's transform and extended Park's transform are used to obtain the two-dimensional dq current and the one-dimensional Park's vector. The obtained one-dimensional Park's vector is processed to obtain an embedding space vector set, and a recursion matrix is ​​calculated, and a real-time recursion graph is obtained based on the recursion matrix; The obtained Park's vector is processed to obtain an embedding space vector set, and a recursion matrix is ​​calculated. A real-time recursion graph is obtained based on the recursion matrix, including: Determine the embedding dimension m and the time delay τ according to the one-dimensional Park's vector, and construct an embedding space vector constituting a vector set; A recursion matrix RR is calculated for the vector set, and the obtained recursion matrix is ​​plotted in a two-dimensional graph to obtain a real-time recursion graph. No binarization process is performed when calculating the recursion matrix.

2. The method according to claim 1, wherein: The method for constructing the multi-operating condition recursive graph sample set under different inter-turn short-circuit ratios, different loads, and power supply imbalance combinations is as follows: Simulate the operating conditions of a three-phase induction motor under different turn-to-turn short-circuit ratios, loads, and supply voltages, and collect the corresponding three-phase stator current data; Preprocessing the three-phase stator current data, thereby sequentially applying Park's transformation and extended Park's transformation to the three-phase stator current to obtain a two-dimensional dq current and a one-dimensional Park's vector; The recursion graph is used to convert the one-dimensional Park's vector time series signal into a two-dimensional image, and the two-dimensional images corresponding to different operating conditions constitute the multi-operating condition recursion graph sample set.

3. The method according to claim 2, wherein: Simulating the operating conditions of a three-phase induction motor under different turn-to-turn short-circuit ratios, loads, and supply voltages, and collecting corresponding three-phase stator current data, including simulating the operating conditions of the induction motor under normal operating conditions and collecting stator current data under different loads; The operating conditions of the induction motor when it is subjected to an inter-turn short-circuit fault with a short-circuit ratio of less than 5% are simulated, and stator current data under different loads are collected. The operating conditions of the induction motor when it is connected to an unbalanced voltage are simulated, and stator current data under different loads are collected.

4. The method according to claim 1, wherein: A CNN diagnostic model for motor turn-to-turn short circuit is constructed using convolutional neural network training, including the use of optimization methods to optimize the hyperparameters of the CNN diagnostic model for motor turn-to-turn short circuit; an optimization problem is established with the learning rate, L-2 regularization factor and momentum as optimization variables and the classification objective function as the optimization goal, and the BO algorithm is used to solve the hyperparameter optimization.

5. The method according to claim 4, wherein: The framework of the convolutional neural network includes 4 convolutional layers, 3 pooling layers, 2 fully connected layers, 1 Dropout layer and a Softmax layer. In the convolutional layer, the convolution kernel size is set to 4×4, the number of convolution kernels is set to 64, 32, 16 and 8 respectively, and the step size of each CL is set to 1. The ReLu activation function is used in the convolutional layer; in the pooling layer, the pooling area is set to 2×2, the stride is set to 2; in the Dropout layer, the Dropout Rate is set to 0.

5.

6. A device for diagnosing stator turn-to-turn short circuit faults in an induction motor in combination with a recursive diagram, characterized in that: include, A real-time recurrence graph acquisition module is configured to obtain a real-time recurrence graph by acquiring three-phase stator currents of the three-phase induction motor in real time; and a turn-to-turn short circuit state diagnosis module configured to take the real-time recursive graph as input and utilize a motor turn-to-turn short circuit CNN diagnosis model to diagnose the motor turn-to-turn short circuit state; The motor turn-to-turn short-circuit CNN diagnostic model is constructed based on a multi-operating condition recursive graph sample set under different turn-to-turn short-circuit ratios, different loads, and power supply imbalance combinations, and is trained using a convolutional neural network. The real-time recursive graph acquisition module acquires the three-phase stator current of the three-phase induction motor in real time to obtain a real-time recursive graph, including: The three-phase stator current of the three-phase induction motor is obtained from the sensor in real time, and Park's transform and extended Park's transform are used to obtain the two-dimensional dq current and the one-dimensional Park's vector. The obtained one-dimensional Park's vector is processed to obtain an embedding space vector set, and a recursion matrix is ​​calculated, and a real-time recursion graph is obtained based on the recursion matrix; The obtained Park's vector is processed to obtain an embedding space vector set, and a recursion matrix is ​​calculated. A real-time recursion graph is obtained based on the recursion matrix, including: Determine the embedding dimension m and the time delay τ according to the one-dimensional Park's vector, and construct an embedding space vector constituting a vector set; A recursion matrix RR is calculated for the vector set, and the obtained recursion matrix is ​​plotted in a two-dimensional graph to obtain a real-time recursion graph. No binarization process is performed when calculating the recursion matrix.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; characterized in that: When the processor executes the computer program, the steps of the induction motor stator turn-to-turn short circuit fault diagnosis method combined with a recursion graph are implemented as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program; wherein: When the computer program is executed by a processor, the steps of the method for diagnosing stator turn-to-turn short circuit fault of an induction motor combined with a recursion diagram are implemented.

Images