Fault diagnosis method for IGBT switch tube and current sensor of three-phase AC motor drive circuit

The fault characteristics of three-phase motors are extracted through Park transform and FFT transform, and the TCN time convolution neural model is used for diagnosis, which solves the problem of fault diagnosis and misdiagnosis of IGBT and current sensors in the existing technology, and achieves more accurate and efficient fault detection .

CN119087168AActive Publication Date: 2024-12-06ZHENGZHOU UNIVERSITY OF LIGHT INDUSTRY
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411211658.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-12-06
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

The prior art is difficult to accurately diagnose the faults between the IGBT switch tube and the current sensor in three-phase AC motors at the same time, especially the mutual coupling relationship between the faults is not fully considered, resulting in misdiagnosis.

Method used

Park transform is used to convert three-phase current into two-phase rotation coordinate system, FFT Fourier transform is used to extract the frequency domain fault characteristics, and fault diagnosis is performed through the TCN time convolution neural model, combining causal convolution and expansion convolution layer, using Residual module and Softmax function are classified.

Benefits of technology

It reduces the number of data sets, improves noise immunity, and realizes accurate detection and diagnosis of IGBT and sensor failures, with better diagnostic performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119087168A_ABST
    Figure CN119087168A_ABST
Patent Text Reader

Abstract

The invention provides a fault diagnosis method for an IGBT switch tube and a sensor of a three-phase AC motor drive circuit. The method comprises the following steps: S1, defining a fault mode and a label of an inverter; s2, converting the measured three-phase current in the fault state through Park change to obtain a d-axis current and a q-axis current in a two-phase rotating coordinate system; s3, carrying out FFT (Fast Fourier Transform) on the obtained q-axis current to obtain data of fault characteristics in a frequency domain; and S4, performing fault diagnosis according to the extracted data of the fault features in the frequency domain by using the TCN time convolutional neural model. According to the novel simple IGBT and sensor fault diagnosis method provided by the invention, the obtained three-phase alternating current signal is subjected to Park conversion to obtain the q-axis direct current signal sample, the sine fundamental component of the current signal after Park conversion is eliminated, the extraction of the characteristic signal is facilitated, and the detection accuracy is improved. Meanwhile, q-axis current is selected as a feature extraction sample, so that a data set is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method for an IGBT switch tube and a current sensor of a three-phase AC motor drive circuit. Background Art

[0002] With the rapid development of the power industry and the continuous advancement of power electronics technology, inverters are being used in more and more industries. As an important part of the power electronics system, inverter circuit failure will cause system failure, performance degradation, and even system shutdown in serious cases. Therefore, accurate and rapid fault diagnosis after an inverter failure plays a key role.

[0003] The failure modes of the inverter are mainly divided into short circuit failure and open circuit failure. Short circuit failure of the power switch is generally caused by overvoltage, overheating or incorrect command. Once it occurs, overcurrent may be generated in a very short time to damage the device. Therefore, in the design process of the circuit, a fast fuse is added to quickly cut off the current and convert the short circuit failure into an open circuit failure. Unlike a short circuit failure, when an open circuit failure occurs in the switch device, the system can still operate normally for a short time. However, if it is not handled for a long time, it may cause a secondary failure in the system, causing the system to shut down and cause huge economic losses.

[0004] In the inverter, in addition to the power semiconductor switch tube IGBT failure, the associated sensor failure is also a common fault in the inverter. Compared with the open circuit failure of the power switch tube, when the sensor fails, not only will the measurement signal be distorted, but it will also quickly cause overcurrent and damage other components, which is more serious. After measurement, we know that the fault types of current sensors can be divided into: disconnection fault, stuck fault, and gain fault.

[0005] In the decades of research on three-phase inverter fault diagnosis, diagnostic methods can generally be divided into two categories: mathematical model-based and data-driven. The mathematical model method mainly includes observer method, parameter estimation technology and equivalent space method. For example, in "Current Sensor Fault-Tolerant Control for Encoderless IPMSM Drives Based on Current Space Vector Error Reconstruction", a single-phase current sensor encoderless based on sliding mode observer is established to realize the fault diagnosis and fault-tolerant control of permanent magnet synchronous motor. In "Open-circuit Fault Diagnosis of Midpoint Clamped Three-level Grid-connected Inverter Based on Adaptive Sliding Mode Observer", an adaptive sliding mode observer is established to realize the open-circuit fault diagnosis of three-level grid-connected inverter. Data-driven methods are mainly divided into signal processing methods, knowledge methods, and artificial intelligence technology. Signal processing methods have been widely used in fault diagnosis, and can generally be divided into two types: voltage signal-based and current signal-based. From the above, it can be seen that the fault diagnosis of the switching device IGBT and the sensor in the inverter is generally diagnosed separately, and the mutual coupling relationship between the two faults is not considered, which will cause the switch tube fault to be misdiagnosed when the sensor fails. If there is a method that can diagnose the fault of the switch tube and the sensor at the same time, it can avoid misdiagnosis.

[0006] In order to solve the above problems, many scholars have made research in recent years. A Robust Observer-Based Method for IGBTs and Current Sensors Fault Diagnosis in Voltage-Source Inverters of PMSM Drives proposes a method for diagnosing IGBT and current sensor faults in permanent magnet synchronous motor (PMSM) drives. This method distinguishes IGBT faults from sensor faults by verifying whether the sum of the three-phase currents is zero, and then obtains the residual threshold to achieve fault location. However, this method cannot distinguish the fault type. A New Approach for Current Sensor Fault Diagnosis in PMSG Drives for Wind Energy Conversion Systems proposes a current sensor fault diagnosis method that is different from the traditional state converter-based method. This method requires a system model and corresponding parameters. The proposed diagnostic method only uses the measured phase current, but this method cannot distinguish between specific fault types, sensors and faulty switches. In “An Online Data-Driven Method for Simultaneous Diagnosis of IGBT and Current Sensor Fault of Three-Phase PWM Inverter in Induction Motor Drives”, a data-driven online diagnosis of inverter IGBT open circuit fault and current sensor fault is proposed. However, it is difficult to find the optimal hidden layer weights and output layer weights required for RVFL network training in this diagnostic method. Summary of the invention

[0007] In response to the technical problem of diagnosis and positioning when an IGBT switch tube and a current sensor in a three-phase AC motor fail at the same time, the present invention proposes a fault diagnosis method for the IGBT switch tube and the current sensor in a three-phase AC motor drive circuit, which reduces the number of data sets while improving the anti-noise capability and highlighting the fault characteristics, thereby facilitating accurate detection and diagnosis of faults.

[0008] In order to achieve the above object, the technical solution of the present invention is implemented as follows: a fault diagnosis method for an IGBT switch tube and a sensor of a three-phase AC motor drive circuit, the steps comprising:

[0009] S1: Define the inverter fault modes and labels;

[0010] S2: transform the three-phase current measured under the fault state through Park change to obtain the d-axis current and q-axis current in the two-phase rotating coordinate system;

[0011] S3: Using FFT Fourier transform on the obtained q-axis current to obtain data of fault characteristics in the frequency domain;

[0012] S4: Using the TCN temporal convolution neural model, fault diagnosis is performed based on the fault feature data extracted in the frequency domain.

[0013] The method for defining the fault mode and label of the inverter in step S1 is: the three-phase inverter diagnoses single-tube and double-tube faults, and the fault types are:

[0014]

[0015] Current sensor faults are generally divided into: stuck fault, gain fault, disconnect fault, stuck fault, gain fault, disconnect fault, the definitions of current sensor faults are as follows:

[0016]

[0017] Where i(t) is the output current, i n (t) is the output current under normal operating conditions, C1 is represented by a constant, C2 is a constant less than 0.8 or greater than 1.2, and t1 is the time when the fault occurs;

[0018] The switch tube IGBT fault and current sensor fault have 21 and 9 states respectively, and different numbers are used to represent the switch tube IGBT fault and current sensor fault:

[0019]

[0020]

[0021] The method for obtaining the d-axis and q-axis currents in the two-phase rotating coordinate system in step S2 is:

[0022] The process of converting the measured three-phase current in the natural coordinate system abc into the two-phase current in the two-phase stationary coordinate system αβ is called Clark transformation.

[0023] The Clark transformation formula is:

[0024]

[0025] Among them, i α with i β are the currents in the two-phase stationary coordinate system of α-axis and β-axis respectively, i a 、i b with i c are the corresponding three-phase currents respectively;

[0026] Using Clark transformation Calculated:

[0027]

[0028] Among them, i d with i q are the d-axis and q-axis currents of the corresponding two-phase rotating coordinate system, and θ is the angle between the d-axis and the α-axis;

[0029] By combining the stuck fault current sensor formula and the gain fault current sensor formula, the two-phase currents under the d and q axes can be directly obtained from the natural coordinate system abc.

[0030] The formula for combining the stuck fault current sensor formula and the gain fault current sensor formula is:

[0031]

[0032] The method of utilizing FFT Fourier transform in step S3 is:

[0033] Based on the q-axis variable, the DFT is defined as:

[0034]

[0035] in, is the rotation factor, x(n) is the input time domain signal, and X(k) is the relative amplitude of the N discrete frequency points corresponding to x(n); FFT is achieved by decomposing the N-point DFT into two N / 2-point DFTs, i.e. (N / 2)2+(N / 2)2=N / 2, and continuing to decompose;

[0036] Decompose x(n) into the sum of two sequences of even numbers and odd numbers, that is

[0037] x(n)=x1(n)+x2(n)

[0038] The FFT is defined as:

[0039]

[0040] According to the scalability of the rotation factor, further conversion is obtained:

[0041]

[0042] Among them, x1(n) and x2(n) correspond to the even and odd sequences of x(n) respectively, and X1(k) and X2(k) are the DFT of x1(n) and x2(n) respectively. is the rotation factor for the corresponding even number x1(n) or odd number x2(n);

[0043] After being converted to the frequency domain through FFT, the mean, standard deviation, skewness, and root mean square are used as fault features in the frequency domain.

[0044] The method for fault diagnosis according to the data of fault characteristics extracted in the frequency domain described in step S4 is: classify and number the data of fault characteristics obtained in step S3 according to the fault type, and then input the numbered fault characteristics into the TCN time convolution neural model, and use the TCN time convolution neural model to perform fault diagnosis according to the data of fault characteristics extracted in the frequency domain, and classify the faults.

[0045] Step S4: The TCN temporal convolutional neural model includes at least one residual module, which includes a first dilated causal convolutional layer, a second dilated causal convolutional layer, a first normalization, a second normalization, a first activation function, a second activation function, a first regularization, a second regularization, and a 1*1 convolutional layer. The input end of the first dilated causal convolutional layer is connected to the input end of the convolutional layer, and the output end of the first dilated causal convolutional layer, the first normalization, the first activation function, the first regularization, the second dilated causal convolutional layer, the second normalization, the second activation function, and the second regularization are connected in sequence, and the output end of the second regularization is connected to the output end of the convolutional layer.

[0046] The first activation function and the second activation function are both Relu activation functions, and the first regularization and the second regularization are both droupout regularization.

[0047] The method for performing fault diagnosis based on the data of fault characteristics extracted in the frequency domain in step S4 is:

[0048] The fault characteristics should be fed into the model as input data:

[0049] o = Activation(x + F(x));

[0050] In the above formula, Activation is the activation function, where x is the input data and F(x) is the output of the convolutional layer;

[0051]

[0052]

[0053] in, is the output of the i-th residual module, is the output of the i-1th residual module, It is represented as a batch normalization process, γ and β are the training parameters of batch normalization, and a l It is represented as the output of the batch normalized output of the lth layer after the activation function f. The data for batch normalization.

[0054] The method for classifying faults is: using the Softmax function to classify faults according to fault features.

[0055] The beneficial effects of the present invention are as follows: the present application proposes a new and simple IGBT and sensor fault diagnosis method, which obtains a q-axis DC current signal sample by performing Park transformation on the acquired three-phase AC current signal. The current signal after Park transformation not only eliminates the sine fundamental component, which is helpful for the extraction of characteristic signals, but also selects the q-axis current as the characteristic extraction sample to reduce the data set. Finally, the feature is extracted by utilizing the ability of FFT to highlight features. The use of FFT for fault feature extraction not only improves the calculation speed, but also converts the signal to the frequency domain to improve the signal's anti-interference ability. In terms of fault diagnosis, a TCN time convolutional neural network that can perform efficient parallel operations, ensure the same input and output lengths, and will not leak future data in advance is used to identify and classify fault types. By combining causal convolution with dilated convolution to form an dilated causal convolution layer, not only the input and output lengths are the same, but the output results are only affected by the elements that have occurred before. By selecting residual block stacking connections, the information transmission of the network can achieve cross-layer connections. This method has good diagnostic results for the open circuit fault of the switch tube and the stuck fault, gain fault and disconnection fault of the sensor of the three-phase inverter. Compared with other methods, this method has better performance in fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 The figure is a flow chart of the method of the present invention.

[0058] Figure 2 Schematic diagram of the residual module of the invention.

[0059] Figure 3 It is the TCN network structure.

[0060] Figure 4 This is the recognition accuracy curve of the TCN model in experimental verification.

[0061] Figure 5 This is the model loss curve in experimental verification.

[0062] Figure 6 The following are comparison diagrams of experimental verification results; (a) is the training set result diagram; (b) is the test set result diagram. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, the fault diagnosis method of the IGBT switch tube and the current sensor of the three-phase AC motor drive circuit includes the following steps:

[0065] S1: Define the inverter failure modes and labels.

[0066] The structural topology in the three-phase SPWM pulse width modulation inverter is as follows Figure 1 As shown, each phase of the three-phase inverter has two power switches with anti-parallel diodes and a current sensor.

[0067] When the switch tube T1 fails to open circuit, even if the conduction signal is applied to T1, the upper bridge arm cannot be normally conducted because T1 is in an open circuit state, and there is no positive voltage output in phase A. Only when the switch tube T4 is turned on, there is a negative voltage output in phase A. Therefore, starting from the failure of T1 tube, phase A only has negative current output and no positive current. If the switch tubes T1 and T4 fail at the same time, there is no output current in phase A.

[0068] When a short circuit fault occurs in a three-phase inverter, it is generally converted into an open circuit fault through a fuse. Failures with more than two switch tubes are very rare and the most ideal method of repair is to replace the device. Therefore, the present invention only diagnoses single-tube and double-tube faults. The fault types are shown in Table 1 below:

[0069] Table 1 IGBT fault types

[0070]

[0071] In order to meet safety requirements, three current sensors are generally installed in the inverter. According to the measurement results, current sensor failures are generally divided into: stuck failure, gain failure, and disconnect failure. When a sensor has a stuck failure, it will not only cause current distortion but may also cause circuit overload, leading to safety hazards such as fire and short circuit. When a gain failure occurs, the system will not be able to correctly identify the magnitude of the current, causing the system to lose control. When a drift failure occurs, the sensor output signal will gradually deviate from the true value, making it impossible to accurately control the system. The definitions of stuck failure, gain failure, and disconnect failure current sensor failures are as follows:

[0072]

[0073] Where i(t) is the output current, i n (t) is the output current under normal operating conditions, C1 is represented as a constant, C2 is a constant less than 0.8 or greater than 1.2, and t1 is the time when the fault occurs.

[0074] The faults of the switch tube IGBT and the current sensor have 21 and 9 states respectively. The fault diagnosis of the inverter is actually the identification of its operating state.

[0075] Different inverter faults cause different output currents. In order to accurately diagnose and locate the faults, different numbers are used in this application to represent IGBT faults and current sensor faults, as shown in Table 2. These include not only normal state, IGBT single tube fault, double tube fault, but also sensor stuck, gain, and disconnection faults.

[0076] Table 2 Inverter fault label

[0077]

[0078] S2: The three-phase current measured under the fault state is transformed through Park change to obtain the d-axis and q-axis currents in the two-phase rotating coordinate system.

[0079] Park transformation is a coordinate transformation method that transforms three-phase current from abc to d and q axes. The principle is as follows: First, the measured three-phase current in the natural coordinate system abc is converted to the two-phase current in the two-phase stationary coordinate system αβ. This process is called Clark transformation.

[0080] The Clark transformation formula is:

[0081]

[0082] Among them, i α with i βare the currents in the two-phase stationary coordinate system of α-axis and β-axis respectively, i a 、i b with i c are the corresponding three-phase currents respectively.

[0083] The essence of Park transformation is to transform the stationary αβ coordinate system into the rotating d, q coordinate system, and use Clark transformation to obtain Calculated:

[0084]

[0085] Among them, i d with i q are the d-axis and q-axis currents of the corresponding two-phase rotating coordinate system, and θ is the angle between the d-axis and the α-axis.

[0086] Combining the stuck fault current sensor formula and the gain fault current sensor formula, we can directly obtain the two-phase current formula under the d and q axes from the natural coordinate system abc:

[0087]

[0088] The collected three-phase current becomes a two-phase current after Park transformation, which reduces the data set and can effectively monitor the changes in the AC quantity. The elimination of the sinusoidal fundamental component is helpful for the extraction of fault characteristics.

[0089] S3: Use FFT Fourier transform to obtain the fault characteristic data in the frequency domain.

[0090] In Park transform, the d-axis is overlapped with the α-axis. At the same time, the corresponding q-axis current has different manifestations under different fault conditions. Therefore, only the q-axis current is used as the detection for fault feature extraction. On this basis, the number of data sets is greatly reduced, reducing the calculation time. FFT is a fast algorithm adapted from DFT discrete Fourier transform. At the same time, converting the signal in the time domain to the frequency domain will improve the signal's anti-interference ability, effectively extract feature information, and remove the white noise and high-frequency part of the signal.

[0091] The DFT is defined as:

[0092]

[0093] FFT is achieved by decomposing an N-point DFT into two N / 2-point DFTs. The total amount of calculation for the two N / 2-point DFTs is only half of the original amount, that is, (N / 2)2+(N / 2)2=N / 2. This can be continued, and N / 2 can be decomposed into N / 4-point DFTs, etc. The more values ​​are calculated, the more the amount of calculation is reduced.

[0094] The FFT is defined as:

[0095]

[0096] Among them, x1(n) and x2(n) correspond to the even and odd sequences of x(n) respectively, and X1(k) and X2(k) are the DFT of x1(n) and x2(n) respectively. is the rotation factor for the corresponding even number x1(n) or odd number x2(n);

[0097] After converting to the frequency domain through FFT, the fault characteristics in the frequency domain are obtained using formulas such as mean value, standard deviation, skewness, and root mean square.

[0098] average value:

[0099] Standard Deviation:

[0100] Skewness:

[0101] S4: Perform fault diagnosis based on the data of fault characteristics extracted in the frequency domain.

[0102] Specifically, the data of the fault characteristics obtained in S3 are numbered and then input into the TCN time convolution neural model. The TCN time convolution neural model is used to perform fault diagnosis based on the data of the fault characteristics extracted in the frequency domain.

[0103] The temporal convolutional neural network (TCN) is selected for fault diagnosis. The TCN model has the advantages of efficient parallel computing, equal length of input and output sequences, no gradient explosion and gradient disappearance, etc.

[0104] Different from traditional convolutional neural networks, in order to ensure that the input and output sequences are of equal length and that future data will not be leaked in advance, it is necessary to combine 1D-FCN with causal convolution as dilated causal convolution in temporal convolutional neural networks. Figure 2 As shown, a multi-layer residual block stack is used in the model to replace the previous layer-by-layer connection. Each residual module includes a first dilated causal convolution layer, a second dilated causal convolution layer, a first normalization, a second normalization, a first activation function, a second activation function, a first regularization, a second regularization, and a 1*1 convolution layer. Among them, the input end of the first dilated causal convolution layer is connected to the input end of the convolution layer, the output end of the first dilated causal convolution layer, the first normalization, the first activation function, the first regularization, the second dilated causal convolution layer, the second normalization, the second activation function and the second regularization are connected in sequence, and the output end of the second regularization is connected to the output end of the convolution layer.

[0105] By using the residual module, the output result is only related to the factors that occurred before without losing information. By stacking the residual blocks, the input and output have the same number of channels and solve the gradient disappearance and gradient explosion problems in the model. By adding the relu activation function and droupout regularization in the residual block, the computational complexity and overfitting are simplified. The fault features should be input into the model as input data:

[0106] o = Activation(x + F(x));

[0107] In the above formula, Activation is the activation function, where x is the input data, F(x) is the output after the convolution layer, and the residual module is as follows: Figure 2 shown.

[0108]

[0109] in, is the output of the i-th residual module, is the output of the i-1th residual module, It is represented as a batch normalization process, and γ and β are the training parameters of batch normalization. l It is represented as the output of the batch normalized output of the lth layer after the activation function f.

[0110] The method makes the fault feature extraction not only simple in calculation and improves the anti-noise ability, but also highlights the fault characteristics, which is convenient for accurate detection and diagnosis of faults. The extraction is simple and the anti-noise ability is improved in the Park transform and FFT transform parts. It is the characteristics of the Park transform and FFT transform themselves.

[0111] This application uses Matlab software to implement the programming of the TCN fault diagnosis model. The program design flow chart is as follows: Figure 3 shown.

[0112] like Figure 4 The figure shows the network structure of the TCN model. The TCN model is designed with 6 layers of stacked residual modules. Since the Relu activation function can not only reduce the amount of calculation of TCN but also enable the model to converge quickly, the Relu activation function is used in the connection of the modules, and the Softmax function is added to the final output layer of the TCN model to classify the faults. The Adam algorithm is used to optimize the gradient during the training of the TCN model. For the determination of the consistency of the model input and output targets, this application selects the cross entropy loss function for determination.

[0113] Among them, Relu activation function:

[0114]

[0115] Cross entropy loss function:

[0116]

[0117] Among them, P(y i ) represents the probability distribution of the sample data set, Represents the probability distribution of the prediction dataset.

[0118] The fault features have been extracted in the previous Park transform and FFT transform. This is the last step of the model diagnosis. Through the Softmax function, the maximum probability of each category corresponding to each input data is screened out. The corresponding results are shown in Table 2. In Table 2, each fault is encoded according to type and location. The type category separated by the final Softmax function can be used to determine the location and type of fault. Finally, the cross entropy loss function is used to evaluate the diagnostic accuracy of the model.

[0119] A three-phase inverter circuit model was built and samples were collected to verify the method proposed in this application. When collecting samples, 2000 sampling points were taken every 0.02 seconds as one sample, 60 samples were taken for each operating state, and a total of 480 samples were taken. After the collected samples were shuffled, the samples were divided into training set and test set in a ratio of 7:3.

[0120] Figure 5 This is the model accuracy and loss rate curve when using the TCN model for training. It can be seen that the recognition accuracy and loss rate of the TCN model for the training set tend to stabilize after 40 iterations. The recognition accuracy is finally maintained at 100%, while the loss function eventually approaches zero.

[0121] Input the training set and test set into the trained and saved TCN model respectively, and you can get the following Figure 5 , Figure 6 The test results show that not only the accuracy of the test set reached 100%, but the accuracy of the training set also reached 99.375%, indicating that the model can not only accurately identify the trained data, but also can well identify device faults for data that has not been seen, and has good generalization ability. At the same time, according to the confusion matrix result diagram, it can be seen that the TCN model can accurately identify switch tube faults or sensor faults and the location of each fault type, proving its accuracy and effectiveness in switch tube and sensor fault diagnosis.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A fault diagnosis method for IGBT switch tubes and sensors in a three-phase AC motor drive circuit. It is characterized in that the steps include: S1: Define the inverter fault modes and labels; S2: transform the three-phase current measured under the fault state through Park change to obtain the d-axis current and q-axis current in the two-phase rotating coordinate system; S3: Using FFT Fourier transform on the obtained q-axis current to obtain data of fault characteristics in the frequency domain; S4: Using the TCN temporal convolution neural model, fault diagnosis is performed based on the fault feature data extracted in the frequency domain.

2. The fault diagnosis method for IGBT switch tube and sensor of the three-phase AC motor drive circuit according to claim 1, Features: The method for defining the fault mode and label of the inverter in step S1 is: the three-phase inverter diagnoses single-tube and double-tube faults, and the fault types are: Current sensor faults are generally divided into: stuck fault, gain fault, disconnect fault, stuck fault, gain fault, disconnect fault, the definitions of current sensor faults are as follows: Where, i(t) is the output current, i n (t) is the output current under normal operating conditions, C 1 Expressed as a constant, C 2 is a constant less than 0.8 or greater than 1.2, t 1 is the time when the failure occurred; The switch tube IGBT fault and current sensor fault have 21 and 9 states respectively, and different numbers are used to represent the switch tube IGBT fault and current sensor fault:

3. The fault diagnosis method for IGBT switch tube and sensor of the three-phase AC motor drive circuit according to claim 2, Features: The method for obtaining the d-axis and q-axis currents in the two-phase rotating coordinate system in step S2 is: The process of converting the measured three-phase current in the natural coordinate system abc into the two-phase current in the two-phase stationary coordinate system αβ is called Clark transformation. The Clark transformation formula is: Among them, i α with i β are the currents in the two-phase stationary coordinate system of α-axis and β-axis respectively, i a 、i b with i c are the corresponding three-phase currents respectively; Using Clark transformation Calculated: Among them, i d with i q They are the d-axis and q-axis currents of the corresponding two-phase rotating coordinate system, and θ is the angle between the d-axis and the α-axis. By combining the stuck fault current sensor formula and the gain fault current sensor formula, we can get the value of the natural coordinate system abc The two-phase currents under the d and q axes can be directly obtained.

4. The fault diagnosis method for IGBT switch tube and sensor of a three-phase AC motor drive circuit according to claim 3, Features: The formula for combining the stuck fault current sensor formula and the gain fault current sensor formula is:

5. The fault diagnosis method for IGBT switch tube and sensor of a three-phase AC motor drive circuit according to claim 3 or 4, Features: The method of utilizing FFT Fourier transform in step S3 is: Based on the q-axis variable, the DFT is defined as: in, is the rotation factor, x(n) is the input time domain signal, X(k) is the relative amplitude of the N discrete frequency points corresponding to x(n); FFT is achieved by decomposing the N-point DFT into two N / 2-point DFTs, i.e. (N / 2)2+(N / 2)2=N / 2, Continue to decompose; Decompose x(n) into the sum of two sequences of even numbers and odd numbers, that is x(n)=x 1 (n)+x 2 (n) The FFT is defined as: According to the scalability of the rotation factor, further conversion is obtained: Among them, x 1 (n) and x 2 (n) corresponds to two sequences of x(n) even and odd numbers, X 1 (k) and X 2 (k) are x 1 (n) and x 2 DFT of (n) points, is the corresponding even number x 1 (n) or odd x 2 The rotation factor at (n); After being converted to the frequency domain through FFT, the mean, standard deviation, skewness, and root mean square are used as fault features in the frequency domain.

6. The fault diagnosis method for IGBT switch tube and sensor of the three-phase AC motor drive circuit according to claim 5, Features: The method for fault diagnosis according to the data of fault characteristics extracted in the frequency domain described in step S4 is: classify and number the data of fault characteristics obtained in step S3 according to the fault type, and then input the numbered fault characteristics into the TCN time convolution neural model, and use the TCN time convolution neural model to perform fault diagnosis according to the data of fault characteristics extracted in the frequency domain, and classify the faults.

7. The fault diagnosis method for IGBT switch tube and sensor of a three-phase AC motor drive circuit according to claim 6, Features: Step S4: The TCN temporal convolutional neural model includes at least one residual module, which includes a first dilated causal convolutional layer, a second dilated causal convolutional layer, a first normalization, a second normalization, a first activation function, a second activation function, a first regularization, a second regularization, and a 1*1 convolutional layer. The input end of the first dilated causal convolutional layer is connected to the input end of the convolutional layer, and the output end of the first dilated causal convolutional layer, the first normalization, the first activation function, the first regularization, the second dilated causal convolutional layer, the second normalization, the second activation function, and the second regularization are connected in sequence, and the output end of the second regularization is connected to the output end of the convolutional layer.

8. The fault diagnosis method for IGBT switch tube and sensor of a three-phase AC motor drive circuit according to claim 7, Features: The first activation function and the second activation function are both Relu activation functions, and the first regularization and the second regularization are both droupout regularization.

9. The fault diagnosis method for IGBT switch tube and sensor of a three-phase AC motor drive circuit according to claim 8, Features: The method for performing fault diagnosis based on the data of fault characteristics extracted in the frequency domain in step S4 is: The fault characteristics should be fed into the model as input data: o = Activation(x + F(x)); In the above formula, Activation is the activation function, where x is the input data and F(x) is the output of the convolutional layer; in, is the output of the i-th residual module, is the output of the i-1th residual module, It is represented as a batch normalization process, γ and β are the training parameters of batch normalization, and a l It is represented as the output of the batch normalized output of the lth layer after the activation function f. The data for batch normalization.

10. The fault diagnosis method for IGBT switch tube and sensor of a three-phase AC motor drive circuit according to any one of claims 7 to 9, Features: The method for classifying faults is: using the Softmax function to classify faults according to fault features.

Citation Information

Patent Citations

  • IGBT stuck-open fault diagnosis method for three-phase inverter bridge of frequency converter

    CN102053210A

  • Motor fault diagnosis method and device based on convolutional neural network, and medium

    CN111157894A

  • Fault diagnosis method for current sensor of three-phase motor driving system

    CN112394312A

  • Open-circuit fault diagnosis method for three-phase permanent magnet synchronous motor driving system

    CN113325334A

  • Alternating current motor system power tube open-circuit fault diagnosis method

    CN114545226A