A state detection circuit control method for an IGBT device

Abstract

The present invention relates to the field of power electronics technology, and in particular to a state detection circuit control method for an IGBT device. The method comprises the following steps: wirelessly monitoring the working parameters of the IGBT device through surface acoustic wave sensor technology, thereby obtaining real-time working parameters; using a nano humidity sensor to perform nano ultraviolet excitation treatment on the ambient air, thereby obtaining real-time humidity data; obtaining a historical fault data set, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor working data; performing a time-frequency joint analysis on the fault precursor working data, thereby obtaining a working parameter feature vector. The present invention can improve the efficiency and accuracy of the control of the state detection circuit of the IGBT device.

Description

Technical Field

[0001] The present invention relates to the technical field of power electronics, and in particular to a state detection circuit control method for an IGBT device. Background Art

[0002] IGBT (Insulated Gate Bipolar Transistor) devices are widely used in high-power electronic devices, such as power converters and motor drives. IGBT devices have high voltage and high current carrying capabilities, and have the low on-state voltage drop characteristics of MOSFET (Metal Oxide Semiconductor Field Effect Transistor), so they are widely used in high-power applications. However, IGBT devices may face some problems during operation, such as overcurrent, overvoltage and overtemperature, which may cause damage or failure of the device.

[0003] At present, many methods for detecting the state of IGBT devices have been proposed. Some of these methods use sensors to monitor parameters such as temperature, current and voltage of IGBT devices, and use these parameters to determine the state of the device. However, these methods usually require additional hardware and complex circuits, which increases the cost and complexity of the system. In addition, due to the complex and changeable working environment of IGBT devices, traditional methods still have certain limitations in the accuracy and real-time performance of detecting device states. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a state detection circuit control method for an IGBT device to solve at least one of the above technical problems.

[0005] To achieve the above object, a state detection circuit control method of an IGBT device comprises the following steps:

[0006] Step S1: wirelessly monitor the working parameters of the IGBT device through surface acoustic wave sensor technology to obtain real-time working parameters; use a nano humidity sensor to perform nano ultraviolet excitation treatment on the ambient air to obtain real-time humidity data;

[0007] Step S2: obtaining a historical fault data set, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor work data; performing a time-frequency joint analysis on the fault precursor work data to obtain a work parameter feature vector;

[0008] Step S3: mining the internal correlation relationship of the fault humidity data, fault timestamp data, fault type data and fault precursor work data according to the historical fault data set, so as to obtain the fault correlation knowledge graph;

[0009] Step S4: dividing the fault humidity data into humidity ranges according to the fault timestamp data, thereby obtaining a first humidity range, a second humidity range, and a third humidity range; performing humidity sensitivity calculations on the first humidity range, the second humidity range, and the third humidity range, respectively, thereby obtaining a first humidity sensitivity factor, a second humidity sensitivity factor, and a third humidity sensitivity factor;

[0010] Step S5: diagnose the fault types of the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor according to the fault cause data and the fault type data, thereby obtaining causal probability distribution data; perform causal reasoning adjustment on the fault association knowledge graph using the causal probability distribution data, thereby obtaining a fault causal mechanism diagram;

[0011] Step S6: predicting the fault type of the real-time working parameters based on the fault cause-effect mechanism diagram and the real-time humidity data, thereby obtaining the predicted type;

[0012] Step S7: performing evolution trajectory behavior learning on the fault precursor working data according to the working parameter feature vector, thereby obtaining a working parameter prediction model;

[0013] Step S8: using the working parameter prediction model to predict the fault time range of the prediction type, thereby obtaining prediction time data;

[0014] Step S9: predict and evaluate the fault urgency based on the prediction type and prediction time data according to the fault cause-effect mechanism diagram, thereby obtaining risk level data; and control the IGBT device accordingly using a preset drive circuit based on the risk level data.

[0015] The present invention can realize wireless monitoring of the working parameters and environmental humidity of the IGBT device through surface acoustic wave sensor technology and nano humidity sensor technology, so as to obtain real-time data, which provides a basis for subsequent analysis and processing, so that the working state and fault risk of the IGBT device can be evaluated in real time. First, by obtaining a historical fault data set and performing a time-frequency joint analysis on the fault precursor working data therein, the working parameter feature vectors are obtained, which can reflect the characteristics and changes of the IGBT device under different working conditions, such as current, voltage, temperature, etc. Then, according to the historical fault data set, the fault humidity data, fault timestamp data, fault type data and fault precursor working data are mined for internal correlation, so as to obtain a fault correlation knowledge map, which can provide the rules and causes of different types of faults of the IGBT device under different humidity conditions, such as high humidity may cause insulation breakdown, low humidity may cause electrostatic discharge, etc. Next, the fault humidity data is divided into humidity ranges according to the fault timestamp data, and the humidity sensitivity of different humidity ranges is calculated to obtain different humidity sensitivity factors. These humidity sensitivity factors can reflect the sensitivity of IGBT devices to faults under different humidity conditions. For example, a high humidity sensitivity factor indicates that faults are more likely to occur under high humidity, and a low humidity sensitivity factor indicates that faults are more likely to occur under low humidity. Next, the fault type is diagnosed for different humidity sensitivity factors according to the fault cause data and fault type data, and the fault association knowledge graph is adjusted by causal reasoning using the cause probability distribution data to obtain a fault causal mechanism diagram. This mechanism diagram can provide the causal relationship and probability of different types of faults occurring in IGBT devices under different humidity conditions, such as the probability of insulation breakdown under high humidity, the probability of electrostatic discharge under low humidity, etc. Thirdly, based on the fault cause-effect mechanism diagram and the real-time humidity data, the fault type is predicted for the real-time working parameters, and the evolution trajectory behavior of the fault precursor working data is learned according to the working parameter feature vector, so as to obtain the working parameter prediction model, and the model is used to predict the fault time range of the predicted type, so as to obtain the predicted time data, which can provide the possible fault type, time and evolution process of the IGBT device, such as what kind of fault may occur within how long, how the fault is caused by other parameters, etc. Finally, the present invention predicts and evaluates the fault urgency of the predicted type and the predicted time data according to the fault cause-effect mechanism diagram, and uses the preset driving circuit to control the IGBT device accordingly according to the risk level data, so as to realize the intelligent protection and optimization of the IGBT device, such as early warning before the fault occurs, protection when the fault occurs, and adjustment after the fault is recovered.In summary, the invention provides a novel, efficient and intelligent method for IGBT device status detection, thereby improving the reliability and safety of IGBT devices, reducing failure risks and maintenance costs, and enhancing the performance and efficiency of IGBT devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0017] Figure 1 A schematic flow chart of the steps of a method for controlling a state detection circuit of an IGBT device according to an embodiment is shown.

[0018] Figure 2 A detailed flow chart of step S5 of an embodiment is shown.

[0019] Figure 3 A detailed flowchart diagram of step S56 of an embodiment is shown.

[0020] Figure 4 A detailed flow chart of step S9 of an embodiment is shown. DETAILED DESCRIPTION

[0021] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0023] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0024] To achieve this, please refer to Figures 1 to 4 The present invention provides a state detection circuit control method for an IGBT device, the method comprising the following steps:

[0025] Step S1: wirelessly monitor the working parameters of the IGBT device through surface acoustic wave sensor technology to obtain real-time working parameters; use a nano humidity sensor to perform nano ultraviolet excitation treatment on the ambient air to obtain real-time humidity data;

[0026] This embodiment uses surface acoustic wave sensor technology to install the surface acoustic wave sensor on the surface of the IGBT device, thereby realizing wireless monitoring of the operating voltage, operating current, operating temperature and other parameters of the IGBT device, and wirelessly transmits the monitoring data to the data processing center, thereby obtaining real-time operating parameters. At the same time, this embodiment uses a nano humidity sensor to perform nano ultraviolet excitation treatment on the ambient air, so that the water molecules in the air have a photoelectric effect, thereby changing the resistance value of the nano humidity sensor, and converting the resistance value into a humidity value, thereby obtaining real-time humidity data.

[0027] Step S2: obtaining a historical fault data set, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor work data; performing a time-frequency joint analysis on the fault precursor work data to obtain a work parameter feature vector;

[0028] This embodiment obtains a historical fault data set from the manufacturer or maintenance organization of the IGBT device, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor working data. Then, this embodiment performs a time-frequency joint analysis on the fault precursor working data, such as short-time Fourier transform, wavelet transform, or empirical mode decomposition, to extract the frequency characteristics, energy characteristics, phase characteristics, and other information of the working parameters of the IGBT device in the fault precursor period, and combines this information into a working parameter feature vector.

[0029] Step S3: mining the internal correlation relationship of the fault humidity data, fault timestamp data, fault type data and fault precursor work data according to the historical fault data set, so as to obtain the fault correlation knowledge graph;

[0030] This embodiment uses a knowledge graph construction method to mine the intrinsic correlation relationship of fault humidity data, fault timestamp data, fault type data, and fault precursor work data based on historical fault data sets, thereby obtaining a fault correlation knowledge graph. Specifically, this embodiment uses the fault type of the IGBT device as the node of the knowledge graph, and uses the fault humidity data, fault timestamp data, and fault precursor work data as the edge of the knowledge graph, and assigns the edge a corresponding weight value, thereby indicating the degree of correlation between different types of faults.

[0031] Step S4: dividing the fault humidity data into humidity ranges according to the fault timestamp data, thereby obtaining a first humidity range, a second humidity range, and a third humidity range; performing humidity sensitivity calculations on the first humidity range, the second humidity range, and the third humidity range, respectively, thereby obtaining a first humidity sensitivity factor, a second humidity sensitivity factor, and a third humidity sensitivity factor;

[0032] This embodiment arranges all the fault timestamp data in the historical fault data set in chronological order and divides them into three time periods of equal length, and then calculates the average humidity value in each time period based on the humidity values ​​corresponding to all the faults occurring in the time period, and determines a reasonable humidity interval based on this, thereby obtaining the first humidity range, the second humidity range, and the third humidity range. Next, this embodiment calculates the humidity sensitivity of the first humidity range, the second humidity range, and the third humidity range, respectively, thereby obtaining the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor. Specifically, this embodiment calculates the average working parameter feature vector in each humidity range based on the working parameter feature vectors corresponding to all the faults occurring in the humidity range, and performs similarity calculation with the global average working parameter feature vector, thereby obtaining the humidity sensitivity of the humidity range, that is, the humidity sensitivity factor.

[0033] Step S5: diagnose the fault types of the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor according to the fault cause data and the fault type data, thereby obtaining causal probability distribution data; perform causal reasoning adjustment on the fault association knowledge graph using the causal probability distribution data, thereby obtaining a fault causal mechanism diagram;

[0034] This embodiment uses the Bayesian network method to calculate the probability of occurrence of different types of faults corresponding to each humidity sensitivity factor according to the degree of association between each humidity sensitivity factor and different types of faults, and combines these probabilities into causal probability distribution data. Then, the causal probability distribution data is used to perform causal reasoning adjustment on the fault-related knowledge graph to obtain a fault causal mechanism diagram. Specifically, this embodiment updates the weights of the edges in the knowledge graph according to the causal probability distribution data, and uses causal reasoning algorithms, such as the Pearl algorithm or the Do-calculus algorithm, to obtain the causal relationship between the faults, and represents these relationships in the fault causal mechanism diagram.

[0035] Step S6: predicting the fault type of the real-time working parameters based on the fault cause-effect mechanism diagram and the real-time humidity data, thereby obtaining the predicted type;

[0036] This embodiment determines the humidity range of the current IGBT device based on the real-time humidity data, and uses a classification algorithm, such as a support vector machine, a decision tree, or a neural network, based on the humidity sensitivity factor corresponding to the humidity range and the cause-effect relationship in the fault cause-effect mechanism diagram, to predict the type of fault that may occur in the current IGBT device and use it as the prediction type.

[0037] Step S7: performing evolution trajectory behavior learning on the fault precursor working data according to the working parameter feature vector, thereby obtaining a working parameter prediction model;

[0038] This embodiment performs evolution trajectory behavior learning on the fault precursor working data according to the working parameter feature vector, thereby obtaining a working parameter prediction model. Specifically, this embodiment uses deep learning methods, such as long short-term memory networks, convolutional neural networks, or variational autoencoders, to learn the evolution laws of the working parameter feature vectors of the IGBT device in different types of fault precursor periods, and express these laws as a working parameter prediction model.

[0039] Step S8: using the working parameter prediction model to predict the fault time range of the prediction type, thereby obtaining prediction time data;

[0040] This embodiment determines the corresponding working parameter prediction model according to the prediction type, and inputs the real-time working parameter feature vector of the current IGBT device into the model, thereby obtaining the predicted value of the working parameter feature vector in a future period of time, and performs similarity matching with the same type of fault precursor working data in the historical fault data set based on these predicted values, thereby determining the time range in which the current IGBT device may experience this type of fault, and using it as the predicted time data.

[0041] Step S9: predict and evaluate the fault urgency based on the prediction type and prediction time data according to the fault cause-effect mechanism diagram, thereby obtaining risk level data; and control the IGBT device accordingly using a preset drive circuit based on the risk level data.

[0042] This embodiment determines the severity of the possible fault of the current IGBT device according to the prediction type and prediction time data, and judges the probability and impact range of other faults that may be caused by the fault according to the causal relationship in the fault causal mechanism diagram, thereby comprehensively evaluating the fault risk of the current IGBT device and dividing it into different risk levels, such as low risk, medium risk or high risk, and using it as risk level data. Finally, this embodiment uses a preset drive circuit to control the IGBT device accordingly according to the risk level data, such as adjusting working parameters, cutting off power or alarming, so as to ensure the safe operation of the IGBT device.

[0043] The present invention can realize wireless monitoring of the working parameters and environmental humidity of the IGBT device through surface acoustic wave sensor technology and nano humidity sensor technology, so as to obtain real-time data, which provides a basis for subsequent analysis and processing, so that the working state and fault risk of the IGBT device can be evaluated in real time. First, by obtaining a historical fault data set and performing a time-frequency joint analysis on the fault precursor working data therein, the working parameter feature vectors are obtained, which can reflect the characteristics and changes of the IGBT device under different working conditions, such as current, voltage, temperature, etc. Then, according to the historical fault data set, the fault humidity data, fault timestamp data, fault type data and fault precursor working data are mined for internal correlation, so as to obtain a fault correlation knowledge map, which can provide the rules and causes of different types of faults of the IGBT device under different humidity conditions, such as high humidity may cause insulation breakdown, low humidity may cause electrostatic discharge, etc. Next, the fault humidity data is divided into humidity ranges according to the fault timestamp data, and the humidity sensitivity of different humidity ranges is calculated to obtain different humidity sensitivity factors. These humidity sensitivity factors can reflect the sensitivity of IGBT devices to faults under different humidity conditions. For example, a high humidity sensitivity factor indicates that faults are more likely to occur under high humidity, and a low humidity sensitivity factor indicates that faults are more likely to occur under low humidity. Next, the fault type is diagnosed for different humidity sensitivity factors according to the fault cause data and fault type data, and the fault association knowledge graph is adjusted by causal reasoning using the cause probability distribution data to obtain a fault causal mechanism diagram. This mechanism diagram can provide the causal relationship and probability of different types of faults occurring in IGBT devices under different humidity conditions, such as the probability of insulation breakdown under high humidity, the probability of electrostatic discharge under low humidity, etc. Thirdly, based on the fault cause-effect mechanism diagram and the real-time humidity data, the fault type is predicted for the real-time working parameters, and the evolution trajectory behavior of the fault precursor working data is learned according to the working parameter feature vector, so as to obtain the working parameter prediction model, and the model is used to predict the fault time range of the predicted type, so as to obtain the predicted time data, which can provide the possible fault type, time and evolution process of the IGBT device, such as what kind of fault may occur within how long, how the fault is caused by other parameters, etc. Finally, the present invention predicts and evaluates the fault urgency of the predicted type and the predicted time data according to the fault cause-effect mechanism diagram, and uses the preset driving circuit to control the IGBT device accordingly according to the risk level data, so as to realize the intelligent protection and optimization of the IGBT device, such as early warning before the fault occurs, protection when the fault occurs, and adjustment after the fault is recovered.

[0044] Preferably, step S1 comprises the following steps:

[0045] Step S11: using a surface acoustic wave sensor to collect acoustic wave data of the working current and voltage of the IGBT device in real time through a preset wireless module, thereby obtaining a working acoustic wave signal;

[0046] In this embodiment, a surface acoustic wave sensor is attached to the surface of the IGBT device and communicates with a data processing center through a wireless module, thereby realizing wireless monitoring of the working current and voltage of the IGBT device, converting the monitoring data into an acoustic wave signal, and wirelessly transmitting the acoustic wave signal to the data processing center, thereby obtaining a working acoustic wave signal.

[0047] Step S12: performing wavelet transform on the working sound wave signal to obtain a sound wave feature matrix;

[0048] This embodiment uses wavelet transform methods, such as discrete wavelet transform, continuous wavelet transform or wavelet packet transform, to decompose the working sound wave signal into sub-signals of different scales and frequencies, and extract the energy, frequency, phase and other characteristics of each sub-signal, and combine these characteristics into a sound wave feature matrix.

[0049] Step S13: performing principal component analysis on the acoustic wave feature matrix to obtain a principal component feature vector;

[0050] This embodiment uses the principal component analysis method to reduce the dimension of the acoustic wave feature matrix, extract the most representative and distinguishing features therein, namely, the principal components, and combine these principal components into a principal component feature vector.

[0051] Step S14: standardizing the principal component feature vector to obtain real-time voltage and current data;

[0052] This embodiment uses a standardization method, such as Z-score standardization, maximum and minimum standardization, or unit vector standardization, to convert the principal component feature vectors into numerical values ​​that meet certain specifications and ranges, and restore these numerical values ​​to real-time voltage and current data of the IGBT device according to a pre-set mapping relationship.

[0053] Step S15: obtaining real-time temperature data by a thermal paint method; using the real-time temperature data, real-time voltage and current data as real-time working parameters;

[0054] In this embodiment, the thermal paint is applied to the surface of the IGBT device, and the color change of the thermal paint is observed in real time by an optical instrument, and the color change is converted into a temperature value according to the preset corresponding relationship between color and temperature, so as to obtain real-time temperature data. Then, this embodiment uses the real-time temperature data, real-time voltage and current data as real-time working parameters.

[0055] Step S16: Exciting the hygroscopic material of the nano humidity sensor by ultraviolet irradiation, and monitoring the resistance change of the nano humidity sensor in real time, thereby obtaining a resistance signal;

[0056] In this embodiment, an ultraviolet lamp is used to periodically irradiate the hygroscopic material of the nano humidity sensor, and a circuit is used to measure the resistance value of the nano humidity sensor, and the resistance value is converted into a resistance signal, and the resistance signal is wirelessly transmitted to a data processing center, thereby obtaining a resistance signal.

[0057] Step S17: Calculate the Shannon entropy change of the resistance signal to obtain real-time humidity data.

[0058] This embodiment uses the Shannon entropy method to calculate the information content and uncertainty of the resistance signal, and converts the Shannon entropy value into a humidity value according to a preset mapping relationship, thereby obtaining real-time humidity data.

[0059] The present invention can realize wireless monitoring of the working parameters and environmental humidity of the IGBT device through the acoustic wave and humidity sensor technology, thereby obtaining real-time data, which provides a basis for subsequent analysis and processing, so that the working state and failure risk of the IGBT device can be evaluated in real time. First, the surface acoustic wave sensor is used to collect the acoustic wave data of the working current and voltage of the IGBT device through a preset wireless module in real time, thereby obtaining the working acoustic wave signal, which can reflect the working characteristics and changes of the IGBT device, such as the amplitude, frequency, phase, etc. of the signal. These signals can avoid wired connection of the IGBT device, reduce interference and loss, and improve monitoring efficiency and accuracy. Then, by performing wavelet transformation on the working acoustic wave signal, the acoustic wave feature matrix is ​​obtained, which can provide the acoustic wave feature information of the IGBT device at different time scales and frequency scales, such as the energy, entropy, variance, etc. of the signal. This information can effectively extract useful information and remove noise from the signal, and enhance the distinguishability and identifiability of the signal. Next, the principal component analysis is performed on the acoustic wave feature matrix to obtain the principal component feature vectors, which can provide acoustic wave feature information of the IGBT device in different dimensions, such as the main component and secondary component of the signal. This information can effectively reduce the dimension and complexity of the data and highlight the main features and change trends in the signal. Next, the real-time voltage and current data are obtained by standardizing the principal component feature vectors, which can provide real-time operating parameter information of the IGBT device, such as voltage of 220V, current of 10A, etc. These data can effectively convert the acoustic wave signal into an electrical signal and maintain the consistency and comparability between the signals. Again, the present invention obtains real-time temperature data through the thermosensitive paint method; the real-time temperature data, real-time voltage and current data are used as real-time operating parameters; these data can provide real-time working status information of the IGBT device, such as temperature of 50°C, voltage of 220V, current of 10A, etc. These data can effectively monitor the temperature rise of the IGBT device and conduct a comprehensive analysis with the voltage and current data. Finally, the hygroscopic material of the nano humidity sensor is excited by ultraviolet irradiation, and the resistance change of the nano humidity sensor is monitored in real time to obtain the resistance signal; the Shannon entropy change of the resistance signal is calculated to obtain real-time humidity data. These data can provide environmental humidity information, such as humidity of 60%, Shannon entropy of 0.8, etc. These data can effectively use the hygroscopicity of nanomaterials and ultraviolet excitation technology to detect environmental humidity, and use Shannon entropy changes to quantify humidity changes.

[0060] Preferably, the Shannon entropy change calculation in step S17 is to calculate the change of the resistance signal through the humidity Shannon entropy detection formula, so as to obtain real-time humidity data, wherein the humidity Shannon entropy detection formula is as follows:

[0061]

[0062] In the formula, H(R) is the real-time humidity data, R is the average value of the resistance signal, n is the total number of resistance signals, i is the serial number of the resistance signal value, σ is the standard deviation of the resistance signal, and R i is the value of the i-th resistance signal, α is the sine adjustment coefficient, β is the sine adjustment frequency, γ is the cosine adjustment coefficient, and δ is the cosine adjustment frequency.

[0063] The present invention constructs a humidity Shannon entropy detection formula, which utilizes the concept of Shannon entropy. Shannon entropy is an indicator for measuring the amount of information, which reflects the uncertainty and complexity of information. The higher the Shannon entropy, the more uncertain and complex the information is. The lower the Shannon entropy, the more certain and simple the information is. In the formula, the resistance signal is a kind of information, which reflects the state of the humidity sensor. The change of the resistance signal can represent the change of humidity. The more stable the resistance signal is, the more stable the humidity is; the more fluctuating the resistance signal is, the more fluctuating the humidity is. Therefore, the Shannon entropy of the resistance signal can represent the uncertainty and complexity of humidity. By calculating the ratio of the square of the difference between each resistance signal value and the average resistance to the standard deviation, the discrete degree of the resistance signal value can be reflected. At the same time, the sine and cosine adjustment coefficients are introduced to adjust the sine and cosine changes, which increases the complexity of the calculation process. The sine function can reflect the periodic change law of the resistance signal, and the cosine function can reflect the fluctuation of the resistance signal value. Adjusting the frequencies β and δ controls the change rates of sine and cosine, and enhances the dynamic adaptability of the detection calculation. Finally, the Shannon entropy, that is, the real-time humidity value, is obtained by summing. This formula comprehensively considers the average value, discreteness, periodic changes, fluctuations and dynamic adjustment factors of the resistance signal, performs complex function conversion, extracts resistance characteristics, obtains accurate and dynamic humidity data, and improves the accuracy and real-time performance of humidity detection.

[0064] Preferably, step S2 comprises the following steps:

[0065] Step S21: Acquire a historical fault data set, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor work data;

[0066] This embodiment obtains a historical fault data set from the manufacturer or maintenance organization of the IGBT device, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor work data. These data record different types of faults and related information of the IGBT device under different environments and working conditions, thereby providing a basis for subsequent fault analysis and prediction.

[0067] Step S22: dividing the fault time into fault time periods according to the fault timestamp data, thereby obtaining a fault time period set, wherein the fault time period set includes multiple fault time periods;

[0068] This embodiment sorts all fault timestamp data in the historical fault data set in chronological order and divides them into a number of time periods of equal length, such as daily, weekly or monthly, and uses these time periods as fault time periods, and combines these fault time periods into a fault time period set.

[0069] Step S23: performing fault count for each fault time period in the fault time period set, thereby obtaining fault count data;

[0070] This embodiment calculates the number of failures in each failure time period according to the number of all failures occurring in the time period, and records the number of failures as failure number data.

[0071] Step S24: calculating the probability of failure occurrence for the number of failures in each failure time period in the failure time period set, thereby obtaining the failure probability of the time period;

[0072] This embodiment calculates the probability of any one or more failures occurring in each failure time period according to the number of failures in the time period and the total number of monitoring times in the time period, and records these probabilities as the time period failure probability.

[0073] Step S25: performing multi-level decomposition on the fault precursor working data to obtain a frequency band signal component set, wherein the frequency band signal component set includes multiple frequency band signal components;

[0074] This embodiment uses a multi-level decomposition method, such as multi-resolution analysis, empirical mode decomposition or sparse representation, to decompose the fault precursor working data into different frequency band signal components, and combines these frequency band signal components into a frequency band signal component set.

[0075] Step S26: extracting the time-frequency energy features of each frequency band signal component in the frequency band signal component set, thereby obtaining an energy feature matrix;

[0076] This embodiment calculates the time-frequency energy spectrum of each frequency band signal component by using a time-frequency energy feature extraction method, such as short-time Fourier transform, Hilbert-Huang transform or Wigner-Ville distribution, and extracts features such as energy peak value, energy mean, energy variance, and combines these features into an energy feature matrix.

[0077] Step S27: Probability weighting is performed on the failure probabilities in different time periods by using the maximum likelihood estimation method, thereby obtaining a failure probability weight vector;

[0078] This embodiment assumes that the probability of failure in each time period follows a binomial distribution; then, based on the relationship between the probability of failure in each time period and the number of failures that occurred in the same time period in the historical fault data set, this embodiment writes a likelihood function of the number of failures in each time period; then, this embodiment takes the logarithm of the likelihood function and simplifies it; then, this embodiment solves the extreme value of the log-likelihood function, which is the maximum likelihood estimate of the probability of failure in each time period; finally, this embodiment checks whether the obtained parameters are reasonable, such as whether they are within the interval [0, 1], whether they are consistent with historical data, etc., and uses these parameters as the optimal weights of the probability of failure in each time period, and combines these weights into a failure probability weight vector.

[0079] Step S28: perform element-wise multiplication of the fault probability weight vector and the energy characteristic matrix to obtain a working parameter characteristic vector.

[0080] This embodiment first multiplies the fault probability weight vector by each row of the energy feature matrix to obtain a weighted energy feature matrix; then, each column of the weighted energy feature matrix is ​​summed to obtain a working parameter feature vector. In this way, the working parameter feature vector can reflect the impact of the fault probability on the energy feature in different time periods, thereby improving the distinguishability and effectiveness of the working parameter feature vector.

[0081] The present invention can realize the effective extraction of the working parameter characteristics of the IGBT device through the fault probability weight technology, so as to obtain useful data, which provides a basis for subsequent analysis and processing, so that the fault type and time of the IGBT device can be accurately predicted. First, by obtaining the historical fault data set, and dividing the fault time into fault time periods according to the fault timestamp data, a fault time period set is obtained, and these time periods can reflect the situation of the IGBT device failing in different time periods, such as morning, noon, and evening. Then, by counting the number of faults in each fault time period in the fault time period set, and calculating the probability of failure of each fault time period, the time period failure probability is obtained. These probabilities can reflect the possibility of failure of the IGBT device in different time periods, such as the probability of failure in the morning is 0.2, the probability of failure at noon is 0.4, etc. Next, by performing multi-level decomposition on the fault precursor working data, a frequency band signal component set is obtained, and the time-frequency energy feature extraction is performed on each frequency band signal component in the frequency band signal component set, so as to obtain an energy feature matrix. These matrices can provide the time-frequency energy feature information of the IGBT device in different frequency bands, such as the energy, entropy, and variance of the signal in the low frequency band, the medium frequency band, and the high frequency band. Next, the present invention performs probability weighting on the fault probability of different time periods by the maximum likelihood estimation method, so as to obtain the fault probability weight vector. These vectors can provide the weight information of the fault occurring in different time periods, such as the weight of the fault occurring in the morning is 0.3, the weight of the fault occurring at noon is 0.5, etc. Thirdly, by performing element-wise multiplication of the fault probability weight vector and the energy feature matrix, the working parameter feature vector is obtained. These vectors can provide the working parameter feature information of the IGBT device in different frequency bands, and the influencing factors of the fault occurring in different time periods are considered, such as the energy of the signal in the low frequency band multiplied by the weight of the fault occurring in the morning is 0.6, the energy of the signal in the medium frequency band multiplied by the weight of the fault occurring at noon is 0.8, etc.

[0082] Preferably, step S3 comprises the following steps:

[0083] Step S31: calculating the fault interval duration of the fault timestamp data, thereby obtaining an interval duration data set, wherein the interval duration data set includes a plurality of interval duration data;

[0084] This embodiment subtracts the timestamps of two adjacent faults in the fault timestamp data to obtain the time interval between each two faults, thereby calculating the fault interval duration. Then, all fault interval duration data are stored in a data set as an interval duration data set. The interval duration data set includes multiple interval duration data, each interval duration data representing the duration of a fault interval.

[0085] Step S32: taking each fault in the historical fault data set as a fault sample, thereby obtaining a fault sample set;

[0086] This embodiment constructs a fault sample set by treating each fault in the historical fault data set as an independent sample. The fault sample set includes multiple fault samples, each of which contains relevant information about the fault, such as fault humidity data, fault timestamp data, fault type data, and fault precursor work data.

[0087] Step S33: hierarchically clustering the fault sample set using the interval duration data set, thereby obtaining a fault clustering result;

[0088] This embodiment uses a hierarchical clustering algorithm to perform cluster analysis on the fault sample set, thereby obtaining a fault clustering result. The hierarchical clustering algorithm is an unsupervised machine learning method that can classify fault samples into different categories based on the similarity or distance between them. In this embodiment, the interval duration data is used as the basis for clustering, that is, the interval duration between each two fault samples is used to determine whether they belong to the same category. Finally, a hierarchical clustering result is obtained, in which each category corresponds to a group of fault samples with similar or similar interval durations.

[0089] Step S34: Calculate the Pearson coefficient for the fault humidity data, fault timestamp data, fault type data, and fault precursor work data corresponding to each category in the fault clustering result, so as to obtain the intra-class correlation coefficient;

[0090] In this embodiment, the Pearson coefficient is calculated for the fault humidity data, fault timestamp data, fault type data, and fault precursor work data corresponding to each category in the fault clustering result, thereby obtaining the intra-class correlation coefficient. The Pearson coefficient is a statistical indicator that measures the degree of linear correlation between two variables. Its value range is [-1,1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation. In this embodiment, the Pearson coefficient is used to analyze the correlation between various data within each category, thereby revealing the fault characteristics and laws of different categories.

[0091] Step S35: using a preset correlation coefficient threshold to perform critical value filtering on the intra-class correlation coefficient, thereby obtaining strongly correlated variable pairs;

[0092] This embodiment uses a preset correlation coefficient threshold to perform critical value filtering on the intra-class correlation coefficient, thereby obtaining strongly correlated variable pairs. The correlation coefficient is a statistical indicator that measures the strength of the linear relationship between two variables, and its value range is -1 to 1. The larger the absolute value, the stronger the correlation. The preset correlation coefficient threshold is a reasonable value determined according to task requirements and data characteristics, such as 0.8 or 0.9. Through critical value filtering, variable pairs with high correlation can be screened out, thereby obtaining strongly correlated variable pairs.

[0093] Step S36: performing association analysis on strongly correlated variable pairs of each category, thereby obtaining frequent item sets within the category;

[0094] This embodiment performs association analysis on strongly correlated variable pairs of each category to obtain frequent item sets within the category. Association analysis is a method for mining hidden association rules and patterns in data. Its basic idea is to find the combination of items that appear simultaneously in the data, i.e., frequent item sets. Through association analysis, variable pairs that often co-occur or affect faults in the same category can be found, thereby obtaining frequent item sets within the category.

[0095] Step S37: performing association rule mining on frequent item sets within the class to obtain association rules;

[0096] This embodiment obtains association rules by mining association rules on frequent item sets within a class. An association rule is an expression that represents a causal relationship or conditional probability between variables. Its general form is X->Y, which represents the probability that Y will also occur when X occurs. Through association rule mining, rules with high confidence and support can be extracted from frequent item sets within a class to obtain association rules.

[0097] Step S38: construct a knowledge graph for strongly correlated variable pairs and association rules, thereby obtaining a fault-related knowledge graph.

[0098] This embodiment constructs a knowledge graph for strongly correlated variable pairs and association rules to obtain a fault-related knowledge graph. A knowledge graph is a graphical model that represents knowledge structure and semantic relationships, and is composed of entity, attribute, and relationship triplets. Through knowledge graph construction, strongly correlated variable pairs can be used as entities, and association rules can be used as attributes and relationships to construct a knowledge graph that describes the complex connections between faults, thereby obtaining a fault-related knowledge graph.

[0099] The present invention can realize the effective construction of the fault-related knowledge graph of the IGBT device through the fault interval time technology, so as to obtain useful data, which provides a basis for subsequent analysis and processing, so that the fault causes and laws of the IGBT device can be deeply understood. First, by calculating the fault interval time of the fault timestamp data, the interval time data set is obtained. These data sets can reflect the time interval between different faults of the IGBT device, for example, the interval time between the first fault and the second fault is 10 minutes, and the interval time between the second fault and the third fault is 20 minutes. Then, by taking each fault in the historical fault data set as a fault sample, a fault sample set is obtained, and the fault sample set is hierarchically clustered using the interval time data set to obtain the fault clustering results. These results can classify fault samples with similar interval time characteristics into different categories, for example, the first category contains 10 fault samples, the second category contains 15 fault samples, etc. Next, the Pearson coefficient is calculated for the fault humidity data, fault timestamp data, fault type data, and fault precursor work data corresponding to each category in the fault clustering results, so as to obtain the intra-class correlation coefficient. These coefficients can reflect the strength of the correlation between different variables. For example, the correlation coefficient between humidity data and type data is 0.8, indicating that there is a strong positive correlation between them. Next, the intra-class correlation coefficient is filtered by the preset correlation coefficient threshold to obtain strongly correlated variable pairs. These variable pairs can reflect the significant influence relationship between different variables, such as humidity data and type data is a strongly correlated variable pair. Thirdly, by performing association analysis on the strongly correlated variable pairs of each category, frequent item sets within the class are obtained, and association rules are mined for frequent item sets within the class to obtain association rules. These rules can reflect the deterministic or probabilistic causal or logical relationship between different variables, such as humidity above 80% leading to insulation breakdown is a deterministic causal relationship rule. Finally, by constructing a knowledge graph for strongly correlated variable pairs and association rules, a fault-related knowledge graph is obtained. This knowledge graph can provide associated knowledge information of different types of faults occurring in IGBT devices under different categories and display it in a graphical manner.

[0100] Preferably, step S4 comprises the following steps:

[0101] Step S41: sorting the fault humidity data by time according to the fault timestamp data, thereby obtaining humidity sorting data;

[0102] This embodiment obtains the fault timestamp data and the fault humidity data, and then sorts the fault humidity data in chronological order according to the fault timestamp data, thereby obtaining the humidity sorting data.

[0103] Step S42: curve fitting and visualization of humidity sorting data to obtain humidity evolution trajectory;

[0104] In this embodiment, a mathematical function describing the change of humidity over time is obtained by curve fitting on the humidity sorting data, and then the function is plotted into a curve graph using a visualization tool to obtain the humidity evolution trajectory.

[0105] Step S43: Analyze the change pattern of the humidity evolution trajectory to obtain the humidity evolution law and humidity sensitive range;

[0106] This embodiment analyzes the change pattern of the humidity evolution trajectory to identify the trend of humidity growth, decrease or stability in different time periods, as well as the fluctuation amplitude and frequency of humidity in different time periods, so as to obtain the humidity evolution law and humidity sensitive interval. The humidity sensitive interval refers to the interval where humidity changes have a greater impact on the occurrence of faults.

[0107] Step S45: dividing the historical fault data set into samples according to the temperature range according to the first humidity range, the second humidity range and the third humidity range, thereby obtaining the first sample set, the second sample set and the third sample set;

[0108] This embodiment divides the historical fault data set into samples according to the temperature range according to the first humidity range, the second humidity range and the third humidity range, thereby obtaining the first sample set, the second sample set and the third sample set. The historical fault data set refers to a data set that includes fault timestamp data, fault humidity data and fault type data, wherein the fault type data refers to the classification label of the fault, such as short circuit, overheating or leakage. By dividing the historical fault data set into samples according to the temperature range, faults occurring under different humidity conditions can be classified into different sample sets, thereby obtaining the first sample set, the second sample set and the third sample set. The first sample set refers to the sample set of faults occurring within the first humidity range, the second sample set refers to the sample set of faults occurring within the second humidity range, and the third sample set refers to the sample set of faults occurring within the third humidity range.

[0109] Step S46: performing humidity-induced fault counts on the first sample set, the second sample set, and the third sample set according to the corresponding fault cause data, thereby obtaining corresponding humidity-induced frequencies;

[0110] This embodiment analyzes the cause of the fault for each sample in the first sample set, the second sample set, and the third sample set to determine whether it is a fault induced by humidity, and records it in the corresponding fault cause data. Then, according to the fault cause data, the number of humidity-induced faults in each sample set is counted to obtain the corresponding humidity-induced frequency. For example, if there are 10 samples in the first sample set, of which 3 samples are faults induced by humidity, the humidity-induced frequency of the first sample set is 3 / 10=0.3.

[0111] Step S47: Calculate the humidity sensitivity of the humidity-induced frequency in each sample set, so as to obtain a first humidity sensitivity factor, a second humidity sensitivity factor and a third humidity sensitivity factor.

[0112] This embodiment uses the formula F1=1-e (-kQ) To calculate the humidity sensitivity in each sample set, F is the humidity sensitivity, Q is the relative humidity, and k is the humidity sensitivity coefficient. According to the humidity induced frequency in each sample set as the humidity sensitivity value, the humidity sensitivity coefficient corresponding to each sample set can be solved to obtain the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor. For example, if the humidity induced frequency of the first sample set is 0.3, the first humidity sensitivity factor can be solved as

[0113] The present invention can realize effective analysis of the humidity sensitivity of IGBT devices through humidity evolution trajectory technology, thereby obtaining useful data, which provides a basis for subsequent analysis and processing, so that the fault risk of IGBT devices under different humidity conditions can be evaluated and predicted. Among them, the humidity evolution trajectory technology is a technology for analyzing the humidity changes of IGBT devices at different time points. It can obtain humidity evolution trajectories by sorting the fault humidity data by time, curve fitting, and visualization, thereby reflecting the humidity change characteristics and sensitivity of IGBT devices. First, the fault humidity data is sorted by time according to the fault timestamp data to obtain humidity sorting data, and the humidity sorting data is curve fitted and visualized to obtain humidity evolution trajectories. These trajectories can reflect the humidity changes of IGBT devices at different time points, such as 50% humidity in the morning, 70% at noon, and 60% at night. Then, by analyzing the change pattern of the humidity evolution trajectory, the humidity evolution law and humidity sensitive interval are obtained. These laws and intervals can reflect the humidity change characteristics and sensitivity of IGBT devices in different time periods. For example, the humidity has a sharp upward trend at noon, and it is a high sensitivity interval when it is above 70%. Next, the humidity data is divided into temperature ranges according to the humidity evolution law and humidity sensitive interval, and the historical fault data set is divided into samples according to the temperature range according to the first humidity range, the second humidity range and the third humidity range, so as to obtain the first sample set, the second sample set and the third sample set. These sample sets can classify fault samples with similar humidity characteristics into different categories, for example, the first category contains 10 fault samples, the second category contains 15 fault samples, etc. Next, the number of humidity-induced faults is counted for the first sample set, the second sample set and the third sample set according to the corresponding fault cause data, and the humidity sensitivity of the humidity-induced frequency in each sample set is calculated, so as to obtain the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor, which can reflect the humidity sensitivity of IGBT devices to different types of faults under different categories. For example, the humidity sensitivity factor of insulation breakdown under the first category is 0.8, indicating that high humidity is easy to cause insulation breakdown under this category.

[0114] Preferably, step S5 comprises the following steps:

[0115] Step S51: establishing a correlation matrix between the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor and the fault type data, thereby obtaining a humidity-sensitive fault probability matrix;

[0116] This embodiment obtains the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor, such as temperature, humidity, air pressure, etc., and establishes a correlation matrix according to different fault types, such as open circuit, short circuit, leakage, etc., to represent the relationship strength between each humidity sensitivity factor and each fault type. In this way, a humidity sensitivity fault probability matrix can be obtained to represent the probability of various fault types occurring under given humidity sensitivity factors.

[0117] Step S52: performing Bayesian posterior probability calculation on the fault type data corresponding to the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor based on the humidity sensitivity failure probability matrix, thereby obtaining a posterior probability matrix;

[0118] This embodiment uses the Bayesian formula to calculate the posterior probability of the fault type data corresponding to each humidity sensitive factor based on the humidity sensitive fault probability matrix, that is, when a certain fault type is known to occur, the probability of each humidity sensitive factor is inferred. In this way, a posterior probability matrix can be obtained to represent the possibility of each humidity sensitive factor under a given fault type.

[0119] Step S53: diagnose the fault type of each humidity sensitive factor according to the posterior probability matrix, so as to obtain the humidity sensitive factor diagnosis result;

[0120] This embodiment performs fault type diagnosis on each humidity sensitive factor according to the posterior probability matrix, that is, according to the posterior probability value of each humidity sensitive factor, determines the fault type that is most likely to cause the humidity sensitive factor to be abnormal. In this way, a humidity sensitive factor diagnosis result can be obtained to indicate the most likely fault type corresponding to each humidity sensitive factor.

[0121] Step S54: using the fault cause data and the fault type data to verify the fault type of the humidity sensitive factor diagnosis result, so as to obtain a verified fault probability distribution;

[0122] This embodiment verifies the fault type of the humidity sensitivity factor diagnosis result by using the fault cause data and the fault type data, that is, determines whether the humidity sensitivity factor diagnosis result is consistent with the actual fault type, and calculates the probability of their consistency, thereby obtaining a verified fault probability distribution. For example, if the first humidity sensitivity factor diagnosis result is a class A fault, and the actual fault type data is a class B fault, the verified fault probability distribution is P(A|B).

[0123] Step S55: using the verification fault probability distribution to adjust the weights of the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor, thereby obtaining causal probability distribution data;

[0124] This embodiment uses the verified fault probability distribution to adjust the weights of the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor, that is, to give different weight coefficients according to their consistency with the actual fault type, thereby obtaining the cause probability distribution data. For example, if the consistency probability of the first humidity sensitivity factor and the actual fault type is 0.8, and the consistency probability of the second humidity sensitivity factor and the actual fault type is 0.6, then the first humidity sensitivity factor can be given a higher weight coefficient, such as 0.4, and the second humidity sensitivity factor can be given a lower weight coefficient, such as 0.2.

[0125] Step S56: Use the causal probability distribution data to perform causal reasoning adjustment on the fault-related knowledge graph to obtain a fault causal mechanism diagram.

[0126] This embodiment uses the causal probability distribution data to perform causal reasoning adjustment on the fault-related knowledge graph, that is, according to the degree of influence of different humidity sensitivity factors on different fault types, the causal relationship in the knowledge graph is strengthened or weakened, and a corresponding fault causal mechanism diagram is generated. For example, if the first humidity sensitivity factor has a higher degree of influence on type A faults, and the second humidity sensitivity factor has a lower degree of influence on type B faults, then the causal edge from the first humidity sensitivity factor to type A faults can be added to the knowledge graph, and the causal edge from the second humidity sensitivity factor to type B faults can be reduced.

[0127] The present invention establishes a humidity-sensitive fault probability matrix through an association matrix, and this matrix can reflect the relationship between different humidity-sensitive factors and fault types, providing a basis for subsequent analysis and diagnosis. The Bayesian posterior probability calculation can estimate the probability of the fault type corresponding to each humidity-sensitive factor according to the humidity-sensitive fault probability matrix, thereby obtaining the posterior probability matrix. This matrix can provide the possibility of each humidity-sensitive factor under different fault types, providing a basis for subsequent diagnosis and verification. The fault type diagnosis can judge the fault type of each humidity-sensitive factor according to the posterior probability matrix, thereby obtaining the humidity-sensitive factor diagnosis result. This result can provide the fault type to which each humidity-sensitive factor belongs, for example, the first humidity-sensitive factor belongs to a circuit short circuit fault, the second humidity-sensitive factor belongs to a battery overcharge fault, etc., providing information for subsequent verification and processing. The fault type verification can use the fault cause data and the fault type data to verify the humidity-sensitive factor diagnosis result, thereby obtaining the verified fault probability distribution. The distribution can provide the accuracy and credibility of the diagnosis results of each humidity sensitive factor. For example, the accuracy of the diagnosis results of the first humidity sensitive factor is 95%, and the accuracy of the diagnosis results of the second humidity sensitive factor is 80%, etc., which provides a reference for subsequent weight adjustment and processing. The weight adjustment can use the verified fault probability distribution to assign weights to each humidity sensitive factor, thereby obtaining the causal probability distribution data. This data can provide the role and influence of each humidity sensitive factor in causing the final failure. For example, the weight of the first humidity sensitive factor is 0.6, and the weight of the second humidity sensitive factor is 0.3, etc., which provides a basis for subsequent causal reasoning and processing. The causal reasoning adjustment can use the causal probability distribution data to adjust the fault association knowledge graph to obtain the fault causal mechanism diagram. This diagram can provide the causal relationship and path between various humidity sensitive factors in the storage device and the final failure. For example, the first humidity sensitive factor causes a circuit short circuit, and the circuit short circuit causes damage to the storage device, etc., which provides guidance for subsequent processing and repair.

[0128] Preferably, step S56 includes the following steps:

[0129] Step S561: Calculate the node weights of the corresponding fault type data according to the cause probability distribution data, so as to obtain the Bayesian probability of the fault;

[0130] This embodiment performs statistical analysis on the fault type data, calculates the occurrence frequency and severity of each fault type, and thus obtains the cause probability distribution data of the fault type. Then, according to the cause probability distribution data, a corresponding weight value is assigned to each fault type node in the fault association knowledge graph to indicate the possibility of the occurrence of the fault type, thereby obtaining the Bayesian probability of the fault.

[0131] Step S562: using the fault Bayesian probability to perform Markov random field probability propagation adjustment on the fault-related knowledge graph, thereby obtaining the fault knowledge graph;

[0132] This embodiment uses the Bayesian probability of fault as the initial condition to perform Markov random field probability propagation adjustment on the fault association knowledge graph. Specifically, by iteratively updating the probability value of each node so that it satisfies the local Markov property and global stationary property of the Markov random field, a fault knowledge graph is obtained. The fault knowledge graph can reflect the degree of association and impact range between different fault types.

[0133] Step S563: Perform causal reasoning analysis on the fault knowledge graph to obtain a fault causal mechanism diagram.

[0134] This embodiment uses a Bayesian network or other causal reasoning methods to infer the causal relationship between different fault types based on the node probability values ​​and edge weight values ​​in the fault knowledge graph, and represents it as a directed edge. Finally, a fault causal mechanism diagram is obtained, which can clearly show the cause and result of the fault.

[0135] The present invention can obtain the Bayesian probability of faults through node weight calculation. This probability can reflect the possibility of occurrence of different fault types under different humidity sensitivity factors, providing a basis for subsequent analysis and reasoning. The Markov random field probability propagation adjustment can use the fault Bayesian probability to adjust the fault association knowledge graph, thereby obtaining the fault knowledge graph. This graph can provide the association relationship and probability distribution between various humidity sensitivity factors and fault types in the storage device. For example, the association probability of the first humidity sensitivity factor and the circuit short circuit fault is 0.8, and the association probability of the second humidity sensitivity factor and the battery overcharge fault is 0.6, etc., providing information for subsequent reasoning and analysis. Causal reasoning analysis can perform causal reasoning on the fault knowledge graph, thereby obtaining a fault causal mechanism diagram, which can provide the causal relationship and path between various humidity sensitivity factors in the storage device and the final fault, for example, the first humidity sensitivity factor causes a circuit short circuit -> the circuit short circuit causes damage to the storage device, etc., providing guidance for subsequent processing and repair.

[0136] Preferably, step S7 comprises the following steps:

[0137] Step S71: filling missing values ​​of the fault precursor working data according to the working parameter feature vector, so as to obtain the precursor working data;

[0138] This embodiment performs data cleaning and data quality assessment on the fault precursor working data and finds that there are some missing values ​​or abnormal values ​​therein, which may affect subsequent analysis and modeling. Therefore, according to the working parameter feature vector, interpolation, regression or other appropriate methods are used to fill the missing values ​​of the fault precursor working data to obtain the precursor working data.

[0139] Step S72: Analyze the evolution law of the precursor work data over time, so as to obtain evolution law data;

[0140] This embodiment mines the time-varying regularity of each working parameter in the precursor working data by using time series analysis, trend analysis or other appropriate methods, thereby obtaining evolution regularity data, which can reflect the changing trends and periodicity of different working parameters.

[0141] Step S73: performing evolution behavior analysis on the evolution law data to obtain an evolution behavior pattern;

[0142] This embodiment uses cluster analysis, correlation analysis or other appropriate methods to extract and classify the evolution behavior characteristics of each operating parameter in the evolution law data, thereby obtaining an evolution behavior pattern. The evolution behavior pattern can reflect the evolution characteristics and mutual relationships of different operating parameters.

[0143] Step S74: performing evolution trajectory learning on the evolution behavior pattern to obtain a working parameter prediction model.

[0144] This embodiment uses a neural network, support vector machine or other suitable methods to model and learn the evolution trajectory of each working parameter in the evolution behavior pattern, thereby obtaining a working parameter prediction model. The working parameter prediction model can predict future working parameter values ​​based on current and historical working parameter values.

[0145] The present invention can obtain precursor working data through missing value filling. This data can fill in the missing values ​​in the working parameter feature vector, making the data more complete and accurate, providing a basis for subsequent analysis and prediction. The time evolution law analysis can perform time series analysis on the precursor working data, so as to obtain evolution law data. This data can reflect the changing trend and law of the working parameters over time, such as the change curves of parameters such as temperature, pressure, and current, providing information for subsequent analysis and prediction. The evolution behavior analysis can perform behavior pattern recognition on the evolution law data, so as to obtain the evolution behavior pattern. This pattern can provide the behavior characteristics and types of the working parameters in different time periods, such as rising, falling, stable, fluctuating and other behaviors, providing a basis for subsequent prediction and learning. The evolution trajectory learning can perform machine learning on the evolution behavior pattern, so as to obtain the working parameter prediction model. This model can predict and estimate future data based on historical data and current data, such as predicting the change range and possible values ​​of parameters such as temperature, pressure, and current in the next hour, providing guidance for subsequent processing and control.

[0146] Preferably, step S9 comprises the following steps:

[0147] Step S91: Obtaining historical failure times and aging degree data of the IGBT device;

[0148] This embodiment obtains the historical failure times and aging degree data of the IGBT device from the manufacturer or user of the IGBT device. The historical failure times indicate the number of failures of the IGBT device during use, and the aging degree data indicate the degree of performance degradation caused by the temperature, voltage, current and other factors of the IGBT device during use.

[0149] Step S92: Evaluate the durability of the working capability of the IGBT device according to the number of historical failures, thereby obtaining a reliability index;

[0150] This embodiment evaluates the durability of the working ability of the IGBT device according to the number of historical failures, thereby obtaining a reliability index. Specifically, by using Weibull distribution, exponential distribution or other appropriate methods, statistical analysis is performed on the number of historical failures to calculate the average life, failure rate, reliability and other indicators of the IGBT device, thereby obtaining a reliability index. The reliability index can reflect the working ability and stability of the IGBT device.

[0151] Step S93: evaluating the humidity sensitivity of the IBGT device according to the aging degree data, thereby obtaining a humidity sensitivity index;

[0152] This embodiment evaluates the humidity sensitivity of the IGBT device based on the aging degree data, thereby obtaining a humidity sensitivity index. Specifically, by using linear regression, nonlinear regression or other appropriate methods, the aging degree data is fitted and analyzed to obtain a relationship function between the aging degree of the IGBT device and humidity, thereby obtaining a humidity sensitivity index. The humidity sensitivity index can reflect the working performance and life changes of the IGBT device under different humidity conditions.

[0153] Step S94: performing a second accurate prediction of the prediction time data according to the reliability index and the humidity sensitivity index, thereby obtaining a second prediction time;

[0154] This embodiment makes a secondary accurate prediction of the prediction time data according to the reliability index and the humidity sensitivity index, thereby obtaining the secondary prediction time. Specifically, by using Bayesian reasoning, Monte Carlo simulation or other suitable methods, the prediction time data is probability analyzed and error corrected, and the influencing factors of the reliability index and the humidity sensitivity index are considered to obtain the secondary prediction time. The secondary prediction time can more accurately predict the remaining life and failure time of the IGBT device.

[0155] Step S95: performing production line asset evaluation on the working area where the IGBT device is located, thereby obtaining regional asset value data;

[0156] In this embodiment, the value of production line equipment, raw materials, products, etc. in the work area is estimated by using asset evaluation methods, such as depreciation method, income method or other appropriate methods, so as to obtain regional asset value data. The regional asset value data can reflect the production capacity and economic benefits of the work area.

[0157] Step S96: performing fault loss asset assessment according to the prediction type and the regional asset value data, thereby obtaining fault loss value data;

[0158] This embodiment estimates the losses such as production line shutdown, product scrapping, and repair costs that may be caused by the failure according to the prediction type, such as the failure type, failure degree, and failure range, so as to obtain the failure loss value data. The failure loss value data can reflect the impact and damage of the failure on the working area.

[0159] Step S97: predict and evaluate the fault urgency according to the fault loss value data and the secondary prediction time, so as to obtain risk level data; and control the IGBT device accordingly using a preset drive circuit according to the risk level data.

[0160] This embodiment uses a risk assessment method, such as a risk matrix method, a risk index method, or other appropriate methods, to comprehensively analyze the fault loss value data and the secondary prediction time, obtain the possibility and severity of the fault, and divide it into different risk levels, thereby obtaining risk level data. The risk level data can reflect the urgency and processing priority of the fault. According to the risk level data, the IGBT device is controlled accordingly using a preset drive circuit, such as adjusting operating parameters, switching backup devices, or stopping work, so as to reduce or avoid losses caused by the fault.

[0161] The present invention can obtain the reliability index of the IGBT device through the number of historical failures. This index can reflect the durability of the working ability of the IGBT device. For example, the more historical failures, the lower the reliability index, which provides a basis for subsequent analysis and prediction. The aging degree data can obtain the humidity sensitivity index of the IGBT device. This index can reflect the sensitivity of the IGBT device to humidity changes. For example, the higher the aging degree, the higher the humidity sensitivity index, which provides information for subsequent analysis and prediction. The secondary precise prediction can adjust the prediction time data according to the reliability index and the humidity sensitivity index to obtain the secondary prediction time. This time can provide the time point and range where the IGBT device may fail. For example, the secondary prediction time is between September 20, 2023 and September 30, 2023, which provides a basis for subsequent evaluation and control. The production line asset evaluation can calculate the asset value of the working area where the IGBT device is located, thereby obtaining regional asset value data. This data can provide The production capacity and value of the working area, for example, the regional asset value data is RMB 10 million, which provides a reference for subsequent evaluation and control. The fault loss asset assessment can calculate the fault loss value according to the prediction type and regional asset value data, so as to obtain the fault loss value data. This data can provide the economic loss that may be caused when the IGBT device fails. For example, the fault loss value data is RMB 2 million, which provides a basis for subsequent evaluation and control. The fault urgency prediction and assessment can divide the risk level according to the fault loss value data and the secondary prediction time, so as to obtain the risk level data. This data can provide the urgency and processing priority when the IGBT device fails. For example, the risk level data is a high risk level, which provides guidance for subsequent control. The drive circuit control can use the preset drive circuit to control the IGBT device accordingly according to the risk level data, so as to ensure the normal operation of the IGBT device or stop working in time to avoid greater losses.

[0162] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0163] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A state detection circuit control method for an IGBT device, characterized in that: The following steps are involved: Step S1: wirelessly monitor the working parameters of the IGBT device through surface acoustic wave sensor technology to obtain real-time working parameters; use a nano humidity sensor to perform nano ultraviolet excitation treatment on the ambient air to obtain real-time humidity data; Step S2: Acquire a historical fault data set, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor work data; Conduct time-frequency joint analysis on the fault precursor working data to obtain the working parameter feature vector; Step S3: mining the internal correlation relationship of the fault humidity data, fault timestamp data, fault type data and fault precursor work data according to the historical fault data set, so as to obtain the fault correlation knowledge graph; Step S4: dividing the fault humidity data into humidity ranges according to the fault timestamp data, thereby obtaining a first humidity range, a second humidity range, and a third humidity range; performing humidity sensitivity calculations on the first humidity range, the second humidity range, and the third humidity range, respectively, thereby obtaining a first humidity sensitivity factor, a second humidity sensitivity factor, and a third humidity sensitivity factor; Step S5: diagnose the fault types of the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor according to the fault cause data and the fault type data, thereby obtaining causal probability distribution data; perform causal reasoning adjustment on the fault association knowledge graph using the causal probability distribution data, thereby obtaining a fault causal mechanism diagram; Step S6: predicting the fault type of the real-time working parameters based on the fault cause-effect mechanism diagram and the real-time humidity data, thereby obtaining the predicted type; Step S7: performing evolution trajectory behavior learning on the fault precursor working data according to the working parameter feature vector, thereby obtaining a working parameter prediction model; Step S8: using the working parameter prediction model to predict the fault time range of the prediction type, thereby obtaining prediction time data; Step S9: predicting and evaluating the fault urgency of the prediction type and prediction time data according to the fault cause and effect mechanism diagram, thereby obtaining risk level data; The IGBT device is controlled accordingly using a preset drive circuit according to the risk level data.

2. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: using a surface acoustic wave sensor to collect acoustic wave data of the working current and voltage of the IGBT device in real time through a preset wireless module, thereby obtaining a working acoustic wave signal; Step S12: performing wavelet transform on the working sound wave signal to obtain a sound wave feature matrix; Step S13: performing principal component analysis on the acoustic wave feature matrix to obtain a principal component feature vector; Step S14: standardizing the principal component feature vector to obtain real-time voltage and current data; Step S15: obtaining real-time temperature data by a thermal paint method; using the real-time temperature data, real-time voltage and current data as real-time working parameters; Step S16: Exciting the hygroscopic material of the nano humidity sensor by ultraviolet irradiation, and monitoring the resistance change of the nano humidity sensor in real time, thereby obtaining a resistance signal; Step S17: Calculate the Shannon entropy change of the resistance signal to obtain real-time humidity data.

3. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: The Shannon entropy change calculation in step S17 is to calculate the change of the resistance signal through the humidity Shannon entropy detection formula, so as to obtain real-time humidity data, wherein the humidity Shannon entropy detection formula is as follows: In the formula, H(R) is the real-time humidity data, R is the average value of the resistance signal, n is the total number of resistance signals, i is the serial number of the resistance signal value, σ is the standard deviation of the resistance signal, and R i is the value of the i-th resistance signal, α is the sine adjustment coefficient, β is the sine adjustment frequency, γ is the cosine adjustment coefficient, and δ is the cosine adjustment frequency.

4. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: Acquire a historical fault data set, wherein the historical fault data set includes fault humidity data, fault cause data, fault timestamp data, fault type data, and fault precursor work data; Step S22: dividing the fault time into fault time periods according to the fault timestamp data, thereby obtaining a fault time period set, wherein the fault time period set includes multiple fault time periods; Step S23: performing fault count for each fault time period in the fault time period set, thereby obtaining fault count data; Step S24: calculating the probability of failure occurrence for the number of failures in each failure time period in the failure time period set, thereby obtaining the failure probability of the time period; Step S25: performing multi-level decomposition on the fault precursor working data to obtain a frequency band signal component set, wherein the frequency band signal component set includes multiple frequency band signal components; Step S26: extracting the time-frequency energy features of each frequency band signal component in the frequency band signal component set, thereby obtaining an energy feature matrix; Step S27: Probability weighting is performed on the failure probabilities in different time periods by using the maximum likelihood estimation method, thereby obtaining a failure probability weight vector; Step S28: perform element-wise multiplication of the fault probability weight vector and the energy characteristic matrix to obtain a working parameter characteristic vector.

5. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: calculating the fault interval duration of the fault timestamp data, thereby obtaining an interval duration data set, wherein the interval duration data set includes a plurality of interval duration data; Step S32: taking each fault in the historical fault data set as a fault sample, thereby obtaining a fault sample set; Step S33: hierarchically clustering the fault sample set using the interval duration data set, thereby obtaining a fault clustering result; Step S34: Calculate the Pearson coefficient for the fault humidity data, fault timestamp data, fault type data, and fault precursor work data corresponding to each category in the fault clustering result, so as to obtain the intra-class correlation coefficient; Step S35: using a preset correlation coefficient threshold to perform critical value filtering on the intra-class correlation coefficient, thereby obtaining strongly correlated variable pairs; Step S36: performing association analysis on strongly correlated variable pairs of each category, thereby obtaining frequent item sets within the category; Step S37: performing association rule mining on frequent item sets within the class to obtain association rules; Step S38: construct a knowledge graph for strongly correlated variable pairs and association rules, thereby obtaining a fault-related knowledge graph.

6. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: sorting the fault humidity data by time according to the fault timestamp data, thereby obtaining humidity sorting data; Step S42: curve fitting and visualization of humidity sorting data to obtain humidity evolution trajectory; Step S43: Analyze the change pattern of the humidity evolution trajectory to obtain the humidity evolution law and humidity sensitive range; Step S44: dividing the humidity data into temperature ranges according to the humidity evolution law and the humidity sensitive interval, thereby obtaining the first humidity range, the second humidity range and the third humidity range; Step S45: dividing the historical fault data set into samples according to the temperature range according to the first humidity range, the second humidity range and the third humidity range, thereby obtaining the first sample set, the second sample set and the third sample set; Step S46: performing humidity-induced fault counts on the first sample set, the second sample set, and the third sample set according to the corresponding fault cause data, thereby obtaining corresponding humidity-induced frequencies; Step S47: Calculate the humidity sensitivity of the humidity-induced frequency in each sample set, so as to obtain a first humidity sensitivity factor, a second humidity sensitivity factor and a third humidity sensitivity factor.

7. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: establishing a correlation matrix between the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor and the fault type data, thereby obtaining a humidity-sensitive fault probability matrix; Step S52: performing Bayesian posterior probability calculation on the fault type data corresponding to the first humidity sensitivity factor, the second humidity sensitivity factor and the third humidity sensitivity factor based on the humidity sensitivity failure probability matrix, thereby obtaining a posterior probability matrix; Step S53: diagnose the fault type of each humidity sensitive factor according to the posterior probability matrix, so as to obtain the humidity sensitive factor diagnosis result; Step S54: using the fault cause data and the fault type data to verify the fault type of the humidity sensitive factor diagnosis result, so as to obtain a verified fault probability distribution; Step S55: using the verification fault probability distribution to adjust the weights of the first humidity sensitivity factor, the second humidity sensitivity factor, and the third humidity sensitivity factor, thereby obtaining causal probability distribution data; Step S56: Use the causal probability distribution data to perform causal reasoning adjustment on the fault-related knowledge graph to obtain a fault causal mechanism diagram.

8. The state detection circuit control method of the IGBT device according to claim 7, characterized in that: Step S56 includes the following steps: Step S561: Calculate the node weights of the corresponding fault type data according to the cause probability distribution data, so as to obtain the Bayesian probability of the fault; Step S562: using the fault Bayesian probability to perform Markov random field probability propagation adjustment on the fault-related knowledge graph, thereby obtaining the fault knowledge graph; Step S563: Perform causal reasoning analysis on the fault knowledge graph to obtain a fault causal mechanism diagram.

9. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S7 includes the following steps: Step S71: filling missing values ​​of the fault precursor working data according to the working parameter feature vector, so as to obtain the precursor working data; Step S72: Analyze the evolution law of the precursor work data over time, so as to obtain evolution law data; Step S73: performing evolution behavior analysis on the evolution law data to obtain an evolution behavior pattern; Step S74: performing evolution trajectory learning on the evolution behavior pattern to obtain a working parameter prediction model.

10. The state detection circuit control method of the IGBT device according to claim 1, characterized in that: Step S9 includes the following steps: Step S91: Obtaining historical failure times and aging degree data of the IGBT device; Step S92: Evaluate the durability of the working capability of the IGBT device according to the number of historical failures, thereby obtaining a reliability index; Step S93: evaluating the humidity sensitivity of the IBGT device according to the aging degree data, thereby obtaining a humidity sensitivity index; Step S94: performing a second accurate prediction of the prediction time data according to the reliability index and the humidity sensitivity index, thereby obtaining a second prediction time; Step S95: performing production line asset evaluation on the working area where the IGBT device is located, thereby obtaining regional asset value data; Step S96: performing fault loss asset assessment according to the prediction type and the regional asset value data, thereby obtaining fault loss value data; Step S97: predict and evaluate the fault urgency according to the fault loss value data and the secondary prediction time, so as to obtain risk level data; and control the IGBT device accordingly using a preset drive circuit according to the risk level data.

Images