IGBT driver fault prediction method and safety system

Abstract

The invention discloses an IGBT driver fault prediction method and a safety system, and relates to the technical field of power electronics, and the method comprises the steps: constructing a dynamic causal reasoning matrix based on degradation characteristics, and carrying out the reverse reasoning of an accidental fault mode of an IGBT driver through the dynamic causal reasoning matrix; establishing a multi-stage fault prediction model based on the accidental fault mode of the IGBT driver, and predicting the fault occurrence frequency of each life cycle stage of the IGBT driver by using the multi-stage fault prediction model; and identifying a vulnerable element in the IGBT driver based on the fault occurrence frequency, performing fault tracing analysis on the vulnerable element, and optimizing the multi-stage fault prediction model according to a tracing analysis result. According to the method, the causal structure model with the dynamic evolution ability and the causal tracing ability is constructed, the search space of the causal relationship is effectively compressed, and the accuracy and expression ability of the causal path are improved.

Description

Technical Field

[0001] The present invention relates to the field of power electronics technology, and in particular to an IGBT driver fault prediction method and a safety system. Background Art

[0002] An IGBT driver is a key module for controlling and driving the normal operation of IGBTs (insulated-gate bipolar transistors). It is typically used in high-voltage, high-current applications such as motor control, electric vehicles, rail transit, frequency converters, and photovoltaic inverter systems. Its primary function is to apply a suitable voltage waveform to the IGBT's gate, enabling reliable switching between on and off, and providing protection against overcurrent, overvoltage, and short-circuit conditions.

[0003] Because IGBT drivers are subject to temperature rise, thermal cycling, electrical stress, and aging over long periods of operation, their internal components, such as gate resistors, capacitors, drive transformers, and optocouplers, may gradually degrade. While initial degradation may not be noticeable, reaching a critical state can lead to serious failures such as false triggering, conduction failure, and short circuits. The key to predictive maintenance lies in identifying abnormal trends in IGBT drivers before degradation turns into failure.

[0004] Many early signs of degradation are manifested in tiny cycle-level waveform deviations, such as a slight increase in on-state voltage, an extension of the turn-off tail current, and irregular jitter in the gate waveform. If these subtle degradation characteristics cannot be captured, the fault prediction model will not be able to accurately distinguish the subtle differences between the different degradation stages of the IGBT driver, and thus will not be able to issue timely warnings before a fault occurs, which will in turn affect the overall prediction accuracy of the fault prediction model.

[0005] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0006] In response to the problems in the related art, the present invention proposes an IGBT driver fault prediction method and a safety system to overcome the above technical problems existing in the existing related art.

[0007] To this end, the specific technical solutions adopted in the present invention are as follows:

[0008] According to one aspect of the present invention, a method for predicting faults of an IGBT driver is provided, the method comprising:

[0009] S1. Extract degradation features from the operating data of the IGBT driver, build a dynamic causal reasoning matrix based on the degradation features, and use the dynamic causal reasoning matrix to reversely reason about the occasional failure mode of the IGBT driver;

[0010] S2. Establish a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver, and use the multi-stage fault prediction model to predict the failure frequency of the IGBT driver in each life cycle stage;

[0011] S3. Identify vulnerable components in the IGBT driver based on the fault frequency, perform fault source analysis on the vulnerable components, and optimize the multi-stage fault prediction model based on the source analysis results.

[0012] Preferably, the degradation characteristics are extracted from the operating data of the IGBT driver, a dynamic causal reasoning matrix is constructed based on the degradation characteristics, and the dynamic causal reasoning matrix is used to reversely reason about the occasional failure mode of the IGBT driver, including:

[0013] S11. Extract the degradation characteristics of each component from the operating data of the IGBT driver and use the causal discovery algorithm to explore the causal relationship between the degradation characteristics and external environmental factors;

[0014] S12. Construct an accelerated degradation model based on external environmental factors, and simulate the dominant degradation trajectory of the IGBT driver under different external environmental interferences through the accelerated degradation model;

[0015] S13. Introduce the dominant degradation trajectory into the dynamic causal reasoning matrix, and reversely deduce the occasional failure mode of the IGBT driver through the dynamic causal reasoning matrix.

[0016] Preferably, an accelerated degradation model based on external environmental factors is constructed, and the leading degradation trajectory of the IGBT driver under different external environmental interferences is simulated by the accelerated degradation model, including:

[0017] S121. Constructing a cumulative degradation function of the IGBT driver based on the causal relationship between the degradation characteristics and the external environmental factors, and finding an extreme value of the cumulative degradation function to obtain parameters to be estimated;

[0018] S122. Constructing an accelerated degradation model based on environmental factors based on the parameters to be estimated, and inputting different external environmental interference combinations into the accelerated degradation model;

[0019] S123. Based on the output results of the accelerated degradation model, a Monte Carlo technique is used to simulate and generate a set of degradation trajectories of the performance parameters of each component of the IGBT driver evolving over time, and a dominant degradation trajectory is iteratively searched from the set of degradation trajectories.

[0020] Preferably, the dominant degradation trajectory is introduced into the dynamic causal reasoning matrix, and the occasional failure mode of the IGBT driver is reversely deduced through the dynamic causal reasoning matrix, including:

[0021] S131. Perform an autoregressive analysis on the dominant degradation trajectory of each component, and based on the autoregressive analysis results, use the degradation characteristics corresponding to the dominant degradation trajectory as child variables and the external environmental factors as parent variables;

[0022] S132. Construct a dynamic causal inference matrix based on the strength of the causal relationship between the independent variable and the parent variable, and randomly select a causal path traced back from the independent variable using a sparse matrix decomposition algorithm, delete duplicate variables in the same causal path, and obtain an optimized dynamic causal inference matrix.

[0023] S133. Based on the optimized dynamic causal reasoning matrix, the failure mode of the IGBT driver is reversed, and the failure mode category is identified using the pre-built graph convolutional neural network to screen out the occasional failure mode of the IGBT driver.

[0024] Preferably, the causal path randomly selected from the independent variable and traced back in combination with the sparse matrix decomposition algorithm includes:

[0025] Obtain all causal paths traced from the independent variables in the dynamic causal inference matrix, and introduce elements of sparse constraint singular vectors to ensure the sparseness of the dynamic causal inference matrix;

[0026] The dynamic causal inference matrix after sparseness is approximated into several low-rank matrices, and the rank of the low-rank matrix is reduced by selecting the principal components through the principal hierarchy analysis method to obtain a low-rank sparse causal matrix;

[0027] An independent variable is randomly selected as the starting point, and a sparse causal path starting from the current independent variable is selected through the strong connection in the sparse causal matrix.

[0028] Preferably, a multi-stage fault prediction model is established based on the occasional failure mode of the IGBT driver. The multi-stage fault prediction model is used to predict the failure frequency of each life cycle stage of the IGBT driver, including:

[0029] S21. Assign risk weights to the occasional failure modes and divide the IGBT driver into several life cycle stages based on the risk weights. The life cycle stages of the IGBT driver include the initial degradation stage, the failure acceleration stage, and the failure critical stage.

[0030] S22. Construct a fault prediction model for each life cycle stage of the IGBT driver, and predict the frequency of failures in each life cycle stage based on the output results of each fault prediction model.

[0031] Preferably, a fault prediction model for each life cycle stage of the IGBT driver is constructed, including:

[0032] Based on the assigned accidental failure mode weights, the total failure frequency index of the IGBT driver is calculated. A stage identifier is then built based on the total failure frequency index to determine the current life cycle stage of the IGBT driver.

[0033] If the IGBT driver is in the initial degradation stage, the long short-term memory network is selected as the model framework to build the initial fault prediction model;

[0034] If the IGBT driver is in the failure acceleration stage, the degradation rate time point analysis module is introduced into the initial fault prediction model to obtain the accelerated fault prediction model;

[0035] If the IGBT driver is in the critical failure stage, a survival life analysis module is introduced into the accelerated failure prediction model to obtain a critical failure prediction model.

[0036] Preferably, a stage identifier is constructed based on the total fault frequency indicator, and the stage identifier is used to determine the current life cycle stage of the IGBT driver. The stage identifier includes:

[0037] Using occasional failure patterns as training data, a parameter estimation algorithm is used to train two sets of hidden Markov models. The hidden states in each hidden Markov model are mapped to the life cycle stages of the IGBT driver.

[0038] The life cycle state of the IGBT driver is estimated, and the state estimation results output by two sets of hidden Markov models are fused through DS evidence theory.

[0039] The current life cycle stage of the IGBT driver and the transition probability from one life cycle stage to another are calculated based on the state estimation results.

[0040] Preferably, the expression of the accelerated degradation model is:

[0041]

[0042] Where A total (t) represents the comprehensive acceleration effect of IGBT driver degradation at time t; A0 represents the initial degradation reference value; T(t) represents the temperature at time t; E a represents temperature sensitivity; RH(t) represents humidity at time t; RH0 represents reference humidity; V(t) represents voltage stress at time t; V0 represents reference voltage; S(t) represents vibration stress at time t; S0 represents reference vibration; k B represents the Boltzmann constant; m represents the humidity acceleration index; p represents the vibration acceleration index; and n represents the voltage acceleration index.

[0043] According to another aspect of the present invention, there is also provided an IGBT driver safety system, the system comprising:

[0044] A fault identification module is used to extract degradation features from the operating data of the IGBT driver, construct a dynamic causal reasoning matrix based on the degradation features, and use the dynamic causal reasoning matrix to reversely infer the occasional failure mode of the IGBT driver;

[0045] A fault prediction module is used to establish a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver, and use the multi-stage fault prediction model to predict the failure frequency of the IGBT driver in each life cycle stage;

[0046] The fault analysis module is used to identify vulnerable components in the IGBT driver based on the frequency of fault occurrence, perform fault source analysis on the vulnerable components, and optimize the multi-stage fault prediction model based on the source analysis results.

[0047] The beneficial effects of the present invention are:

[0048] 1. The present invention extracts degradation features from the operating data of the IGBT driver, constructs a dynamic causal reasoning matrix based on the degradation features, and then uses the matrix to reversely infer occasional failure modes. It can extract the dominant degradation trajectory that is highly correlated with external factors from the degradation behavior of each driver component under various environmental disturbances, and construct a causal structure model with dynamic evolution and causal traceability capabilities. Combining sparse matrix decomposition and path optimization, it effectively compresses the search space of causal relationships, improves the accuracy and expression ability of causal paths, and efficiently identifies and classifies fault modes in the dynamic causal structure, thereby significantly enhancing the ability to accurately identify occasional failures of the IGBT driver, improving the reliability, interpretability and generalization ability of fault prediction, and providing data support and causal basis for efficient health management and advance maintenance decisions.

[0049] 2. The present invention establishes a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver and predicts the frequency of failure in each life cycle stage. It can achieve detailed modeling and dynamic tracking of the entire process of the driver from initial degradation to critical failure. The life cycle stages are divided by risk weights, and the accuracy of stage identification is improved by combining the hidden Markov model and DS evidence theory. Differentiated prediction strategies are then introduced for different stages, such as long-short-term memory networks to capture initial characteristics, degradation rate analysis modeling acceleration stages, and survival life assessment to characterize critical stages. This effectively improves the model's adaptability and prediction accuracy to the fault development process, enhances the prediction model's pertinence and foresight, and provides reliable support for key equipment health management and preventive maintenance.

[0050] 3. The present invention effectively reduces the probability of electrical fires by identifying abnormal current, voltage jumps and overheating trends in multiple life cycle stages of the equipment, such as initial degradation, accelerated failure and critical failure, and avoids fire accidents caused by problems such as thermal runaway and short circuit of the device. At the same time, accurate prediction of the life cycle status also helps to judge the hidden dangers that lead to electrical control failure in advance, and timely activate the power-off protection device before detecting that the equipment enters the critical stage, to prevent the risk of electric shock caused by problems such as insulation damage and abnormal potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 is a flow chart of a method for predicting faults of an IGBT driver according to an embodiment of the present invention;

[0053] Figure 2 This is a principle block diagram of an IGBT driver safety system according to an embodiment of the present invention.

[0054] In the picture:

[0055] 1. Fault identification module; 2. Fault prediction module; 3. Fault analysis module. DETAILED DESCRIPTION

[0056] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and the advantages of the present invention.

[0057] According to an embodiment of the present invention, a method and a safety system for predicting faults of an IGBT driver are provided.

[0058] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the IGBT driver fault prediction method of an embodiment of the present invention, the method includes:

[0059] S1. Extract degradation features from the operating data of the IGBT driver, build a dynamic causal reasoning matrix based on the degradation features, and use the dynamic causal reasoning matrix to reversely infer the occasional failure mode of the IGBT driver.

[0060] The degradation characteristics are extracted from the operating data of the IGBT driver. A dynamic causal reasoning matrix is constructed based on the degradation characteristics. The dynamic causal reasoning matrix is used to reversely infer the occasional failure modes of the IGBT driver, including:

[0061] S11. Extract the degradation characteristics of each component from the operating data of the IGBT driver, and use the causal discovery algorithm to explore the causal relationship between the degradation characteristics and external environmental factors.

[0062] It should be noted that the degradation characteristics of each component are extracted from the operating data of the IGBT driver, and the causal relationship between the degradation characteristics and external environmental factors is mined using the causal discovery algorithm, including:

[0063] Time series data of IGBT drivers under different operating conditions are collected and preprocessed. Based on the typical degradation performance of each key component, degradation-sensitive features such as increased on-state voltage drop, prolonged off-time, change in current rise rate, and temperature drift are selected, and a time series feature set is constructed using sliding window analysis technology. The causal structure between variables is learned using the selected causal discovery algorithm to obtain a causal relationship network represented by a directed acyclic graph. The causal path between the degradation features and environmental factors is mined in the causal relationship network, and significant causal channels are screened based on the edge weights.

[0064] It should be noted that the causal discovery algorithm is a GES algorithm based on scoring search, which specifically includes: initializing an empty graph as the current causal graph structure; adopting a two-stage greedy search strategy, in the first stage, the scoring function is optimized by gradually adding edges, and each time the legal edges that can maximize the score improvement are selected for addition, until the score can no longer be improved; in the second stage, the scoring function is optimized by gradually deleting edges, and each time the edges that can maximize the score improvement are selected for deletion, until the score can no longer be improved; using the Bayesian Information Criterion as the scoring function to evaluate the quality of the graph structure; and outputting an optimal equivalence class DAG structure.

[0065] Among them, the operating data of the IGBT driver mainly include gate voltage, current, voltage rise time, turn-on time, turn-off time, on-state voltage drop, switching frequency, drive voltage, chip junction temperature, module case temperature, current waveform, voltage waveform, short-circuit protection, etc.; external environmental factors mainly include ambient temperature, relative humidity, power supply voltage fluctuation, electromagnetic interference level, cooling wind speed, vibration acceleration, dust concentration, workload changes, grid harmonic disturbances, long-term thermal cycle times, mechanical shock events, etc.

[0066] S12. Construct an accelerated degradation model based on external environmental factors, and simulate the dominant degradation trajectory of the IGBT driver under different external environmental interferences through the accelerated degradation model.

[0067] Among them, an accelerated degradation model based on external environmental factors is constructed, and the dominant degradation trajectory of the IGBT driver under different external environmental interferences is simulated by the accelerated degradation model, including:

[0068] S121. Constructing a cumulative degradation function of the IGBT driver based on the causal relationship between the degradation characteristics and the external environmental factors, and finding an extreme value of the cumulative degradation function to obtain parameters to be estimated;

[0069] S122. Construct an accelerated degradation model based on environmental factors based on the parameters to be estimated, and input different external environmental interference combinations into the accelerated degradation model.

[0070] The expression of the accelerated degradation model is:

[0071]

[0072] Where A total (t) represents the comprehensive acceleration effect of IGBT driver degradation at time t; A0 represents the initial degradation reference value; T(t) represents the temperature at time t; E a represents temperature sensitivity; RH(t) represents humidity at time t; RH0 represents reference humidity; V(t) represents voltage stress at time t; V0 represents reference voltage; S(t) represents vibration stress at time t; S0 represents reference vibration; k B represents the Boltzmann constant; m represents the humidity acceleration index; p represents the vibration acceleration index; and n represents the voltage acceleration index.

[0073] S123. Based on the output results of the accelerated degradation model, a Monte Carlo technique is used to simulate and generate a set of degradation trajectories of the performance parameters of each component of the IGBT driver evolving over time, and a dominant degradation trajectory is iteratively searched from the set of degradation trajectories.

[0074] It's important to note that Monte Carlo simulations generate multiple possible degradation trajectories by randomly sampling changes in environmental factors and driver operating parameters. These trajectories reflect the potential degradation paths of the device under varying environmental interference. Within these trajectories, an iterative search method identifies the dominant degradation trajectories that influence overall system failure, helping to focus on the most critical degradation factors and paths. This approach effectively identifies the primary factors that could lead to long-term device failure, while reducing interference from non-dominant factors and improving prediction accuracy.

[0075] S13. Introduce the dominant degradation trajectory into the dynamic causal reasoning matrix, and reversely deduce the occasional failure mode of the IGBT driver through the dynamic causal reasoning matrix.

[0076] Among them, the dominant degradation trajectory is introduced into the dynamic causal reasoning matrix, and the occasional failure modes of the IGBT driver are reversely deduced through the dynamic causal reasoning matrix, including:

[0077] S131. Perform an autoregressive analysis on the dominant degradation trajectory of each component, and based on the autoregressive analysis results, use the degradation characteristics corresponding to the dominant degradation trajectory as child variables and the external environmental factors as parent variables;

[0078] S132. A dynamic causal inference matrix is constructed based on the strength of the causal relationship between the independent variable and the parent variable. The sparse matrix decomposition algorithm is used to randomly select a causal path traced back from the independent variable, and duplicate variables in the same causal path are deleted to obtain an optimized dynamic causal inference matrix.

[0079] It should be noted that the dynamic causal reasoning matrix based on the causal relationship strength between the independent variable and the parent variable includes:

[0080] The degradation characteristics corresponding to the dominant degradation trajectory are used as child variables (such as on-state voltage drop, turn-off delay time, and chip junction temperature), and external environmental factors are used as parent variables (such as ambient temperature, humidity, and voltage fluctuation).

[0081] Assume there are n parent variables and m independent variables, and construct an m×n dynamic causal inference matrix, which is expressed as:

[0082]

[0083] Where M represents the dynamic causal reasoning matrix; w ij It represents the influence strength of the j-th parent variable on the i-th child variable; n represents the number of parent variables; m represents the number of child variables.

[0084] Among them, the causal paths randomly selected from the independent variables by combining the sparse matrix decomposition algorithm include:

[0085] Obtain all causal paths traced from the independent variables in the dynamic causal inference matrix, and introduce elements of sparse constraint singular vectors to ensure the sparseness of the dynamic causal inference matrix;

[0086] The dynamic causal inference matrix after sparseness is approximated into several low-rank matrices, and the rank of the low-rank matrix is reduced by selecting the principal components through the principal hierarchy analysis method to obtain a low-rank sparse causal matrix;

[0087] An independent variable is randomly selected as the starting point, and a sparse causal path starting from the current independent variable is selected through the strong connection in the sparse causal matrix.

[0088] It should be noted that sparse matrix decomposition introduces sparse constraints to ensure that most elements in the matrix are zero, reduce redundant information, and eliminate unimportant or low-correlated variables in the causal path, thereby focusing on key variables and important causal relationships; by randomly selecting starting independent variables for tracing, causal paths can be efficiently explored in large-scale variable spaces, while avoiding the high computational cost of global search in traditional methods; the optimized dynamic causal reasoning matrix can more accurately reveal the dominant causal relationships in the system in reverse reasoning and fault pattern identification, thereby improving the model's predictive ability and adaptability to complex system behaviors.

[0089] S133. Based on the optimized dynamic causal reasoning matrix, the failure mode of the IGBT driver is reversed, and the failure mode category is identified using the pre-built graph convolutional neural network to screen out the occasional failure mode of the IGBT driver.

[0090] S2. A multi-stage fault prediction model is established based on the occasional failure mode of the IGBT driver, and the failure frequency of the IGBT driver in each life cycle stage is predicted using the multi-stage fault prediction model.

[0091] Among them, a multi-stage fault prediction model is established based on the occasional failure mode of the IGBT driver. The multi-stage fault prediction model is used to predict the failure frequency of each life cycle stage of the IGBT driver, including:

[0092] S21. Assign risk weights to occasional failure modes and divide the IGBT driver into several life cycle stages based on the risk weights. The life cycle stages of the IGBT driver include the initial degradation stage, the failure acceleration stage and the failure critical stage.

[0093] It should be noted that the life cycle stages of IGBT drivers are divided based on their performance degradation laws and fault development characteristics: the initial degradation stage refers to the period when the performance parameters of the equipment begin to show slight deviations in the early stage of operation but the function is normal, mainly manifested as reversible degradation such as gate voltage fluctuations and a slight increase in conduction loss; the failure acceleration stage is a critical period when the performance parameters show nonlinear accelerated degradation, manifested as irreversible changes such as a significant increase in switching losses and an increase in thermal resistance. At this time, the reliability of the equipment decreases rapidly; the failure critical stage is the final stage when the equipment is close to functional failure. The characteristic parameters exceed the safety threshold, manifested as a sudden drop in driving capability, frequent triggering of protection functions, etc. At this time, the equipment is in a dangerous state that may fail completely at any time.

[0094] S22. Construct a fault prediction model for each life cycle stage of the IGBT driver, and predict the frequency of failures in each life cycle stage based on the output results of each fault prediction model.

[0095] The fault prediction model for each life cycle stage of the IGBT driver is constructed, including:

[0096] Based on the assigned accidental failure mode weights, the total failure frequency index of the IGBT driver is calculated, and a stage identifier is constructed based on the total failure frequency index. The stage identifier is used to determine the current life cycle stage of the IGBT driver.

[0097] Among them, a stage identifier is constructed based on the total fault frequency index. The stage identifier is used to determine the current life cycle stage of the IGBT driver, which includes:

[0098] Using occasional failure patterns as training data, a parameter estimation algorithm is used to train two sets of hidden Markov models. The hidden states in each hidden Markov model are mapped to the life cycle stages of the IGBT driver.

[0099] The life cycle state of the IGBT driver is estimated, and the state estimation results output by two sets of hidden Markov models are fused through DS evidence theory.

[0100] The current life cycle stage of the IGBT driver and the transition probability from one life cycle stage to another are calculated based on the state estimation results.

[0101] It should be noted that the stage identifier is constructed based on the total fault frequency index. In actual applications, the stage identifier is used to determine the current life cycle stage of the IGBT driver.

[0102] Step 2: Based on the historical operating data of the IGBT driver (including degradation characteristics and environmental factors), an observation sequence is constructed to initialize the parameters of two sets of hidden Markov models, including the state transition matrix, the observation probability matrix, and the initial state distribution;

[0103] Step 3: Use the Baum-Welch algorithm to iteratively optimize the parameters of the two sets of models, and adjust the parameters through the expectation maximization process until convergence;

[0104] Step 4: After the model training is completed, the new observation sequence is decoded and the most likely hidden state sequence is output, i.e., the life cycle stage judgment. One hidden Markov model focuses on the electrical parameter characteristics, and the other hidden Markov model focuses on the thermodynamic parameters.

[0105] Step 5: Apply DS evidence theory to fuse the results of the two sets of hidden Markov models, calculate the state probability distribution of each model output, and then synthesize the joint confidence through the basic probability distribution function. Select the life cycle stage with the highest confidence after combination as the final judgment, and at the same time calculate the stage transition probability matrix to predict the evolution trend of the next stage.

[0106] It should be noted that the Baum-Welch algorithm is an expectation-maximization algorithm used to train hidden Markov models. It maximizes the conditional probability of a given observation sequence by iteratively estimating the hidden state transition probability and observation probability. The training process includes: E-step calculating the hidden state probability distribution at each time point, and M-step updating the model parameters based on these probabilities. The two sets of hidden Markov models trained using the expectation-maximization algorithm correspond to different observation characteristics (such as electrical characteristics and thermodynamic characteristics) and can independently estimate the life cycle stage of the equipment; and the state estimation results of the two sets of models are fused through the DS evidence theory to eliminate the uncertainty of a single model, improve the accuracy and robustness of the life cycle stage determination, and thus obtain a more reliable comprehensive fault prediction result.

[0107] If the IGBT driver is in the initial degradation stage, the long short-term memory network is selected as the model framework to construct the initial fault prediction model.

[0108] It should be noted that if the IGBT driver is in the initial degradation stage, the long short-term memory network is selected as the model framework, and the initial fault prediction model is constructed including:

[0109] Design the long-short-term memory network structure, determine the input layer dimension (corresponding to the number of features), the number of hidden layer neurons (usually 64-256), and the output layer (failure probability or degree of degradation), and add a dropout layer to prevent overfitting; use the Adam optimizer and mean square error loss function for model training, and control the number of training rounds using the early stopping method; after training, use the validation set to evaluate the model's ability to capture initial degradation features, focusing on monitoring sensitivity to small parameter drifts (such as a 0.5% change in the on-state voltage drop).

[0110] If the IGBT driver is in the failure acceleration stage, the degradation rate time point analysis module is introduced into the initial fault prediction model to obtain an accelerated fault prediction model.

[0111] It should be noted that the degradation rate timing analysis module is an analytical method used to identify critical time points when the degradation rate of an IGBT driver undergoes sudden changes during operation. Its core approach is to detect and determine the accelerating trend of the slope of performance parameter changes through sliding window and curvature analysis. When applied to the initial long-short-term memory network (LSTM) fault prediction model, this module marks or encodes the degradation rate mutation points in the input sequence and uses them as auxiliary features to input into the LSTM network or to dynamically adjust the LSTM's time window weights, allowing the model to focus more on changing behavior during the degradation acceleration phase. The introduction of the degradation rate timing analysis module significantly improves the model's sensitivity to sudden changes in degradation trends and enhances the prediction's ability to capture failure inflection points, thereby enhancing the model's accuracy and foresight during the failure acceleration phase.

[0112] If the IGBT driver is in the critical failure stage, a survival life analysis module is introduced into the accelerated failure prediction model to obtain a critical failure prediction model.

[0113] It should be noted that if the IGBT driver is in the critical failure stage, the survival life analysis module is introduced into the accelerated failure prediction model, and the critical failure prediction model obtained includes:

[0114] It should be noted that the survival life analysis module is a component based on reliability theory and survival analysis methods, used to quantify the remaining service life of equipment under a given degradation state. This module establishes a statistical relationship between degradation parameters and remaining service life by fitting historical failure data, and calculates the failure probability distribution corresponding to the current degradation trajectory in real time. In the accelerated fault prediction model, the survival life analysis module receives the degradation trend prediction results output by LSTM, maps them into the probability density function of remaining service life, and generates the life distribution confidence interval through Monte Carlo simulation, ultimately obtaining a critical failure prediction model. The effect is that the model not only predicts the possibility of failure, but also outputs a statistical estimate of the specific remaining service life, significantly improving the engineering practicality of the prediction results. Among them, the survival life analysis module includes:

[0115] Step 1: Determine the survival time in the model and define the event state;

[0116] Step 2: Select covariates related to survival time, including equipment degradation characteristics, working environment factors, usage load, etc.

[0117] Step 3: Construct a Cox proportional hazard model based on covariates related to survival time. The expression is:

[0118] h(t∣X)=h0(t)·exp(β1X1+β2X2+...+β p X p );

[0119] Where h(t|X) represents the hazard function given the covariate X; h0(t) represents the baseline hazard function (the risk not affected by the covariate); β1,β2,...,β p All represent the regression coefficients of covariates; X1, X 2, ...,X p Both represent covariate vectors.

[0120] Step 4: Estimate the regression coefficient of the covariate by the maximum likelihood estimation method. The partial likelihood function is realized by calculating all possible sequences of events.

[0121] S3. Identify vulnerable components in the IGBT driver based on the fault frequency, perform fault source analysis on the vulnerable components, and optimize the multi-stage fault prediction model based on the source analysis results.

[0122] It should be noted that identifying vulnerable components in IGBT drivers based on fault frequency, accurately locating key components that affect equipment life by analyzing the fault frequency of different components, and conducting fault tracing analysis on these vulnerable components can help reveal the root cause of the failure mode, thereby identifying the dominant role of specific working conditions or environmental factors in component degradation. The parameters and structure of the prediction model can then be dynamically adjusted according to the failure modes and vulnerable component characteristics at different stages, thereby improving the model's adaptability to different degradation stages and the prediction accuracy.

[0123] According to another embodiment of the present invention, Figure 2 As shown, a safety system for an IGBT driver is also provided, the system comprising:

[0124] A fault identification module is used to extract degradation features from the operating data of the IGBT driver, construct a dynamic causal reasoning matrix based on the degradation features, and use the dynamic causal reasoning matrix to reversely infer the occasional failure mode of the IGBT driver;

[0125] A fault prediction module is used to establish a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver, and use the multi-stage fault prediction model to predict the failure frequency of the IGBT driver in each life cycle stage;

[0126] The fault analysis module is used to identify vulnerable components in the IGBT driver based on the frequency of fault occurrence, perform fault source analysis on the vulnerable components, and optimize the multi-stage fault prediction model based on the source analysis results.

[0127] In summary, with the help of the above technical solution of the present invention, the present invention extracts degradation characteristics from the operating data of the IGBT driver, and constructs a dynamic causal reasoning matrix based on the degradation characteristics, and then uses the matrix to reversely infer occasional failure modes, which can realize the extraction of dominant degradation trajectories that are highly correlated with external factors from the degradation behavior of each component of the driver under various environmental disturbances, and construct a causal structure model with dynamic evolution capability and causal traceability capability. Combining sparse matrix decomposition and path optimization, it effectively compresses the search space of causal relationships, improves the accuracy and expression ability of causal paths, and efficiently identifies and classifies fault modes in dynamic causal structures, thereby significantly enhancing the ability to accurately identify occasional failures of the IGBT driver, improving the reliability, explainability and generalization ability of fault prediction, and providing data support and causal basis for efficient health management and advance maintenance decisions; the present invention establishes a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver, and predicts the frequency of failures in each life cycle stage, which can realize the degradation of the driver from initial stage to The detailed modeling and dynamic tracking of the entire failure critical process, the division of life cycle stages by risk weights, the combination of hidden Markov model and DS evidence theory to improve the accuracy of stage identification, and then introduce differentiated prediction strategies for different stages, such as long short-term memory network to capture initial characteristics, degradation rate analysis to model the acceleration stage, and survival life assessment to characterize the critical stage, effectively improve the adaptability and prediction accuracy of the model to the fault development process, enhance the pertinence and foresight of the prediction model, and thus provide reliable support for the health management and preventive maintenance of key equipment; the present invention effectively reduces the probability of electrical fires by identifying abnormal current, voltage jumps and overheating trends in multiple life cycle stages of equipment such as initial degradation, failure acceleration and failure criticality, and avoids fire accidents caused by problems such as thermal runaway and short circuit of devices. At the same time, the accurate prediction of the life cycle status also helps to judge the hidden dangers that lead to electrical control failure in advance, and timely activate the power-off protection device before detecting that the equipment enters the critical stage to prevent the risk of electric shock caused by insulation damage, abnormal potential and other problems.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting IGBT driver failure, characterized in that: The method includes: S1. Extract degradation features from the operating data of the IGBT driver, build a dynamic causal reasoning matrix based on the degradation features, and use the dynamic causal reasoning matrix to reversely reason about the occasional failure mode of the IGBT driver; S2. Establish a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver, and use the multi-stage fault prediction model to predict the failure frequency of the IGBT driver in each life cycle stage; S3. Identify vulnerable components in the IGBT driver based on the fault frequency, perform fault source analysis on the vulnerable components, and optimize the multi-stage fault prediction model based on the source analysis results.

2. The IGBT driver fault prediction method according to claim 1, characterized in that: The method of extracting degradation features from the operating data of the IGBT driver, constructing a dynamic causal reasoning matrix based on the degradation features, and reversely reasoning the occasional failure mode of the IGBT driver using the dynamic causal reasoning matrix includes: S11. Extract the degradation characteristics of each component from the operating data of the IGBT driver and use the causal discovery algorithm to explore the causal relationship between the degradation characteristics and external environmental factors; S12. Construct an accelerated degradation model based on external environmental factors, and simulate the dominant degradation trajectory of the IGBT driver under different external environmental interferences through the accelerated degradation model; S13. Introduce the dominant degradation trajectory into the dynamic causal reasoning matrix, and reversely deduce the occasional failure mode of the IGBT driver through the dynamic causal reasoning matrix.

3. The IGBT driver fault prediction method according to claim 2, characterized in that: The accelerated degradation model based on external environmental factors is constructed, and the leading degradation trajectory of the IGBT driver under different external environmental interferences is simulated by the accelerated degradation model, including: S121. Constructing a cumulative degradation function of the IGBT driver based on the causal relationship between the degradation characteristics and the external environmental factors, and finding an extreme value of the cumulative degradation function to obtain parameters to be estimated; S122. Constructing an accelerated degradation model based on environmental factors based on the parameters to be estimated, and inputting different external environmental interference combinations into the accelerated degradation model; S123. Based on the output results of the accelerated degradation model, a Monte Carlo technique is used to simulate and generate a set of degradation trajectories of the performance parameters of each component of the IGBT driver evolving over time, and a dominant degradation trajectory is iteratively searched from the set of degradation trajectories.

4. The IGBT driver fault prediction method according to claim 3, characterized in that: The method of introducing the dominant degradation trajectory into the dynamic causal reasoning matrix and reversely deducing the occasional failure mode of the IGBT driver through the dynamic causal reasoning matrix includes: S131. Perform an autoregressive analysis on the dominant degradation trajectory of each component, and based on the autoregressive analysis results, use the degradation characteristics corresponding to the dominant degradation trajectory as child variables and the external environmental factors as parent variables; S132. Construct a dynamic causal inference matrix based on the strength of the causal relationship between the independent variable and the parent variable, and randomly select a causal path traced back from the independent variable using a sparse matrix decomposition algorithm, delete duplicate variables in the same causal path, and obtain an optimized dynamic causal inference matrix. S133. Based on the optimized dynamic causal reasoning matrix, the failure mode of the IGBT driver is reversed, and the failure mode category is identified using the pre-built graph convolutional neural network to screen out the occasional failure mode of the IGBT driver.

5. The IGBT driver fault prediction method according to claim 4, characterized in that: The causal path randomly selected and traced back from the independent variable in combination with the sparse matrix decomposition algorithm includes: Obtain all causal paths traced from the independent variables in the dynamic causal inference matrix, and introduce elements of sparse constraint singular vectors to ensure the sparseness of the dynamic causal inference matrix; The dynamic causal inference matrix after sparseness is approximated into several low-rank matrices, and the rank of the low-rank matrix is reduced by selecting the principal components through the principal hierarchy analysis method to obtain a low-rank sparse causal matrix; An independent variable is randomly selected as the starting point, and a sparse causal path starting from the current independent variable is selected through the strong connection in the sparse causal matrix.

6. The IGBT driver fault prediction method according to claim 1, characterized in that: The multi-stage fault prediction model is established based on the occasional failure mode of the IGBT driver. The multi-stage fault prediction model is used to predict the failure frequency of each life cycle stage of the IGBT driver, including: S21. Assign risk weights to the occasional failure modes and divide the IGBT driver into several life cycle stages based on the risk weights, wherein the life cycle stages of the IGBT driver include an initial degradation stage, an accelerated failure stage, and a critical failure stage; S22. Construct a fault prediction model for each life cycle stage of the IGBT driver, and predict the frequency of failures in each life cycle stage based on the output results of each fault prediction model.

7. The IGBT driver fault prediction method according to claim 6, characterized in that: The construction of the fault prediction model for each life cycle stage of the IGBT driver includes: Based on the assigned accidental failure mode weights, the total failure frequency index of the IGBT driver is calculated. A stage identifier is then built based on the total failure frequency index to determine the current life cycle stage of the IGBT driver. If the IGBT driver is in the initial degradation stage, the long short-term memory network is selected as the model framework to build the initial fault prediction model; If the IGBT driver is in the failure acceleration stage, the degradation rate time point analysis module is introduced into the initial fault prediction model to obtain the accelerated fault prediction model; If the IGBT driver is in the critical failure stage, a survival life analysis module is introduced into the accelerated failure prediction model to obtain a critical failure prediction model.

8. The IGBT driver fault prediction method according to claim 7, characterized in that: The stage identifier is constructed based on the total fault frequency indicator, and the stage identifier is used to determine the current life cycle stage of the IGBT driver. Using occasional failure patterns as training data, a parameter estimation algorithm is used to train two sets of hidden Markov models. The hidden states in each hidden Markov model are mapped to the life cycle stages of the IGBT driver. The life cycle state of the IGBT driver is estimated, and the state estimation results output by two sets of hidden Markov models are fused through DS evidence theory. The current life cycle stage of the IGBT driver and the transition probability from one life cycle stage to another are calculated based on the state estimation results.

9. The IGBT driver fault prediction method according to claim 2, characterized in that: The expression of the accelerated degradation model is: Where A total (t) represents the comprehensive acceleration effect of IGBT driver degradation at time t; A0 represents the initial degradation reference value; T(t) represents the temperature at time t; E a represents temperature sensitivity; RH(t) represents humidity at time t; RH0 represents reference humidity; V(t) represents voltage stress at time t; V0 represents reference voltage; S(t) represents vibration stress at time t; S0 represents reference vibration; k B represents the Boltzmann constant; m represents the humidity acceleration index; p represents the vibration acceleration index; and n represents the voltage acceleration index.

10. An IGBT driver safety system, using the IGBT driver fault prediction method according to any one of claims 1 to 9, characterized in that: The system includes: A fault identification module is used to extract degradation features from the operating data of the IGBT driver, construct a dynamic causal reasoning matrix based on the degradation features, and use the dynamic causal reasoning matrix to reversely infer the occasional failure mode of the IGBT driver; A fault prediction module is used to establish a multi-stage fault prediction model based on the occasional failure mode of the IGBT driver, and use the multi-stage fault prediction model to predict the failure frequency of the IGBT driver in each life cycle stage; The fault analysis module is used to identify vulnerable components in the IGBT driver based on the frequency of fault occurrence, perform fault source analysis on the vulnerable components, and optimize the multi-stage fault prediction model based on the source analysis results.

Images