IGBT Device Remaining Life Prediction Method and System

Abstract

The present application provides a method and system for predicting the remaining life of an IGBT device, including: simulating an experiment on the power cycle aging of the IGBT device, collecting historical data of the conduction voltage drop according to the experiment, and respectively obtaining empirical values of model parameters and estimated values of model parameters based on the historical data of the conduction voltage drop; constructing a first state weight according to the current operating state, and smoothly adjusting the estimated values of the model parameters according to the empirical values of the model parameters and the first state weight to obtain a parameter correction result; establishing a state transition equation according to the parameter correction result, and predicting the remaining life of the IGBT device according to the state transition equation; using the present application can improve the accuracy of predicting the remaining life of the IGBT device.

Description

Technical Field

[0001] This application relates to the technical field of power electronic devices, and particularly to a method and system for predicting the remaining life of IGBT devices. Background Art

[0002] Insulated Gate Bipolar Transistor (IGBT) has become the core component of modern high-power power electronic equipment due to its advantages such as small drive power, large current-carrying capacity, and fast switching speed. However, during actual operation, as the service time of the IGBT device increases, its packaging structure will continuously age under the action of cyclic electrothermal stress, resulting in a gradual decline in the electrothermal performance of the device, which may not meet the design requirements of the power electronic equipment, ultimately affecting the reliability of the entire equipment or system, and even having a huge impact on system safety in severe cases. Therefore, accurately predicting the remaining useful lifetime (RUL) of IGBT devices is of great significance for improving system reliability. Accurately obtaining the remaining life of IGBT devices not only helps to reliably evaluate their current health status, but also helps system operation and maintenance, and timely replace the devices about to fail, thus ensuring system safety.

[0003] Currently, extensive research has been carried out on the life modeling of IGBT devices in existing related technologies, and a variety of IGBT device life models have been proposed, such as the Coffin-Manson model based on the analytical method and the stress-strain model based on the physics of failure. The above life models use the operating condition parameters of IGBT devices as inputs, such as the junction temperature fluctuation ΔTj, switching frequency f, and current magnitude IC during device operation, etc., and output the average service life of IGBT devices. However, with input parameters being operating condition parameters and output being the average service life, it is impossible to accurately predict the remaining life of a specific IGBT device. Summary of the Invention

[0004] The purpose of this application is to address the deficiencies of the above existing related technologies, and provide a method and system for predicting the remaining life of IGBT devices, which can improve the accuracy of predicting the remaining life of IGBT devices.

[0005] In a first aspect, the present invention provides a method for predicting the remaining life of an IGBT device, including:

[0006] Simulate the experiment of power cycle aging of the IGBT device, collect the historical data of the conduction voltage drop according to the experiment, and respectively obtain the empirical value and estimated value of the model parameters according to the historical data of the conduction voltage drop;

[0007] Construct a first state weight according to the current operating state, and smooth-adjust the model parameter estimated value according to the empirical value of the model parameter and the first state weight to obtain a parameter correction result;

[0008] Establish a state transition equation according to the parameter correction result, and predict the remaining life of the IGBT device according to the state transition equation.

[0009] This application uses the historical data of the conduction voltage drop to obtain the empirical value of the model parameter and the estimated value of the model parameter, and combines the empirical value of the model parameter and the first state weight constructed according to the current operating state to adjust the estimated value of the model parameter, so as to reduce the error of directly estimating the model parameter and improve the accuracy of the remaining life prediction based on the adjusted input data; at the same time, according to the obtained parameter correction result, a state transition equation is established to predict the remaining life of the IGBT device, so as to realize the adjustment of the prediction accuracy based on the output data and further improve the accuracy of the remaining life prediction.

[0010] Further, the smoothing adjustment of the model parameter estimated value according to the empirical value of the model parameter and the first state weight to obtain a parameter correction result includes:

[0011] Assign the first state weight to the model parameter estimated value to obtain a first weighted result, and assign a second state weight to the empirical value of the model parameter to obtain a second weighted result, and use the sum of the first weighted result and the second weighted result as the parameter correction result; where the sum of the first state weight and the second state weight is a preset value.

[0012] This application assigns a first state weight to the model parameter estimated value and a second state weight to the empirical value of the model parameter to realize the smoothing adjustment of the model parameter estimated value based on historical data and current data, so as to reduce the error of directly estimating the model parameter and improve the accuracy of the remaining life prediction based on the adjusted input data.

[0013] Further, the parameter correction result includes: a first parameter correction result and a second parameter correction result; the model parameter estimated value includes: a first model parameter estimated value and a second model parameter estimated value; the model parameter empirical value includes: a first model parameter empirical value and a second model parameter empirical value; where the first parameter correction result and the second parameter correction result are respectively expressed as:

[0014]

[0015] where α k and β k are the first parameter correction result and the second parameter correction result respectively; and They are the first model parameter estimate and the second model parameter estimate respectively; both α and β are the first model parameter empirical value and the second model parameter empirical value; is the first state weight, is the second state weight.

[0016] Furthermore, the current operating state includes: the current voltage state of the IGBT device; constructing the first state weight according to the current operating state includes:

[0017] Constructing the first state weight according to the ratio of the current voltage state to the failure threshold of the IGBT device.

[0018] Furthermore, collecting the historical data of the conduction voltage drop according to the experiment, and obtaining the model parameter empirical value and the model parameter estimate according to the historical data of the conduction voltage drop, includes:

[0019] Collecting the historical data of the conduction voltage drop according to the experiment, and obtaining multiple sets of time series data of the conduction voltage drop change according to the historical data of the conduction voltage drop, and obtaining the model parameter empirical value according to the time series data.

[0020] Furthermore, obtaining the model parameter empirical value according to the time series data includes:

[0021] Establishing an IGBT degradation process model with the conduction voltage drop data as the characteristic quantity, sampling in the time series data as the training data set of the IGBT degradation process model, obtaining the empirical distribution of the model parameters according to the training data set and the IGBT degradation process model, and obtaining the model parameter empirical value according to the empirical distribution.

[0022] Furthermore, collecting the historical data of the conduction voltage drop according to the experiment, and obtaining the model parameter estimate according to the historical data of the conduction voltage drop, includes:

[0023] Constraining the model parameters according to the time series data and based on the range value of the empirical distribution, and performing parameter estimation on the model parameters through a particle swarm optimization algorithm to obtain the model parameter estimate.

[0024] Furthermore, establishing a state transition equation according to the parameter correction result, and predicting the remaining life of the IGBT device according to the state transition equation, includes:

[0025] According to the current operating state and the parameter correction result, a particle filter state vector and a state transition equation are jointly constructed, and the particle filter state vector is updated multiple times according to the state transition equation until the posterior distribution of the state vector is obtained under the condition of the conduction voltage drop historical data, and then the remaining life of the IGBT device is predicted using the state transition equation at this time.

[0026] Further, predicting the remaining life of the IGBT device using the state transition equation at this time includes:

[0027] Obtain the current particle filter state vector, and according to the state transition equation at this time, predict the future state distribution. If the remaining life is defined as the time required for the particles of the particle filter state vector to first reach the failure threshold from the current moment, then according to the future state distribution, obtain the remaining life of the IGBT device.

[0028] In a second aspect, the present application provides a system for predicting the remaining life of an IGBT device, including: a historical data acquisition module, a correction module, and a prediction module; wherein,

[0029] The historical data acquisition module is used to simulate the experiment of power cycle aging of the IGBT device, collect the conduction voltage drop historical data according to the experiment, and respectively obtain the empirical values of the model parameters and the estimated values of the model parameters according to the conduction voltage drop historical data;

[0030] The correction module is used to construct a first state weight according to the current operating state, and perform a smoothing adjustment on the estimated value of the model parameter according to the empirical value of the model parameter and the first state weight to obtain a parameter correction result;

[0031] The prediction module is used to establish a state transition equation according to the parameter correction result, and predict the remaining life of the IGBT device according to the state transition equation. Description of the Drawings

[0032] Figure 1 is a schematic flowchart of a method for predicting the remaining life of an IGBT device provided by an embodiment of the present application;

[0033] Figure 2 is a schematic flowchart of another method for predicting the remaining life of an IGBT device provided by an embodiment of the present application;

[0034] Figure 3(a) is a schematic diagram of a method for predicting the remaining life provided by an embodiment of the present application;

[0035] Figure 3(b) is t provided by an embodiment of the present application k = a schematic diagram of the remaining life prediction result at 140k;

[0036] Figure 4 It is a schematic diagram of the remaining life prediction results at different prediction times provided by an embodiment of the present application;

[0037] Figure 5 It is a schematic diagram of the remaining life prediction deviation at different prediction times provided by an embodiment of the present application;

[0038] Figure 6 It is a schematic structural diagram of an IGBT device remaining life prediction system provided by an embodiment of the present application. Detailed implementation manners

[0039] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0040] It is worth noting that the existing related technologies use operating condition parameters as input parameters and the average service life as output, fail to correct the model parameters of the IGBT device, and the predicted life is the average life, making it impossible to accurately predict the remaining life of a specific IGBT device. Based on this, the present application provides an IGBT device remaining life prediction method and system, which can improve the accuracy of predicting the remaining life of the IGBT device.

[0041] To more clearly illustrate the solution of the present application, the following embodiments will be described in detail.

[0042] Embodiment 1

[0043] Refer to Figure 1 , which is a schematic flow diagram of an IGBT device remaining life prediction method provided by an embodiment of the present application, including: steps S11 to S13, specifically:

[0044] Step S11, simulate the experiment of power cycle aging of the IGBT device, collect the historical data of the conduction voltage drop according to the experiment, and respectively obtain the empirical value of the model parameter and the estimated value of the model parameter according to the historical data of the conduction voltage drop.

[0045] In some embodiments, collecting the historical data of the conduction voltage drop according to the experiment and obtaining the empirical value of the model parameter and the estimated value of the model parameter according to the historical data of the conduction voltage drop include: collecting the historical data of the conduction voltage drop according to the experiment, obtaining multiple sets of time series data of the change of the conduction voltage drop according to the historical data of the conduction voltage drop, and obtaining the empirical value of the model parameter according to the time series data.

[0046] In some embodiments, according to the time series data, empirical values of model parameters are obtained, including: establishing an IGBT degradation process model with the on-state voltage drop data as the feature quantity, sampling in the time series data as the training data set of the IGBT degradation process model, obtaining the empirical distribution of the model parameters according to the training data set and the IGBT degradation process model, and obtaining the empirical values of the model parameters according to the empirical distribution.

[0047] In some embodiments, power cycling accelerated aging experiments are carried out on N + 1 groups of IGBT devices, and the on-state voltage drop VCE is synchronously extracted each time the cycle is completed. Among them, the time series of the on-state voltage drop of N groups of IGBT devices is selected to establish the model training data set V = {V1, V2, …, V N}, where each group of sequences represents the change data of the on-state voltage drop of an IGBT device.

[0048] In some embodiments, the IGBT degradation process model can be expressed as:

[0049] V CE = 1 + αln(βt + 1);

[0050] where α and β are model parameters, t is any moment, and in the experiment, it is the aging cycle number of the IGBT device.

[0051] In some embodiments, the model parameters are obtained by fitting according to the on-state voltage drop time series data set V. At this time, the obtained parameters are the empirical distributions of the model parameters α and β, which are used for the estimation of the model parameters in the subsequent particle filtering process.

[0052] In some embodiments, the IGBT degradation process model can define the state transition equation and the observation equation.

[0053] In some embodiments, defining the state transition equation and the observation equation includes: establishing the state-space equation of the particle filtering model, and then defining the state transition equation and the state transition equation required by the particle filtering model.

[0054] In some implementations, the general form of the state-space equation can be expressed as:

[0055]

[0056] where x k represents the IGBT device state value at time t k , y k represents the state measurement value, and u k , v k represent the process noise and the observation noise respectively.

[0057] In some embodiments, according to the established IGBT degradation process model, the state transition equation required for the particle filter model can be obtained. First, taking the derivative of it, we get:

[0058]

[0059] In some embodiments, the state transition equation required for the particle filter model can be written in the following discrete form:

[0060]

[0061] where Δt is the time step after discretization.

[0062] In some embodiments, the measurement equation required for the particle filter model can be written as:

[0063] V CE_m,k =V CE,k +v k ;

[0064] where V CE_m,k represents the measured value of the on-state voltage drop of the IGBT device, and v k ~N(0,Q), where Q is the variance of the observation noise.

[0065] In some embodiments, based on the historical data of the on-state voltage drop collected through experiments, and according to the historical data of the on-state voltage drop, the estimated values of the model parameters are obtained, including: based on the time series data, and constraining the model parameters according to the range values of the empirical distribution, and estimating the model parameters through the particle swarm optimization algorithm to obtain the estimated values of the model parameters.

[0066] In some embodiments, according to the historical data set {V CE_m} 1:k obtained by measuring the on-state voltage drop of the IGBT device, the particle swarm optimization algorithm is used to perform parameter estimation of the model parameters under constrained conditions, and the constraint condition is the empirical range of the parameters of the empirical distribution of the model parameters, and the estimated values of the model parameters are obtained. The estimated values of the model parameters include: the first estimated value of the model parameter and the second estimated value of the model parameter, which are denoted as and

[0067] Step S12: According to the current operating state, construct the first state weight, and smooth and adjust the estimated values of the model parameters according to the empirical values of the model parameters and the first state weight to obtain the parameter correction result.

[0068] In some embodiments, the current operating state includes: the current voltage state of the IGBT device; according to the current operating state, constructing the first state weight includes: constructing the first state weight according to the ratio of the current voltage state to the failure threshold of the IGBT device.

[0069] In some embodiments, according to the current voltage state V of the IGBT device CE,k , the first state weight can be calculated as:

[0070]

[0071] where w _th is the set failure threshold of the IGBT device; when the device operates normally, w _th > V CE,k , and after the device fails, w _th ≤V CE,k .

[0072] In some embodiments, according to the empirical values of the model parameters and the first state weight, the estimated value of the model parameters is smoothed and adjusted to obtain the parameter correction result, including: assigning the first state weight to the estimated value of the model parameters to obtain the first weighted result, and assigning the second state weight to the empirical value of the model parameters to obtain the second weighted result, and taking the sum of the first weighted result and the second weighted result as the parameter correction result; wherein, the sum of the first state weight and the second state weight is a preset value.

[0073] In some embodiments, the parameter correction result includes: a first parameter correction result and a second parameter correction result; the estimated value of the model parameters includes: a first estimated value of the model parameters and a second estimated value of the model parameters; the empirical value of the model parameters includes: a first empirical value of the model parameters and a second empirical value of the model parameters.

[0074] It should be noted that, in some embodiments, the first state weight is a state weight constructed according to the ratio of the current motion state and the failure threshold of the IGBT device, and is used to weight the first estimated value of the model parameters, and is denoted by , the second state weight is a state weight for weighting the second estimated value of the model parameters , and the sum of the first state weight and the second state weight is a preset value, which can be denoted by , is a preset value.

[0075] In some embodiments can be 1 or other positive numbers.

[0076] In this application, by assigning the first state weight to the estimated value of the model parameters and the second state weight to the empirical value of the model parameters, the estimated value of the model parameters is smoothed and adjusted based on historical data and current data, so as to reduce the error of directly estimating the model parameters and improve the accuracy of the remaining life prediction based on the adjusted input data.

[0077] In some embodiments, when estimating model parameters, obtaining the model parameter estimation values includes: the first model parameter estimation value and the second model parameter estimation value When simulating the power cycle aging experiment of an IGBT device, by collecting the historical data of the conduction voltage drop, the empirical values of the model parameters are obtained. Collecting the historical data of the conduction voltage drop includes: the first empirical value α of the model parameter and the second empirical value β of the model parameter. When performing a smoothing adjustment on the model parameter estimation values, the parameter correction result is obtained. Specifically, the first model parameter estimation value and the second model parameter estimation value are respectively smoothed. Therefore, the parameter correction result includes: the first parameter correction result α k and the second parameter correction result β k . The first weighted result refers to the weighted result obtained by assigning the first state weight to the first model parameter estimation value and the second model parameter estimation value . That is, after assigning the first state weight to the first model parameter estimation value , the weighted result of the first model parameter estimation value is obtained; after assigning the first state weight to the second model parameter estimation value , the weighted result of the second model parameter estimation value is also obtained; similarly, the second weighted result refers to the weighted result obtained by assigning the second state weight to the first empirical value α of the model parameter and the second empirical value β of the model parameter respectively. That is, after assigning the second state weight to the first empirical value α of the model parameter, the weighted result of the first empirical value α of the model parameter is obtained; after assigning the second state weight to the second empirical value β of the model parameter, the weighted result of the second empirical value β of the model parameter is obtained. After assigning the second state weight to the first empirical value α of the model parameter, the weighted result of the first empirical value α of the model parameter can be obtained; after assigning the second state weight

[0078] In some embodiments, the first parameter correction result and the second parameter correction result are respectively expressed as:

[0079]

[0080] where α k and β k are the first parameter correction result and the second parameter correction result respectively; and They are the first model parameter estimation value and the second model parameter estimation value respectively; both α and β are the first model parameter empirical value and the second model parameter empirical value; is the first state weight, is the second state weight.

[0081] Step S13: Establish a state transition equation according to the parameter correction result, and predict the remaining life of the IGBT device according to the state transition equation.

[0082] In some embodiments, to establish a state transition equation based on the parameter correction result, the particle filter algorithm can be used to estimate the state of the IGBT device and predict the remaining life.

[0083] In some embodiments, according to the parameter correction result, establishing a state transition equation, and predicting the remaining life of the IGBT device according to the state transition equation includes: jointly constructing a particle filter state vector and a state transition equation according to the current operating state and the parameter correction result, and updating the particle filter state vector multiple times according to the state transition equation until the posterior distribution of the state vector under the condition of the on-state voltage drop historical data is obtained, and then predicting the remaining life of the IGBT device with the state transition equation at this time.

[0084] In some embodiments, the voltage V of the IGBT device CE , the model parameters α k and β k are jointly used to construct a particle filter state vector x, and the particle filter state vector x can be expressed as:

[0085] x = [V CE , α, β].

[0086] In some embodiments, updating the particle filter state vector multiple times according to the state transition equation until the posterior distribution of the state vector under the condition of the on-state voltage drop historical data is obtained includes: initializing the particle state of the particle filter state vector to obtain a particle state result, and initializing the particle weight of the particle filter state vector to obtain a particle weight result; updating the particle state result according to the state transition equation and updating the particle weight result according to the observation equation. If the number of effective particles of the updated weight result is less than the threshold, resample the particles until the posterior distribution of the state vector under the condition of the on-state voltage drop historical data is obtained.

[0087] In some embodiments, according to the initial state distribution, initial particles can be randomly generated and the particle filter state vector x can be initialized, which can be expressed as:

[0088]

[0089] where \(i = 1, 2, \ldots, N_s\), \(N_s\) represents the total number of particles, and the particle weight \(\omega\) of each initial particle is set to be the same.

[0090] In some embodiments, the particle weight can be expressed as:

[0091]

[0092] In some embodiments, updating the particle state of the particle filter state vector according to the state transition equation can be expressed as:

[0093]

[0094] In some embodiments, updating the particle weight according to the observation equation can be expressed as:

[0095]

[0096] In some embodiments, normalizing the particle weight can be expressed as:

[0097]

[0098] In some embodiments, calculating the number of effective particles \(N\) eff , can be expressed as:

[0099]

[0100] In some embodiments, to address the problem of particle degeneracy, when the number of effective particles , resample the particles, and reset the particle weights after resampling to \(1 / N_s\).

[0101] Thus, the posterior distribution of the state vector can be obtained under the condition of the known historical data \(\{V\) CE_m \}: 1:k where \(\delta(\cdot)\) represents the Dirac function.

[0102]

[0103] where \(\delta(\cdot)\) represents the Dirac function.

[0104] In some embodiments, predicting the remaining useful life of the IGBT device using the state transition equation at this time includes: obtaining the current particle filter state vector, and predicting the future state distribution according to the state transition equation at this time. If the remaining useful life is defined as the time required for the particle of the particle filter state vector to first reach the failure threshold from the current moment, then the remaining useful life of the IGBT device can be obtained according to the future state distribution.

[0105] In some embodiments, based on the obtained current state, the future state can be predicted according to the state transition equation, and thus the state distribution at the (k + l)-th moment can be known as:

[0106]

[0107] wherein,

[0108] In some embodiments, the remaining life L is defined k as the time required for the particle to first reach the failure threshold w _th from the current moment, that is:

[0109] L k = inf{l | V CE,k+l ≥ w _th | V CE_m,1:k}

[0110] wherein, l represents the time required for the particle to reach failure from the initial time.

[0111] In some embodiments, the future remaining life distribution can be expressed as:

[0112]

[0113] Compared with the traditional analytical method for IGBT life prediction, the IGBT life prediction solution of the present application adds the current state and historical data of the device during the life prediction process, and can predict the remaining life of a specific device; moreover, during the process of using the particle filter algorithm to predict the remaining life of the IGBT device, the state-determined weight is introduced in the parameter estimation stage to correct the estimated parameters, so that the corrected parameter results mainly reflect the empirical parameters in the early stage and can better reflect the specific failure state parameters of the device in the late failure stage.

[0114] Embodiment 2

[0115] The embodiment of the present application gives a specific application example of the IGBT device life distribution probability prediction. Refer to Figure 2 , which is a schematic flowchart of another method for predicting the remaining life of the IGBT device provided by the embodiment of the present application. Specifically in the specific application example:

[0116] (1) Conduct 6 groups of IGBT device power cycle accelerated aging experiments. Among them, with N = 5, the measured data of the on-state voltage drop is divided into a training data set and a validation data set. The training data set contains 5 groups of on-state voltage drop change data, and a degradation process model of the on-state voltage drop V CE is established according to the training data set:

[0117] V CE = 1 + αln(βt + 1);

[0118] Among them, \(t\) is any time, and \(\alpha\) and \(\beta\) are model parameters, which can be obtained by fitting the conduction voltage drop sequence.

[0119] After fitting, the model parameters of 5 groups of IGBT devices are obtained respectively, as shown in Table 1.

[0120] Table 1 Model parameter values

[0121]

[0122] By performing normal statistical fitting on the model parameters \(\alpha\) and \(\beta\) respectively, the prior distribution law of the model parameters can be obtained:

[0123]

[0124] Among them, \(N(\mu,\sigma)\) represents a normal distribution with a mean of \(\mu\) and a variance of \(\sigma\).

[0125] (2) Establish a state equation according to the IGBT device degradation process model:

[0126]

[0127] Among them, \(\Delta t\) represents the discrete time step, \(V\) CE,k represents the conduction voltage drop state value of the IGBT device at the \(t\) k th time, and \(\alpha\) k and \(\beta\) k represent the estimated values of the model parameters at the \(t\) k th time.

[0128] Establish a measurement equation:

[0129] \(V\) CE_m,k = \(V\) CE,k + \(v\) k ;

[0130] Among them, \(V\) CE_m,k represents the measured conduction voltage value at time \(t\) k , \(v\) k represents the measurement noise, which is affected by the measurement device. Here, it is considered to be a white noise \(v\) k ~ \(N(0,0.001)\).

[0131] (3) Use the particle filter algorithm to predict the remaining life of the data validation set at the selected monitoring time \(t\) k . Consider the data of the validation set in the first \(t\) k time as known historical data, and the data after time \(t\) k as future unknown data. Then, the remaining life \(L\) of the IGBT device at the current time \(t\) k can be defined.k is:

[0132] L k = inf{l k |V CE,k ≥ ω|V CE_m,1:k};

[0133] where, {V CE_m} 1:k represents the k - group voltage historical measurement data before time t k , w _th represents the set IGBT device voltage failure threshold, here taking 1.05 times the initial voltage value according to experience, and inf{·} is used to obtain the minimum lower limit of the set.

[0134] (4) Use the particle swarm optimization algorithm with constraints to estimate the model parameters. Among them, the search range of the model parameters is constrained by parameter empirical values to avoid over - fitting problems, that is:

[0135]

[0136] The fitness function fopt of the algorithm is described using the sum of squared errors:

[0137]

[0138] where, represents the voltage value calculated according to the current parameter estimation result, and represent the current obtained model parameter estimation values. The particle swarm optimization algorithm solves the optimal parameter estimation result by minimizing the fitness function within the constraint range.

[0139] The particle swarm optimization algorithm uses historical measurement data to estimate the model parameters. When the historical data is small, there may be problems with poor estimation accuracy. Therefore, the first - state weight is introduced to correct the estimation result of the model parameters:

[0140]

[0141] where, represents the state - decision weight at the current time t k , and its value is defined as the ratio of the conduction voltage of the current IGBT device to the failure threshold. When t k is small, the voltage of the IGBT device differs greatly from the failure threshold. Therefore is less than 1, so that the corrected model parameters are closer to the empirical values, reducing the parameter over - fitting problem when the data volume is small; when t k is large, the actual voltage of the IGBT device is closer to the failure threshold. Therefore It is closer to 1, so that the corrected model parameters are closer to the estimated values, further increasing the accuracy of the model parameter estimation results.

[0142] If t k = 140k is used to predict the remaining life of the device at the current moment, then the parameter estimation values of the particle swarm algorithm can be obtained as The state decision weight at this time is The corrected parameter is α k ~N(-0.0093, 0.0014 2 ), β k ~N(-0.0061, 0.00046 2 ), and thus its probability density function can be written as:

[0143]

[0144] Among them, and are the posterior distribution probabilities of the model parameters α k and β k respectively, μ k and represent their expectations and variances respectively, and at t k = 140, there is

[0145] (4) Use the particle filter algorithm to predict the remaining life of the IGBT device. First, generate an initial particle set according to the above parameter estimation results:

[0146]

[0147] Among them, Ns is the number of particles generated by simulation. Here, Ns = 5000, and are sampled by importance from their corrected empirical distributions and respectively, and an initial weight is assigned to each particle. When the voltage measurement value V k of the IGBT device at time t CE_m,k is obtained, the state and weight of all particles are updated.

[0148] The particle state update equation is:

[0149]

[0150] The particle weight update equation is:

[0151]

[0152] Normalize the updated particle weights:

[0153]

[0154] If the number of effective particles after update needs to be resampled to enhance the accuracy of the prediction results, the particle weights after resampling are redefined as 1 / Ns.

[0155] Among them, the number of effective particles N eff is:

[0156]

[0157] Thus, the posterior distribution of the on-state voltage drop of the IGBT device can be obtained under the condition of the known historical data {V CE_m}: 1:k where δ(·) represents the Dirac function.

[0158]

[0159] Among them, δ(·) represents the Dirac function.

[0160] Furthermore, based on the obtained current state, the future state can be predicted according to the state transition equation, and thus the state distribution at the (k + l)-th moment during failure can be known as:

[0161]

[0162] where

[0163] Furthermore, the future remaining life distribution can be calculated according to the time when the particles reach the failure threshold:

[0164]

[0165] Thus, the remaining life at the start of prediction at t k = 140k is about 27.5k. The remaining life prediction result is shown in Fig. 3(a), which is a schematic diagram of the remaining life prediction method provided by the embodiment of the present application. Fig. 3(b) is a schematic diagram of the remaining life prediction result at t k = 140k provided by the embodiment of the present application. According to the validation set, the true remaining life of the IGBT device is about 26k, and the prediction error is 5.6%, indicating the accuracy of the proposed method. At the same time, see Figure 4 and Figure 5, which are respectively the schematic diagrams of the remaining life prediction results at different prediction times and the remaining life prediction deviation at different prediction times provided by the embodiments of the present application. It can also be seen therefrom that the proposed results have high accuracy.

[0166] Embodiment 3

[0167] Refer to Figure 6 , which is the structural schematic diagram of the IGBT device remaining life prediction system provided by the embodiments of the present application, including: a historical data acquisition module 31, a correction module 32, and a prediction module 33.

[0168] It should be noted that when the IGBT device remaining life prediction system simulates the power cycle aging experiment of the IGBT device in the historical data acquisition module 31, it can be simulated by an online software system or an offline experiment.

[0169] In some embodiments, the historical data acquisition module 31 is used to output the empirical value of the model parameter and the estimated value of the model parameter to the correction module 32. After receiving the empirical value of the model parameter and the estimated value of the model parameter, the correction module 32 outputs the parameter correction result and transmits the parameter correction result to the prediction module 33. After receiving the parameter correction result, the prediction module 33 performs the remaining life prediction on the IGBT device and outputs the life prediction result.

[0170] The historical data acquisition module 31 is used to simulate the power cycle aging experiment of the IGBT device, collect the historical data of the conduction voltage drop according to the experiment, and respectively obtain the empirical value of the model parameter and the estimated value of the model parameter according to the historical data of the conduction voltage drop.

[0171] In some embodiments, collecting the historical data of the conduction voltage drop according to the experiment and obtaining the empirical value of the model parameter and the estimated value of the model parameter according to the historical data of the conduction voltage drop include: collecting the historical data of the conduction voltage drop according to the experiment, and obtaining multiple groups of time series data of the conduction voltage drop change according to the historical data of the conduction voltage drop, and obtaining the empirical value of the model parameter according to the time series data.

[0172] In some embodiments, obtaining the empirical value of the model parameter according to the time series data includes: establishing an IGBT degradation process model with the conduction voltage drop data as the characteristic quantity, sampling in the time series data as the training data set of the IGBT degradation process model, obtaining the empirical distribution of the model parameter according to the training data set and the IGBT degradation process model, and obtaining the empirical value of the model parameter according to the empirical distribution.

[0173] In some embodiments, historical data of the conduction voltage drop is collected according to experiments, and based on the historical data of the conduction voltage drop, an estimated value of the model parameter is obtained, including: based on time series data, constraining the model parameter according to the range value of the empirical distribution, and performing parameter estimation on the model parameter through a particle swarm optimization algorithm to obtain the estimated value of the model parameter.

[0174] A correction module 32, configured to construct a first state weight according to the current operating state, and perform a smoothing adjustment on the estimated value of the model parameter according to the empirical value of the model parameter and the first state weight to obtain a parameter correction result.

[0175] In some embodiments, the current operating state includes: the current voltage state of the IGBT device; constructing a first state weight according to the current operating state, including: constructing a first state weight according to the ratio of the current voltage state to the failure threshold of the IGBT device.

[0176] In some embodiments, performing a smoothing adjustment on the estimated value of the model parameter according to the empirical value of the model parameter and the first state weight to obtain a parameter correction result, including: assigning the first state weight to the estimated value of the model parameter to obtain a first weighted result, and assigning a second state weight to the empirical value of the model parameter to obtain a second weighted result, and using the sum of the first weighted result and the second weighted result as the parameter correction result; wherein, the sum of the first state weight and the second state weight is a preset value.

[0177] In some embodiments, the parameter correction result includes: a first parameter correction result and a second parameter correction result; the estimated value of the model parameter includes: a first estimated value of the model parameter and a second estimated value of the model parameter; the empirical value of the model parameter includes: a first empirical value of the model parameter and a second empirical value of the model parameter.

[0178] In some embodiments, the first parameter correction result and the second parameter correction result are respectively expressed as:

[0179]

[0180] wherein, α k and β k are respectively the first parameter correction result and the second parameter correction result; and are respectively the first estimated value of the model parameter and the second estimated value of the model parameter; both α and β are the first empirical value of the model parameter and the second empirical value of the model parameter; is the first state weight, is the second state weight.

[0181] A prediction module 33, configured to establish a state transition equation according to the parameter correction result, and perform a remaining life prediction on the IGBT device according to the state transition equation.

[0182] In some embodiments, according to the parameter correction result, a state transition equation is established, and the remaining life of the IGBT device is predicted according to the state transition equation, including: jointly constructing a particle filter state vector and a state transition equation according to the current operating state and the parameter correction result, and updating the particle filter state vector multiple times according to the state transition equation until the posterior distribution of the state vector under the condition of the conduction voltage drop historical data is obtained, and then predicting the remaining life of the IGBT device with the state transition equation at this time.

[0183] In some embodiments, the particle filter state vector is updated multiple times according to the state transition equation until the posterior distribution of the state vector under the condition of the conduction voltage drop historical data is obtained, including: initializing the particle state of the particle filter state vector to obtain a particle state result, and initializing the particle weight of the particle filter state vector to obtain an initialization weight result; updating the particle state result according to the state transition equation, and updating the initialization weight result according to the observation equation. If the number of effective particles of the updated weight result is less than the threshold, resampling is performed on the particles until the posterior distribution of the state vector under the condition of the conduction voltage drop historical data is obtained.

[0184] In some embodiments, predicting the remaining life of the IGBT device with the state transition equation at this time includes: obtaining the current particle filter state vector, and predicting the future state distribution according to the state transition equation at this time. If the remaining life is defined as the time required for the particles of the particle filter state vector to reach the failure threshold for the first time from the current moment, then the remaining life of the IGBT device is obtained according to the future state distribution.

[0185] This application collects the conduction voltage drop historical data to obtain the empirical values and estimated values of the model parameters, and adjusts the estimated values of the model parameters by combining the empirical values of the model parameters and the first state weight constructed according to the current operating state, so as to reduce the error of directly estimating the model parameters and improve the accuracy of the remaining life prediction based on the adjusted input data; at the same time, according to the obtained parameter correction result, a state transition equation is established to predict the remaining life of the IGBT device, so as to realize the adjustment of the prediction accuracy based on the output data and further improve the accuracy of the remaining life prediction.

[0186] Those skilled in the art should understand that the embodiments of the present application may also provide a computer program product. Therefore, the present application may be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application may be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0188] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0190] The above is only the preferred embodiment of the present application. It should be noted that for those of ordinary skill in the art of this technology, without departing from the technical principle of the present application, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present application.

Claims

1. A method for predicting the remaining life of an IGBT device, characterized in that: include: An experiment simulating power cycle aging of an IGBT device is performed, and historical data of on-state voltage drop is collected according to the experiment, and empirical values ​​of model parameters and estimated values ​​of model parameters are obtained according to the historical data of on-state voltage drop; According to the current running state, a first state weight is constructed, and according to the model parameter experience value and the first state weight, the model parameter estimation value is smoothly adjusted to obtain a parameter correction result; A state transfer equation is established according to the parameter correction result, and the remaining life of the IGBT device is predicted according to the state transfer equation.

2. The method for predicting the remaining life of an IGBT device according to claim 1, characterized in that: The step of smoothly adjusting the estimated value of the model parameter according to the model parameter empirical value and the first state weight to obtain a parameter correction result includes: The first state weight is assigned to the estimated value of the model parameter to obtain a first weighted result, and the second state weight is assigned to the empirical value of the model parameter to obtain a second weighted result, and the sum of the first weighted result and the second weighted result is taken as the parameter correction result; wherein the sum of the first state weight and the second state weight is a preset value.

3. The method for predicting the remaining life of an IGBT device according to claim 2, characterized in that: The parameter correction result includes: a first parameter correction result and a second parameter correction result; the model parameter estimation value includes: a first model parameter estimation value and a second model parameter estimation value; the model parameter empirical value includes: a first model parameter empirical value and a second model parameter empirical value; wherein the first parameter correction result and the second parameter correction result are respectively expressed as: Among them, α k and β k are the first parameter correction result and the second parameter correction result respectively; and are the estimated values ​​of the first model parameter and the second model parameter respectively; α and β are the empirical values ​​of the first model parameter and the second model parameter; is the first state weight, is the second state weight.

4. The method for predicting the remaining life of an IGBT device according to claim 1, characterized in that: The current operating state includes: the current voltage state of the IGBT device; and constructing the first state weight according to the current operating state includes: A first state weight is constructed according to a ratio of the current voltage state to a failure threshold of the IGBT device.

5. The method for predicting the remaining life of an IGBT device according to claim 1, characterized in that: Collecting the on-state voltage drop historical data according to the experiment, and obtaining the model parameter empirical value and the model parameter estimated value according to the on-state voltage drop historical data, including: According to the experiment, historical data of on-state voltage drop are collected, and according to the historical data of on-state voltage drop, multiple groups of time series data of on-state voltage drop change are obtained, and according to the time series data, empirical values ​​of model parameters are obtained.

6. The method for predicting the remaining life of an IGBT device according to claim 5, characterized in that: The step of obtaining the empirical value of the model parameter according to the time series data includes: An IGBT degradation process model is established using the on-state voltage drop data as a feature quantity, and sampling is performed in the time series data as a training data set for the IGBT degradation process model. Based on the training data set and the IGBT degradation process model, an empirical distribution of model parameters is obtained, and based on the empirical distribution, empirical values ​​of model parameters are obtained.

7. The method for predicting the remaining life of an IGBT device according to claim 6, characterized in that: Collecting the on-state voltage drop historical data according to the experiment, and obtaining the model parameter estimation value according to the on-state voltage drop historical data, including: According to the time series data and based on the range value of the empirical distribution, the model parameters are constrained, and the model parameters are estimated by a particle swarm optimization algorithm to obtain model parameter estimation values.

8. The method for predicting the remaining life of an IGBT device according to claim 1, characterized in that: The step of establishing a state transfer equation according to the parameter correction result and predicting the remaining life of the IGBT device according to the state transfer equation includes: According to the current operating state and the parameter correction result, a particle filter state vector and a state transfer equation are jointly constructed, and the particle filter state vector is updated multiple times according to the state transfer equation until the posterior distribution of the state vector under the condition of the on-state voltage drop historical data is obtained, and the remaining life of the IGBT device is predicted based on the state transfer equation at this time.

9. The method for predicting the remaining life of an IGBT device according to claim 8, characterized in that: The remaining life prediction of the IGBT device using the state transfer equation at this time includes: The current particle filter state vector is obtained, and the future state distribution is predicted according to the state transfer equation at this time. If the remaining life is defined as the time required for the particles of the particle filter state vector to reach the failure threshold for the first time from the current moment, the remaining life of the IGBT device is obtained according to the future state distribution.

10. An IGBT device remaining life prediction system, characterized in that: include: Historical data acquisition module, correction module and prediction module; among them, The historical data acquisition module is used to simulate the experiment of power cycle aging of IGBT devices, collect the on-state voltage drop historical data according to the experiment, and obtain the model parameter empirical value and the model parameter estimated value respectively according to the on-state voltage drop historical data; The correction module is used to construct a first state weight according to the current running state, and to smoothly adjust the estimated value of the model parameter according to the model parameter experience value and the first state weight to obtain a parameter correction result; The prediction module is used to establish a state transfer equation according to the parameter correction result, and predict the remaining life of the IGBT device according to the state transfer equation.

Images