Method for predicting service life of silicon carbide power device based on multi-sensor information fusion

Abstract

The invention discloses a silicon carbide power device life prediction method based on multi-sensor information fusion, and the method comprises the following steps: obtaining operation data of a silicon carbide power device, and constructing a multi-dimensional working condition feature data set of the device, the operation data including data collected by a plurality of sensors; establishing a degradation model of the device health state; fusing the multi-sensor data based on an evidence theory, and processing uncertainty and conflicts of the sensor data by adopting an evidence discount technology to obtain fused health data; using the fused health data to optimize the degradation model of the health state of the device to obtain a multi-sensor fusion life prediction model; and inputting data acquired in real time into the trained multi-sensor fusion life prediction model to realize life prediction output of the silicon carbide power device under a dynamic working condition. According to the invention, through real-time data fusion of various sensors, the service life and reliability of the device in an actual working environment can be evaluated more accurately.

Description

Technical Field

[0001] The present invention belongs to the field of life prediction of power electronic power semiconductor devices, and in particular relates to a life prediction method for silicon carbide power devices based on multi-sensor information fusion. Background Art

[0002] Silicon carbide (SiC) power devices, due to their excellent high-temperature, high-frequency, and high-efficiency characteristics, have gradually become an ideal choice for high-power power electronics applications. Compared with traditional silicon-based power devices, SiC devices have higher switching frequencies and higher temperature tolerance, and therefore have been widely used in electric vehicles, photovoltaic inverters, grid converters, and other fields. However, as SiC devices continue to operate in high-temperature and high-frequency environments, their internal structure is gradually subjected to multiple stresses such as stress, temperature cycling, and current and voltage changes. This can cause device aging and failure, thereby affecting the safety and stability of the entire system.

[0003] Current lifetime prediction methods for SiC power devices are often based on monitoring data from a single sensor. These methods collect single physical quantities, such as temperature and current, from the device under specific operating conditions to establish a lifetime prediction model. However, the amount of information a single sensor can monitor is limited, and it is difficult to fully reflect the device's overall health. For example, a temperature sensor can only capture changes in device surface temperature but cannot accurately reflect the thermal distribution within the device. While a stress sensor can monitor stress changes under high power, it struggles to distinguish between the coupled stress effects caused by current and temperature. These lifetime prediction methods based on single-sensor data often have limitations when operating under complex operating conditions, resulting in low prediction accuracy. Furthermore, the data collected by the sensor is often subject to interference from noise and drift. For example, a temperature sensor may experience zero drift under long-term high-temperature operation, causing the measured value to deviate from the actual temperature. Current sensors may also suffer from measurement errors due to electromagnetic interference. Noise and drift in the data not only affect the input data quality of the lifetime prediction model but also lead to biased prediction results. Therefore, to improve the accuracy of lifetime prediction for SiC power devices, a method is urgently needed to integrate information from multiple sensors and reduce errors caused by the limitations and noise of individual sensors. Lifespan prediction methods based on multi-sensor information fusion have already found application in other fields, such as aerospace and manufacturing, where they leverage information from multiple sensors for equipment condition monitoring and fault diagnosis. However, in the field of silicon carbide power device lifespan prediction, research based on multi-sensor fusion is still in its infancy, and the relevant technologies and applications are still immature.

[0004] In their Chinese patent application, "A Reliability Assessment Method for Silicon Carbide Power Devices (CN118861580A)," Hao Jianyong et al. disclose a reliability assessment method for silicon carbide power devices based on a multivariate relationship prediction model. By collecting device parameters, performing temperature cycling tests, and monitoring key electrical parameters in real time, combined with a deep learning algorithm, this method provides efficient and accurate reliability assessment reports. However, a limitation of this technical solution is that its lifetime assessment relies primarily on temperature cycling tests and single sensor data, using a deep learning model to predict the performance of silicon carbide power devices under different temperature conditions. Summary of the Invention

[0005] In order to solve at least one of the problems existing in the prior art, the present invention provides a method for predicting the life of silicon carbide power devices based on multi-sensor information fusion. The present invention uses multiple sensors (such as temperature, stress, current, and voltage sensors) to collect key data in device operation in real time, and comprehensively obtains the multi-dimensional working condition characteristics of the device under actual working conditions. Combined with machine learning algorithms, through multi-feature fusion, it is possible to effectively fuse different sensor data, capture the interaction relationship between each feature, and establish an accurate life prediction model. Through real-time data fusion of multiple sensors, a more accurate assessment of the life and reliability of the device in an actual working environment can be achieved.

[0006] A method for predicting the life of a silicon carbide power device based on multi-sensor information fusion includes the following steps:

[0007] S1: Multi-sensor data acquisition: Acquire the operating data of the silicon carbide power device and construct a multi-dimensional operating condition characteristic data set of the device. The operating data includes data collected by multiple sensors, including temperature changes, stress fluctuations, current and voltage parameters, etc., and perform pre-processing operations such as cleaning, denoising, and standardization;

[0008] S2: Construction of device health degradation model: Based on random processes, a device health degradation model is established to extract degradation characteristics of multi-source sensors, accurately capture the changing trend of device health status, and support remaining life prediction;

[0009] S3: Multi-sensor data fusion: Based on evidence theory, multi-sensor data is fused and the evidence discounting technique is used to deal with the uncertainty and conflict of sensor data to improve the accuracy and reliability of the fused data and obtain fused health data;

[0010] S4: Life prediction model training: Use the fused health data to optimize the degradation model of the device health status, obtain a multi-sensor fusion life prediction model, and calculate the probability density distribution of the remaining life of the device;

[0011] S5: Input the real-time collected data into the trained multi-sensor fusion life prediction model to realize the life prediction output of silicon carbide power devices under dynamic working conditions.

[0012] The life prediction model can provide life prediction results, including expected life and uncertainty range, to help equipment maintenance management

[0013] Furthermore, in step S1, based on the failure characteristics of power electronic devices, sensors such as temperature, stress, current and voltage are used to collect the working condition data of silicon carbide power devices in real time. s and the acquisition period T s To ensure that the collected multi-dimensional data is complete and timely. The data includes: temperature data T(t), which records the temperature of the device at each moment during operation; stress data σ(t), which collects the stress conditions of the device; current data I(t) and voltage data V(t), which reflect the electrical characteristics of the device. The collected sensor data is cleaned and marked to remove noise data and outliers. The health status data of each sensor is normalized and used as a source of evidence for data fusion. The collected multi-sensor data can form a time series matrix Used to comprehensively reflect the operating status of the device, where i=1,2,...,N represents N sensors.

[0014] Furthermore, in step S2, a random process is used to establish a degradation model for the device health status. Based on the monitoring data, a degradation function for the device health status is defined, and feature information from each sensor data is extracted to support subsequent data fusion. The specific implementation steps are as follows:

[0015] S21: Construct a degradation model for the device health state. The device health state can be represented by a time function. The device health state degradation model is defined as l(t) = l0 + μt + σB(t), where l(t) represents the degradation state at time t; l0 is the initial health state; μ is a drift parameter describing the average rate of decline in the device health state; σ is a diffusion parameter describing the volatility of the health state; and B(t) represents standard Brownian motion, which is used to introduce randomness.

[0016] S22: Use the collected multi-sensor degradation data to determine the unknown parameters in the model through maximum likelihood estimation. Input the sensor data into the model, calculate the degradation increment at each moment, and solve for the degradation rate μ and diffusion coefficient σ that best fit the actual data.

[0017] S23: Use some historical data or simulated experimental data to verify the degradation model of the device health status to ensure that the model fits the actual degradation trend well.

[0018] Furthermore, in step S3, the multi-sensor degradation information is fused through evidence theory to generate a comprehensive equipment health status estimate. The specific implementation steps are as follows:

[0019] S31: Construct the probability distribution of evidence sources. Treat the degradation data of each sensor as an independent evidence source S i , construct its basic probability assignment function (BPA) under the theory of evidence, defined as m i (A)=P(S i =A|(μ,σ)), where A is a hypothetical state in the identification framework to characterize the degree of device degradation, and m i (A) represents the basic probability distribution of the i-th sensor in state A, P(S i =A|(μ,σ) represents the basic probability distribution when the device degradation level is A, the drift parameter is μ, and the diffusion coefficient is σ. The basic probability distribution for each evidence source is calculated using the interval likelihood function method. The basic probability distribution defines the credibility of each piece of evidence for different hypotheses (different lifecycle stages).

[0020] S32: Evidence discount processing. To deal with the uncertainty and conflict of different sensor data, the PROMETHEEII method is used to discount the evidence. First, the evidence uncertainty of each sensor is calculated using the evidence entropy formula. Uncertainty is measured, where Θ is the set of all possible hypothetical states in the sample space. Secondly, the quality of evidence is evaluated using the similarity / difference of sensor data to achieve a dissimilarity measure. The dissimilarity measure between two pieces of evidence can be calculated using the Euclidean distance: Finally, the conflict degree between the evidences of each sensor is measured by Calculated, K(m i ,m j ) represents m i and m j KL divergence between j (A) represents the basic probability distribution of the jth sensor in state A. By weighting the above uncertainty, dissimilarity, and conflict, the discount factor α of each sensor is obtained. i =f(H(m i ),D(m i ,m j ),K(m i, m j )), f(H(m i ),D(m i, m j ),K(m i, mj )) is a metric function used to calculate the discount factor of each sensor. The calculated discount factor is used to adjust the basic probability distribution of each sensor and improve the reliability of data fusion.

[0021] S33: Evidence fusion. The improved Dempster-Shafer (DS) evidence combination rule is used to fuse the degradation information of multiple sensors to obtain a comprehensive estimate of the overall equipment health status. The fusion formula is Where K is the conflict coefficient, m i,j (A) represents the basic probability distribution of state A after fusing sensor i and sensor j. It is used for normalization to avoid excessive conflicts. For the fusion of multiple sensor data, a recursive method is used to gradually fuse all evidence.

[0022] Step 4: Further optimize the device health degradation model based on the fused health data to calculate the device’s current health state l(t), degradation rate μ, and diffusion parameter σ.

[0023] Using the degradation model of the Wiener process, the life prediction model is derived Where f(L(t)) is the probability density function of the predicted remaining life of the equipment; L(t) represents the remaining life of the equipment at time t; the uncertainty range of the remaining life prediction is σ and μ are the previously estimated diffusion parameter and drift parameter, respectively; t is the current time.

[0024] The life prediction loss function is calculated based on the predicted remaining life of the equipment and the actual remaining life of the monitored equipment: Where Loss represents the life prediction loss function value, N all represents the number of training samples, Represents the true remaining life value of sample n at time t. Figure 2 As shown, the device lifetime prediction model parameters are updated by back propagation according to the lifetime prediction loss function until the maximum number of training times e is reached, and a dynamically updated device lifetime prediction model is obtained.

[0025] The parameters of the degradation model can be updated based on subsequent sensor data to enhance the accuracy of the prediction model.

[0026] Furthermore, in step S5, the multi-sensor data collected in real time is input into the life prediction model to output the remaining life L(t) of the silicon carbide device. When the life approaches the preset threshold, L th , generate alarms to notify maintenance needs.

[0027] The present invention also provides a silicon carbide power device life prediction system based on multi-sensor information fusion.

[0028] The present invention also provides a computer device.

[0029] The present invention also provides a computer-readable storage medium.

[0030] Compared with the prior art, the present invention has at least the following beneficial effects:

[0031] 1. This invention effectively reduces data uncertainty and conflict through multi-sensor data fusion, improving the accuracy of remaining life prediction. It also addresses inconsistencies in sensor data through evidence discounting, enhancing the reliability of data fusion.

[0032] 2. This paper establishes a degradation model based on random processes, which can extract degradation characteristics of health status from multi-source sensor data, capture long-term degradation trends of equipment, and support accurate lifespan prediction. This process can provide more efficient degradation analysis for complex systems.

[0033] 3. The present invention realizes predictive maintenance through the remaining life prediction results, and can provide real-time monitoring and early warning functions of the equipment health status, effectively reducing the risk of unexpected downtime and maintenance costs, improving the stability and safety of equipment operation, and has wide application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is an overall flow chart of a method for predicting the life of a silicon carbide device provided by an embodiment of the present invention.

[0035] Figure 2 This is a multi-sensor information fusion life prediction flow chart provided by an embodiment of the present invention.

[0036] Figure 3 3 is a prediction curve diagram of the method proposed in the embodiment of the present invention in prediction task 1.

[0037] Figure 4 3 is a schematic diagram of the RMSE indicator of the method proposed in the embodiment of the present invention in four types of migration tasks. DETAILED DESCRIPTION

[0038] In order to make the technical solutions and purposes of the present invention more clearly understood, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific implementation steps described herein are only used to better illustrate the application of the present invention, but the technical features involved in the implementation methods of the present invention are not limited thereto.

[0039] See also Figure 1 The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion provided by an embodiment of the present invention specifically includes the following steps:

[0040] Step 1: Multi-sensor data acquisition: Based on the failure characteristics of power electronic devices, four types of sensors, temperature, stress, current, and voltage, are used to collect real-time operating data such as temperature changes, stress fluctuations, current, and voltage of silicon carbide power devices. The collected data is preprocessed to construct a multi-dimensional operating characteristic data set of the device.

[0041] In one embodiment of the present invention, by setting an appropriate sampling rate f s and the acquisition period T s , ensuring the integrity and timeliness of the collected multi-dimensional data. This data includes: temperature data T(t), which records the temperature of the device at every moment during operation; stress data σ(t), which captures the stress conditions on the device; and current data I(t) and voltage data V(t), which reflect the device's electrical characteristics.

[0042] The preprocessing operation includes: cleaning and labeling the collected sensor data to remove noise data and outliers; performing sliding window sampling on the multi-sensor data to obtain the health status data of each sensor; and normalizing the health status data of each sensor and using it as the evidence source for data fusion. The collected multiple sensor data can form a time series matrix It is used to comprehensively reflect the operating status of the device. The time series matrix X(t) is the multi-dimensional working condition feature data set, which contains data collected by multiple sensors at multiple time points, where i = 1, 2, ..., N represents N sensors, and x i (t) represents the observation value of the i-th sensor at time point t, x N (t) represents the observation value of the Nth sensor at time point t.

[0043] Step 2: Constructing a degradation model of the device health status: Using a random process (in one embodiment of the present invention, the random process is a Wiener process) to establish a degradation model of the device health status. Based on the Wiener process, the degradation model of the device health status is established, and the degradation characteristics in the operating data collected by multi-source sensors are extracted to accurately capture the changing trend of the equipment health status to support the remaining life prediction.

[0044] Based on the monitoring data, the degradation function of the equipment health status is defined, and the characteristic information in each sensor data is extracted to provide support for subsequent data fusion. The specific implementation steps are as follows:

[0045] S21: Construct a degradation model for the device health state. The device health state can be represented by a time function. The device health state degradation model is defined as l(t) = l0 + μt + σB(t), where l(t) represents the degradation state at time t; l0 is the initial health state; μ is a drift parameter describing the average rate of decline in the device health state; σ is a diffusion parameter describing the volatility of the health state; and B(t) represents standard Brownian motion, which is used to introduce randomness.

[0046] S22: Using the multi-sensor data collected in step 1, determine unknown parameters in the degradation model of the device health state through a maximum likelihood estimation method.

[0047] The multi-sensor data is input into the degradation model of the device health state, the degradation increment at each moment is calculated, and the drift parameter μ and diffusion coefficient σ that best fit the actual data are solved.

[0048] S23: Use some historical data or simulated experimental data to verify the degradation model of the device health state to ensure that the degradation model of the device health state fits the actual degradation trend well.

[0049] Step 3: Multi-sensor data fusion: Fusion of multi-sensor degradation information is performed using evidence theory, and evidence discounting techniques are used to handle uncertainty and conflict in sensor data to improve the accuracy and reliability of fused data and generate a comprehensive estimate of the equipment health status. The specific implementation steps are as follows:

[0050] S31: Construct the probability distribution of evidence sources. Treat the degradation data of each sensor as an independent evidence source S i , construct its basic probability assignment function (BPA) under the theory of evidence, defined as m i (A)=P(S i =A|(μ,σ)), where A is a hypothetical state in the identification framework to characterize the degree of device degradation, and m i (A) represents the basic probability distribution of the i-th sensor in state A, P(S i =A|(μ,σ) represents the basic probability distribution when the device degradation level is A, the drift parameter is μ, and the diffusion coefficient is σ. The basic probability distribution for each evidence source is calculated using the interval likelihood function method. The basic probability distribution defines the credibility of each piece of evidence for different hypotheses (different lifecycle stages).

[0051] S32: Evidence Discounting. To address the uncertainty and conflict of different sensor data, the PROMETHEEII method is used to discount the evidence. Evidence discounting is used to address the uncertainty and conflict of sensor data to improve the accuracy and reliability of the fused data.

[0052] First, the evidence uncertainty of each sensor is calculated using the evidence entropy formula To measure uncertainty, Θ is the set of all possible hypothetical states in the sample space.

[0053] Secondly, the similarity / difference of sensor data is used to evaluate the quality of evidence and achieve dissimilarity measurement. The dissimilarity measure between two pieces of evidence can be calculated using the Euclidean distance: And i≠j.

[0054] Finally, the conflict degree between the evidences of each sensor is measured by Calculated, K(m i ,m j ) represents m i KL divergence between mj and mj, m j (A) represents the basic probability distribution representing the basic probability distribution of the j-th sensor in state A.

[0055] By weighting the above uncertainty, dissimilarity, and conflict, the discount factor α of each sensor is obtained. i =f(H(m i ),D(m i ,m j ),K(m i ,m j )), f(H(m i ),D(m i ,m j ),K(m i ,m j )) is a metric function used to calculate the discount factor of each sensor. The calculated discount factor is used to adjust the basic probability distribution of each sensor to improve the reliability of data fusion.

[0056] S33: Evidence Fusion: Using the improved Dempster-Shafer (DS) evidence combination rule, the degradation information of multiple sensors is fused to obtain a comprehensive estimate of the overall equipment health status.

[0057] The fusion formula is Where K is the conflict coefficient, which is used to normalize and avoid excessive conflicts; m i,j(A) represents the basic probability distribution in state A after fusing sensor i and sensor j. For the fusion of multiple sensor data, a recursive method is used to gradually fuse all evidence.

[0058] Step 4: Further optimize the degradation model of the device health status based on the fused health data to obtain a multi-sensor fusion life prediction model; calculate the current health status of the device and provide life prediction results, including expected life and uncertainty range, to assist in equipment maintenance and management.

[0059] Using the degradation model of the device health state of the Wiener process, the device life prediction model is derived The probability density distribution of the remaining life of the device can be calculated through the device life prediction model. Among them, f(L(t)) is the probability density distribution of the predicted remaining life of the device; L(t) represents the remaining life of the device at time t; the uncertainty range of the remaining life prediction is σ and μ are the previously estimated diffusion parameter and drift parameter, respectively; t is the current time.

[0060] The life prediction loss function is calculated based on the predicted remaining life of the equipment and the actual remaining life of the monitored equipment: Where Loss represents the life prediction loss function value, N all represents the number of training samples, Indicates the true remaining life value of sample n at time t, L n (t) represents the predicted value of the remaining life of the equipment for sample n at time t.

[0061] like Figure 2 As shown in the figure, the device lifetime prediction model parameters are updated by back propagation according to the lifetime prediction loss function. The updated model parameters include: the drift parameter μ, the diffusion parameter σ in the model, and the weight coefficient and discount factor of each sensor in the fusion, etc., to dynamically reflect the changes in the contribution of different sensors until the maximum number of training times e is reached, and the dynamically updated device lifetime prediction model is obtained, that is, the multi-sensor fusion lifetime prediction model.

[0062] This application can update the parameters of the device life prediction model based on subsequent sensor data to enhance the accuracy of the prediction model.

[0063] Step 5: Real-time prediction, alarm and maintenance: Input the multi-sensor data collected in real time into the device life prediction model dynamically updated in step 4, output the remaining life L(t) of the silicon carbide device, and generate a detailed life assessment report later. When the life approaches the preset threshold L th When a fault occurs, an alarm is generated to notify maintenance needs.

[0064] The embodiments of the present invention can provide power device life prediction results and uncertainty ranges, and realize equipment failure early warning and predictive maintenance.

[0065] In one embodiment of the present invention, the threshold L is set based on historical life test data or industry standards. th In other embodiments, it can also be set based on maintenance records and expert experience.

[0066] In one of the embodiments of the present invention, further explanation is given with reference to the accompanying drawings and experimental cases. In order to evaluate the performance of the proposed silicon carbide power device life prediction method based on multi-sensor information fusion, this embodiment uses a public IGBT aging data set from the University of Padova, Italy, which provides IGBT aging data under four different working conditions (Device 2, Device 3, Device 4, Device 5) for testing. The package temperature in the experiment is controlled outside the rated temperature of the device to accelerate the aging process of the device. The monitored parameters include collector current, collector voltage, gate voltage, and package temperature.

[0067] Comparative analysis of various parameters revealed that the collector-emitter voltage (VCE) exhibits a very clear monotonic variation throughout the entire lifecycle. This monotonicity not only effectively reflects the degradation process of IGBT devices, but also accurately tracks the performance decline trend of devices during aging. Further analysis shows that the collector-emitter voltage (VCE) is highly robust, meaning it can stably characterize the device's failure progression despite varying operating conditions and noise interference. Therefore, the collector-emitter voltage (VCE) has been selected as the primary failure parameter for evaluating IGBT failure time, enabling relatively accurate prediction of the device's failure moment, providing a reliable basis for fault prediction and lifespan assessment.

[0068] The IGBT datasets are labeled as Device2, Device3, Device4, and Device5, and prediction tasks T1, T2, T3, and T4 are constructed on the four datasets. The first 80% of the samples in each dataset are used as the training set, and the last 20% of the samples are used as the test set. The collector-emitter voltage (VCE) prediction curve for task T1 is shown in Figure 1. Figure 3 As shown, the dotted line is the predicted value and the solid line is the true value. It can be seen that the method of the present invention can well capture the degradation characteristics in multi-sensor information and simulate the degradation trend of the device.

[0069] In this embodiment, the effectiveness of the device lifetime prediction model is further evaluated using the common model evaluation metric, the root mean square error (RMSE). RMSE is a commonly used metric to measure the difference between predicted and actual values. It reflects the magnitude of the model's prediction error; a smaller RMSE indicates a more accurate model prediction. The RMSE is calculated by taking the square root of the square mean of the prediction errors. It imposes a higher penalty on large errors and is therefore more sensitive to model errors. The RMSE metric is calculated using the final prediction results for each power electronic device. A smaller RMSE value indicates a stronger predictive capability of the model.

[0070] like Figure 4 As shown in the figure, in the four life prediction tasks, the RMSE values ​​obtained by the method provided by the embodiment of the present invention are all low, which proves that the method described in the embodiment of the present invention can not only effectively capture the life degradation trend of silicon carbide devices, but also learn its unique feature information, thereby verifying the importance of multi-sensor information fusion strategy in realizing accurate life assessment of electronic devices.

[0071] In one embodiment of the present invention, a silicon carbide power device life prediction system based on multi-sensor information fusion is provided, comprising the following modules:

[0072] A data acquisition module, configured to acquire operating data of a silicon carbide power device and construct a multi-dimensional operating condition characteristic data set of the device, wherein the operating data includes data collected by multiple sensors;

[0073] A model building module is used to build a degradation model of the device health status based on the Wiener process and extract degradation features from the operating data collected by multiple sensors;

[0074] The fusion module is used to fuse multi-sensor data based on evidence theory and use evidence discounting technology to deal with the uncertainty and conflict of sensor data to obtain fused health data;

[0075] An optimization module is used to optimize the degradation model of the device health state using the fused health data to obtain a multi-sensor fusion life prediction model; and calculate the probability distribution of the remaining life of the device;

[0076] The prediction module is used to input the real-time collected data into the trained multi-sensor fusion life prediction model to realize the life prediction output of silicon carbide power devices under dynamic working conditions.

[0077] In one embodiment of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method provided in the aforementioned embodiment when executing the computer program.

[0078] In one embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method provided in the above embodiment is implemented.

[0079] In summary, the silicon carbide power device life prediction method, system equipment, and medium based on multi-sensor information fusion proposed in the embodiments of the present invention can solve the problems of low multi-sensor information utilization and poor generalization ability of the model under different working conditions in silicon carbide device life prediction. By utilizing evidence theory to fully extract degradation features from multi-sensor data, this method significantly improves the adaptation speed and life prediction accuracy of the model in device life prediction tasks under different working conditions. In actual industrial applications, this method can effectively enhance the stability of equipment operation and maintenance efficiency, showing broad application potential and important value.

[0080] Finally, it should be noted that although the implementation of the present invention has been described in detail with reference to examples, it is easy for those skilled in the art to understand that any modifications, substitutions and improvements made without departing from the spirit and principles of the present invention as described in the appended claims should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the life of silicon carbide power devices based on multi-sensor information fusion, characterized in that: The real-time collected data is input into a trained multi-sensor fusion life prediction model to realize the life prediction output of the silicon carbide power device under dynamic working conditions, wherein the multi-sensor fusion life prediction model is obtained by the following steps: Acquiring operating data of a silicon carbide power device and constructing a multi-dimensional operating condition characteristic data set of the device, wherein the operating data includes data collected by multiple sensors; Establish a degradation model of device health status based on random processes and extract degradation features from operating data collected by multiple sensors; Based on evidence theory, multi-sensor data is fused and evidence discounting technology is used to deal with the uncertainty and conflict of sensor data to obtain fused health data; The fused health data is used to optimize the degradation model of the device health status to obtain a multi-sensor fusion life prediction model.

2. The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion according to claim 1, characterized in that: The step of obtaining the operating data of the silicon carbide power device and constructing a multi-dimensional operating condition characteristic data set of the device includes: Real-time collection of SiC power device operating data, including temperature changes, stress fluctuations, current and voltage parameters, through a variety of sensors; Preprocess the collected operating data to form a time series matrix It is used to fully reflect the operating status of the device, where i = 1, 2, ..., N represents N sensors, x i (t) represents the observation value of the i-th sensor at time point t, x N (t) represents the observation value of the Nth sensor at time point t, and the time series matrix is ​​the multi-dimensional working condition characteristic data set of the device.

3. The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion according to claim 2, characterized in that: The step of establishing a degradation model of the device health state based on a random process includes: The degradation model of the device health state is defined as l(t) = l0 + μt + σB(t), where l(t) represents the degradation state at time t; l0 is the initial health state; μ is the drift parameter; σ is the diffusion parameter; and B(t) represents the standard Brownian motion. The degradation data collected by multiple sensors are used to determine the unknown parameters in the degradation model of the device health state through the maximum likelihood estimation method.

4. The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion according to claim 1, characterized in that: The multi-sensor data is fused based on evidence theory, and the uncertainty and conflict of sensor data are processed by evidence discounting technology to obtain fused health data, including: The degradation data of each sensor is used as an independent source of evidence S i , construct its basic probability distribution function under the evidence theory, defined as m i (A)=P(S i =A|(μ,σ)), where A is an assumption in the identification framework to characterize the degree of device degradation, and m i (A) represents the basic probability distribution of the i-th sensor in state A, P(S i =A|(μ,σ)) represents the basic probability distribution when the device degradation degree is A, the drift parameter is μ and the diffusion coefficient is σ; Discounting the evidence to obtain a discount factor used to adjust the base probability assignment for each sensor; The degradation information from multiple sensors is fused to obtain a comprehensive estimate of the overall equipment health status.

5. The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion according to claim 4, characterized in that: The discount calculation of the evidence to obtain the discount factor for adjusting the basic probability distribution of each sensor includes: The evidence uncertainty of each sensor is calculated using the evidence entropy formula; Use similarities / differences in sensor data to assess the quality of evidence; Measure the degree of conflict between the evidence of each sensor; The discount factor is obtained through weighted calculation.

6. The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion according to claim 4, characterized in that: The improved Dempster-Shafer evidence combination rule is used to fuse the degradation information of multiple sensors to obtain a comprehensive estimate of the overall equipment health status. The fusion formula is: Where K is the conflict coefficient and Θ is the set of all possible hypothesis states in the sample space.

7. The method for predicting the life of a silicon carbide power device based on multi-sensor information fusion according to any one of claims 1 to 6, characterized in that: The method uses the fused health data to optimize the degradation model of the device health state, obtains a multi-sensor fusion life prediction model, and calculates the probability density distribution of the remaining life of the device, including: Using the degradation model of device health status of the Wiener process, a device life prediction model is derived; The probability density distribution of the remaining life of the device is calculated through the device life prediction model; Calculate the life prediction loss function based on the predicted remaining life of the equipment and the actual remaining life of the monitored equipment; The device lifetime prediction model parameters are updated based on the back propagation of the lifetime prediction loss function to obtain a multi-sensor fusion lifetime prediction model.

8. A silicon carbide power device life prediction system based on multi-sensor information fusion, characterized in that: Includes the following modules: A data acquisition module, configured to acquire operating data of a silicon carbide power device and construct a multi-dimensional operating condition characteristic data set of the device, wherein the operating data includes data collected by multiple sensors; A model building module is used to build a degradation model of the device health status based on the Wiener process and extract degradation features from the operating data collected by multiple sensors; The fusion module is used to fuse multi-sensor data based on evidence theory and use evidence discounting technology to deal with the uncertainty and conflict of sensor data to obtain fused health data; An optimization module is used to optimize the degradation model of the device health state using the fused health data to obtain a multi-sensor fusion life prediction model; and calculate the probability distribution of the remaining life of the device; The prediction module is used to input the real-time collected data into the trained multi-sensor fusion life prediction model to realize the life prediction output of silicon carbide power devices under dynamic working conditions.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images