A method for detecting silicon carbide power devices with self-detection of hot spots

Abstract

The present invention discloses a method for detecting a silicon carbide power device with self-detection of hot spots, which belongs to the technical field of semiconductor power device detection. The method specifically comprises: monitoring device temperature, environment and electrical parameters in real time by integrating temperature-sensitive elements and multi-source data acquisition and fusion technology, and constructing accurate device status data; processing thermal imaging images by using a deep learning algorithm, accurately identifying normal and abnormal heating areas, and locating hot spots in combination with electrical parameters; constructing a multi-dimensional risk assessment model, quantitatively assessing heating risks, and automatically generating a detection report; and automatically formulating and implementing optimal control measures through a real-time feedback mechanism and an intelligent decision-making system, thereby ensuring the safe and stable operation of silicon carbide power devices and improving system reliability and maintenance efficiency.

Description

Technical Field

[0001] The invention belongs to the technical field of semiconductor power device detection, and in particular is a silicon carbide power device detection method with self-detection of hot spots. Background Art

[0002] Silicon carbide power devices have shown great potential in high-temperature, high-frequency, and high-power applications in electric vehicles, smart grids, and renewable energy systems due to their excellent properties of high breakdown electric field, high thermal conductivity, and low dielectric constant. However, in actual applications, silicon carbide power devices may produce local hot spots under high-intensity working conditions. If these hot spots cannot be detected and handled in time, they will seriously affect the performance and life of the device, and even cause the failure of the entire system.

[0003] Existing detection methods mostly focus on static electrical performance, dynamic switching characteristics and overall reliability testing, while there are few direct detection methods for hot spots inside devices. The operations are complex and costly, making it difficult to detect hot spots inside devices in real time and accurately. At the same time, it is impossible to promptly warn and prevent potential failures.

[0004] For example, the patent with the authorization announcement number CN107179494B discloses a method and system for detecting faults of components on a universal circuit board, including: by collecting visible light images and infrared thermal images of the circuit board under test, firstly using the image registration algorithm to calibrate the infrared thermal image coordinate system with the visible light image as the reference, then using the power device packaging database as the template image, searching the visible light image of the circuit board under test, determining the position coordinates and area size of the power device on the circuit board under test, and then processing the infrared thermal image of the circuit board under test with the black body as the background to calibrate the infrared thermal image, extracting the temperature, position coordinates and area size of the hot spot in the infrared thermal image; finally, judging and locating the faulty component through analysis and matching. This technical solution does not require the prior collection of infrared thermal images of circuit boards of the same type and model that are working normally, and can be used not only for factory mass production, but also for various application occasions such as equipment maintenance.

[0005] For example, the patent with publication number CN112180227A discloses a non-contact online detection system and method for junction temperature of silicon carbide power devices, including: controlling the silicon carbide power device to work at a set operating temperature through a temperature control unit, controlling the silicon carbide power device to work through a driving unit, so that current flows through the parasitic body diode of the silicon carbide power device to generate electroluminescence; detecting and obtaining the operating temperature, conduction current and luminous intensity of the silicon carbide power device, and establishing a function model; and detecting and calculating the working junction temperature of the silicon carbide power device according to the function model. This technical solution derives the junction temperature by detecting the luminous intensity of the impurity or defect energy level of the silicon carbide power device, realizes non-contact measurement, has the characteristics of inherent electrical isolation, and is particularly suitable for online junction temperature detection of silicon carbide power devices working in high temperature, high pressure and large current applications, with high accuracy and real-time performance.

[0006] The above existing technologies all have the following problems: CN107179494B mainly relies on image registration and template matching technology to identify and locate faults. Although this method is highly versatile, its accuracy and depth may not be as good as those based on temperature-sensitive components and deep learning algorithms when dealing with complex or subtle faults. In addition, it may take a long time and high computing resources to process images and match templates, especially when dealing with high-resolution images and complex circuit boards; CN112180227A mainly relies on the electroluminescence phenomenon of silicon carbide power devices to derive junction temperature. Although this method is non-contact and has electrical isolation characteristics, it requires the device to have sufficient electroluminescence effect and is not suitable for all types of silicon carbide power devices, especially those with weak luminescence effects. At the same time, its measurement accuracy may be affected by many factors, such as the luminous efficiency, spectral distribution, ambient temperature and sensitivity of the detection equipment of the device. In addition, the relationship between luminous intensity and junction temperature may be affected by device aging, material defects or process changes, resulting in a decrease in measurement accuracy and a lack of comprehensive assessment of the overall status and potential risks of the device. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention proposes a silicon carbide power device detection method with self-detection of hot spots. By integrating temperature-sensitive elements and multi-source data acquisition and fusion technology, the device temperature, environment and electrical parameters are monitored in real time to construct accurate device status data; deep learning algorithms are used to process thermal images to accurately identify normal and abnormal heating areas, and hot spots are located in combination with electrical parameters; a multi-dimensional risk assessment model is constructed to quantitatively assess heating risks and automatically generate detection reports; through a real-time feedback mechanism and an intelligent decision-making system, optimal control measures are automatically formulated and implemented to ensure the safe and stable operation of silicon carbide power devices and improve system reliability and maintenance efficiency.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method for detecting a silicon carbide power device with self-detection of hot spots, comprising:

[0010] Step S1: monitor the temperature change data of the silicon carbide power device in real time through a temperature sensitive element, and at the same time, collect the environmental data and electrical parameter data of the silicon carbide power device when it is working, and perform preprocessing and multi-source data fusion to obtain device status data;

[0011] Step S2: Generate a thermal image according to the temperature change data, and use a deep learning algorithm to extract features and recognize patterns on the thermal image to automatically distinguish between normal temperature distribution and abnormal heating areas. At the same time, use an electrical parameter algorithm to locate the heating point inside the heating area.

[0012] Step S3: construct a multi-dimensional risk assessment model based on the device status data, conduct quantitative assessment on the identified hot spots, and automatically generate a test report based on the quantitative assessment results;

[0013] Step S4: Establish a feedback mechanism to transmit the test report to the test equipment in real time, and automatically formulate the optimal control measures based on the evaluation results through the built-in intelligent decision-making system.

[0014] Specifically, the specific steps of step S2 include:

[0015] S2.1: Acquire temperature change data of the silicon carbide power device captured by the algorithm built into the infrared thermal imager, and generate a thermal image based on the temperature change data using a predefined color mapping scheme;

[0016] S2.2: Preprocess the thermal image and use the pre-trained thermal imaging-deep learning model to automatically extract features from the pre-processed thermal image;

[0017] S2.3: Input the extracted features into the classifier of the thermal imaging-deep learning model for pattern recognition, and output the probability p of each feature category n , and the category with the largest probability max(p n ) as the temperature distribution state of the area, where n represents the number of feature categories.

[0018] Specifically, the specific steps of step S2 also include:

[0019] S2.4: Set the probability threshold h1,

[0020] If max(p n )

[0021] ​If max(p n )≥h1, the heating area is an abnormal heating area;

[0022] S2.5: The identified abnormal heating area is integrated with the corresponding electrical parameter data, and the electrical parameter algorithm is used in combination with the temperature change data to analyze the heating cause in the abnormal heating area;

[0023] S2.6: Based on the analysis results, use the improved outlier detection algorithm to identify abnormal temperature points in the temperature change data, match the spatial data with the identified abnormal temperature points, determine the actual physical location of the abnormal temperature points in the silicon carbide power device, and verify it.

[0024] Specifically, the specific steps of S2.5 include:

[0025] S2.51: Obtain electrical parameter algorithm data, and use a weighted fusion method to fuse the temperature data of the abnormal heating area with the electrical parameter data according to time and space dimensions;

[0026] S2.52: Use the electrical parameter algorithm to analyze the fused data and calculate the current density J, resistance distribution R, and power loss P in the abnormal heating area;

[0027] S2.53: Use the correlation analysis method to analyze the relationship between the electrical parameter analysis result Y=f(J, R, P) and the temperature change data, where f(·) is the electrical parameter analysis function;

[0028] S2.54: construct and train an abnormal fever analysis model based on the correlation analysis results, and use the trained abnormal fever analysis model to predict and analyze new abnormal fever areas;

[0029] S2.55: Analyze the cause of abnormal heating based on the output results of the abnormal heating analysis model and combined with domain knowledge.

[0030] Specifically, the specific steps of identifying abnormal temperature points in S2.6 include:

[0031] S2.61: Obtain temperature change data in the abnormal heating area X = {x1, ..., x N}, where x N represents the temperature change data in the Nth abnormal heating area, where N represents the number of temperature change data in the abnormal heating area;

[0032] S2.62: Use the improved outlier detection algorithm to detect X = {x1, ..., x N} for identification, the formula of the improved outlier detection algorithm is:

[0033]

[0034] Among them, u i represents the recognition result of the ith temperature change data in the abnormal heating area, Z(x i ) represents the median of X, Φ -1 (·) represents the inverse cumulative distribution function of the standard normal distribution, α represents the quantile position index, |*| represents the absolute deviation, Z(|x i -Z(x i )|) means |x i -Z(x i )|median;

[0035] S2.63: Set the temperature threshold h2,

[0036] If u i ≤h2, then x i It is a normal point in the abnormally hot area;

[0037] If u i >h2, then x i It is an abnormal point in the abnormal heating area.

[0038] Specifically, the specific steps of step S3 include:

[0039] S3.1: Set risk assessment dimensions and evaluation indicators, use machine learning methods, and combine domain knowledge to build a multi-dimensional risk assessment model;

[0040] S3.2: The identified hotspot data is input into the multidimensional risk assessment model. The multidimensional risk assessment model calculates the quantitative score of the hotspot based on the input data. The formula is:

[0041]

[0042] Among them, core wen Indicates the quantitative score of the hot spot, W max Indicates the upper limit of normal temperature, W min Indicates the lower limit of normal temperature, K up represents the temperature rise rate, t represents the length of time the temperature remains at the current level or above the temperature threshold, δ, ε, η represents the weight coefficient, q W It represents the number of times the temperature exceeds the temperature threshold within a period of time Δt, and m represents the exponential factor;

[0043] S3.3: Generate a preliminary test report based on the quantitative assessment results, and review and revise the generated report.

[0044] Specifically, the color mapping scheme in S2.1 is represented by red and blue tones, where red represents high temperature and blue represents low temperature.

[0045] Specifically, the risk assessment dimensions in S3.1 include: temperature, current, voltage, power fluctuation, and historical temperature anomaly records; the assessment indicators include: temperature rise rate, current overload multiple, and voltage fluctuation range.

[0046] Specifically, the control measures in step S4 include adjusting working parameters, optimizing heat dissipation design, early warning notification or automatic shutdown.

[0047] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a method for detecting a silicon carbide power device with self-detection of hot spots are implemented.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention proposes a silicon carbide power device detection method with self-detection of hot spots, which can instantly capture the temperature changes of silicon carbide power devices during operation, ensure the real-time and accuracy of data, and combine the feature extraction and pattern recognition of thermal imaging images with deep learning algorithms. It can automatically distinguish between normal and abnormal temperature distributions, realize accurate positioning of hot spots, and improve the efficiency and accuracy of fault detection.

[0050] 2. The present invention proposes a silicon carbide power device detection method with self-detection of hot spots, which not only focuses on temperature data, but also integrates environmental data and electrical parameter data. Through preprocessing and multi-source data fusion technology, comprehensive device status data is obtained. Based on these data, a multi-dimensional risk assessment model is constructed to quantitatively evaluate hot spots, comprehensively consider multiple factors, and provide more comprehensive and in-depth risk assessment results. At the same time, the built-in intelligent decision-making system can automatically formulate optimal control measures according to the evaluation results, realizing the intelligence and automation of detection and control, which not only reduces the burden of manual operation, but also improves the response speed and decision-making accuracy, which helps to timely discover and solve potential problems and ensure the stable operation of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a framework diagram of a method for detecting silicon carbide power devices with self-detection of hot spots according to the present invention;

[0052] Figure 2 This is a working flow chart of a method for detecting a silicon carbide power device with self-detection of hot spots according to the present invention;

[0053] Figure 3 A flow chart of determining the location of a hot spot in a method for detecting a silicon carbide power device with self-detection of hot spots according to the present invention;

[0054] Figure 4 The present invention provides a hot spot self-detection method for a silicon carbide power device and a hot spot evaluation flow chart. DETAILED DESCRIPTION

[0055] Example 1

[0056] See also Figure 1-Figure 2 , an embodiment provided by the present invention: a method for detecting a silicon carbide power device with self-detection of a hot spot, comprising the following steps:

[0057] Step S1: monitor the temperature change data of the silicon carbide power device in real time through a temperature sensitive element, and at the same time, collect the environmental data and electrical parameter data of the silicon carbide power device when it is working, and perform preprocessing and multi-source data fusion to obtain device status data;

[0058] Among them, electrical parameter data include: current, voltage, resistance, power; environmental data include: temperature, humidity, air pressure.

[0059] Step S2: Generate a thermal image according to the temperature change data, and use a deep learning algorithm to extract features and recognize patterns on the thermal image to automatically distinguish between normal temperature distribution and abnormal heating areas. At the same time, use an electrical parameter algorithm to locate the heating point inside the heating area.

[0060] Furthermore, the steps of generating the thermal image include:

[0061] (1) Data acquisition: Use an infrared thermal imager to scan the target object and obtain infrared radiation data on its surface;

[0062] (2) Data conversion: converting infrared radiation data into temperature values, usually done through algorithms or software built into the infrared thermal imager;

[0063] (3) Image generation: Generate a thermal image based on the temperature value, where different colors or brightness represent different temperature ranges.

[0064] Step S3: construct a multi-dimensional risk assessment model based on the device status data, conduct quantitative assessment on the identified hot spots, and automatically generate a test report based on the quantitative assessment results;

[0065] In the present invention, the constructed multi-dimensional risk assessment model comprehensively considers the temperature level, duration, location distribution, and impact on electrical parameters of the hot spots, and conducts an in-depth quantitative assessment of the hot spots. During the assessment process, the design specifications, operating conditions, and historical fault records of the device are also considered to ensure the accuracy and comprehensiveness of the assessment results. Then, based on the quantitative assessment results, a test report containing detailed assessment information, risk level, recommended measures, and priority ranking is automatically generated to provide support for subsequent decision-making.

[0066] Step S4: Establish a feedback mechanism to transmit the test report to the test equipment in real time, and automatically formulate the optimal control measures based on the evaluation results through the built-in intelligent decision-making system.

[0067] Furthermore, the control measures in step S4 include adjusting working parameters, optimizing heat dissipation design, early warning notification or automatic shutdown.

[0068] Example 2

[0069] See also Figure 3-Figure 4 In this embodiment, the specific steps of step S2 include:

[0070] S2.1: Acquire temperature change data of the silicon carbide power device captured by the algorithm built into the infrared thermal imager, and generate a thermal image based on the temperature change data using a predefined color mapping scheme;

[0071] Further, the color mapping scheme in S2.1 is represented by red and blue tones, where red represents high temperature and blue represents low temperature, wherein the mapping is achieved through a lookup table or an interpolation algorithm, and the mapped color code is used to generate a thermal image in an image processing software or a programming environment.

[0072] S2.2: Preprocess the thermal image and use the pre-trained thermal imaging-deep learning model to automatically extract features from the pre-processed thermal image;

[0073] Among them, the extracted features include: temperature gradient, temperature distribution pattern, and abnormal hot spots.

[0074] Furthermore, the specific steps of S2.2 include:

[0075] (1) Load the original thermal imaging data from the infrared thermal imager, ensure that the image data format is correct, usually grayscale or pseudo-color, and perform preprocessing;

[0076] (2) Load the pre-trained thermal imaging-deep learning model to ensure that the model structure, weights, and configuration are consistent with those during training;

[0077] (3) passing the preprocessed thermal image as input to the feature extraction part of the thermal imaging-deep learning model, where the feature extraction part usually includes multiple convolutional layers, pooling layers, and possible activation layers, which work together to extract useful feature representations from the image;

[0078] (4) Obtain output features from the feature extraction part of the thermal imaging-deep learning model, which will be used for subsequent classification and regression.

[0079] S2.3: Input the extracted features into the classifier of the thermal imaging-deep learning model for pattern recognition, and output the probability p of each feature category n , and the category with the largest probability max(p n ) as the temperature distribution state of the region, where n represents the number of feature categories;

[0080] S2.4: Set the probability threshold h1,

[0081] If max(p n )

[0082] If max(p n )≥h1, the heating area is an abnormal heating area;

[0083] S2.5: The identified abnormal heating area is integrated with the corresponding electrical parameter data, and the electrical parameter algorithm is used in combination with the temperature change data to analyze the heating cause in the abnormal heating area;

[0084] S2.6: Based on the analysis results, use the improved outlier detection algorithm to identify abnormal temperature points in the temperature change data, match the spatial data with the identified abnormal temperature points, determine the actual physical location of the abnormal temperature points in the silicon carbide power device, and verify it.

[0085] The specific steps of S2.5 include:

[0086] S2.51: Obtain electrical parameter algorithm data, and use a weighted fusion method to fuse the temperature data of the abnormal heating area with the electrical parameter data according to time and space dimensions;

[0087] S2.52: Use the electrical parameter algorithm to analyze the fused data and calculate the current density J, resistance distribution R, and power loss P in the abnormal heating area;

[0088] Further, P=I 2 R,

[0089] ​Wherein, A represents the cross-sectional area of ​​the abnormal heating region, I represents the current passing through the cross-sectional area A, g represents the length of the silicon carbide power device, and ρ represents the resistivity.

[0090] S2.53: Use the correlation analysis method to analyze the relationship between the electrical parameter analysis result Y=f(J, R, P) and the temperature change data, where f(·) is the electrical parameter analysis function;

[0091] S2.54: construct and train an abnormal fever analysis model based on the correlation analysis results, and use the trained abnormal fever analysis model to predict and analyze new abnormal fever areas;

[0092] Furthermore, the specific steps of S2.54 include:

[0093] (1) Collect temperature change data including normal and abnormal heating areas from infrared thermal imagers and clean the data, including removing noise, processing missing values, and standardizing or normalizing to ensure data quality and consistency;

[0094] (2) determine the characteristics associated with abnormal fever, such as the rate, duration, and spatial distribution of temperature change;

[0095] (3) Using statistical methods to calculate the correlation between features to identify factors that have a significant impact on abnormal fever. In the present invention, the Pearson correlation coefficient is selected for correlation calculation. The Pearson correlation coefficient calculation formula is the prior art content in this field and is not an inventive solution of the present application, so it will not be described in detail here.

[0096] (4) According to the characteristics of the data and the analysis objectives, a machine learning algorithm based on random forest is selected to construct an abnormal fever analysis model, wherein the process of constructing the model using random forest is the prior art content in this field and is not the inventive solution of the present application, and is not described in detail here;

[0097] (5) Use a training data set containing known data of normal and abnormal fever areas to train a random forest model so that the model can learn the difference between normal and abnormal fever;

[0098] (6) Use the validation dataset to evaluate the performance of the random forest model, such as accuracy, recall, and F1 score;

[0099] (7) Adjust the random forest model parameters or algorithm selection based on the evaluation results to improve the performance of the random forest model;

[0100] (8) Apply the trained random forest model to new data or actual scenarios to predict abnormal fever areas;

[0101] (9) Analyze the prediction results, including the location, degree, and possible causes of abnormal fever.

[0102] S2.55: Analyze the cause of abnormal heating based on the output results of the abnormal heating analysis model and combined with domain knowledge.

[0103] Among them, the output results of the abnormal fever analysis model include: the area, degree and probability of abnormal fever.

[0104] The specific steps for identifying abnormal temperature points in S2.6 include:

[0105] S2.61: Obtain temperature change data X={x1,...,x N}, where x N represents the temperature change data in the Nth abnormal heating area, where N represents the number of temperature change data in the abnormal heating area;

[0106] S2.62: Use the improved outlier detection algorithm to detect X = {x1, ..., x N} for identification, the formula of the improved outlier detection algorithm is:

[0107]

[0108] Among them, u i represents the recognition result of the ith temperature change data in the abnormal heating area, Z(x i ) represents the median of X, Φ -1 (·) represents the inverse cumulative distribution function of the standard normal distribution, α represents the quantile position index, usually α=0.75, |*| represents the absolute deviation, Z(|x i -Z(x i )|) means |x i -Z(x i )|median;

[0109] S2.63: Set the temperature threshold h2,

[0110] If u i ≤h2, then x i It is a normal point in the abnormally hot area;

[0111] If u i >h2, then x i It is an abnormal point in the abnormal heating area.

[0112] The specific steps of step S3 include:

[0113] S3.1: Set risk assessment dimensions and assessment indicators, use machine learning methods, and combine domain knowledge to build a multi-dimensional risk assessment model. The process of building a model using machine learning methods is the prior art content in this field, and is not an inventive solution of this application, and will not be elaborated here;

[0114] Furthermore, the risk assessment dimensions in S3.1 include: temperature, current, voltage, power fluctuation, and historical temperature anomaly records; the assessment indicators include: temperature rise rate, current overload multiple, and voltage fluctuation range.

[0115] S3.2: The identified hotspot data is input into the multidimensional risk assessment model. The multidimensional risk assessment model calculates the quantitative score of the hotspot based on the input data. The formula is:

[0116]

[0117] in, Indicates the quantitative score of the hot spot, W max Indicates the upper limit of normal temperature, W min Indicates the lower limit of normal temperature, [W min ,W max ] defines the temperature range in which the device operates normally, K up represents the temperature rise rate, that is, how fast the temperature changes, t represents the length of time the temperature remains at the current level or above the temperature threshold, δ, ε, η represents the weight coefficient, which is used to adjust the influence of different parameters on the overall risk score. These coefficients can be determined by experiments, expert evaluation or machine learning algorithms. W It indicates the number of times the temperature exceeds the temperature threshold within a period of time Δt. m indicates an exponential factor, and m>0. It is used to adjust the risk growth rate when the temperature approaches the upper limit of the normal temperature. When m is large, the risk score will increase rapidly when the temperature is slightly close to the upper limit.

[0118] The formula for the comprehensive risk assessment model is:

[0119] core toal =F(core wen ,core I ,core V );

[0120] Among them, core toal Indicates that core I Indicates the current score, core V represents the voltage score, and F(·) represents a function trained by a machine learning algorithm, which can comprehensively consider the scores of multiple dimensions and output a comprehensive risk score.

[0121] S3.3: Generate a preliminary test report based on the quantitative assessment results, and review and revise the generated report.

[0122] Example 3

[0123] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a method for detecting a silicon carbide power device with self-detection of hot spots are implemented. For details, please refer to the above method embodiment, which will not be repeated here.

[0124] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take 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.

[0125] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes 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 a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the purpose and scope of protection of the present invention, and all of these are within the protection of the present invention.

Claims

1. A method for detecting silicon carbide power devices with self-detection of hot spots, characterized in that: include: Step S1: monitor the temperature change data of the silicon carbide power device in real time through a temperature sensitive element, and at the same time, collect the environmental data and electrical parameter data of the silicon carbide power device when it is working, and perform preprocessing and multi-source data fusion to obtain device status data; Step S2: Generate a thermal image according to the temperature change data, and use a deep learning algorithm to extract features and recognize patterns on the thermal image to automatically distinguish between normal temperature distribution and abnormal heating areas. At the same time, use an electrical parameter algorithm to locate the heating point inside the heating area. Step S3: construct a multi-dimensional risk assessment model based on the device status data, conduct quantitative assessment on the identified hot spots, and automatically generate a test report based on the quantitative assessment results; Step S4: Establish a feedback mechanism to transmit the test report to the test equipment in real time, and automatically formulate the optimal control measures according to the evaluation results through the built-in intelligent decision-making system; The specific steps of step S2 include: S2.1: Acquire temperature change data of the silicon carbide power device captured by the algorithm built into the infrared thermal imager, and generate a thermal image based on the temperature change data using a predefined color mapping scheme; S2.2: Preprocess the thermal image and use the pre-trained thermal imaging-deep learning model to automatically extract features from the pre-processed thermal image; S2.3: Input the extracted features into the classifier of the thermal imaging-deep learning model for pattern recognition, and output the probability p of each feature category n , and the category with the largest probability max(p n ) as the temperature distribution state of the region, where n represents the number of feature categories; The specific steps of step S2 also include: S2.4: Set the probability threshold h1, If max(p n )<h1, then the heating area is the normal temperature distribution area; If max(p n )≥h1, the heating area is an abnormal heating area; S2.5: The identified abnormal heating area is integrated with the corresponding electrical parameter data, and the electrical parameter algorithm is used in combination with the temperature change data to analyze the heating cause in the abnormal heating area; S2.6: Based on the analysis results, use the improved outlier detection algorithm to identify abnormal temperature points in the temperature change data, match the spatial data with the identified abnormal temperature points, determine the actual physical location of the abnormal temperature points in the silicon carbide power device, and verify; The specific steps of S2.5 include: S2.51: Obtain electrical parameter algorithm data, and use a weighted fusion method to fuse the temperature data of the abnormal heating area with the electrical parameter data according to time and space dimensions; S2.52: Use the electrical parameter algorithm to analyze the fused data and calculate the current density J, resistance distribution R, and power loss P in the abnormal heating area; S2.53: Use the correlation analysis method to analyze the relationship between the electrical parameter analysis result Y=f(J, R, P) and the temperature change data, where f(·) is the electrical parameter analysis function; S2.54: construct and train an abnormal fever analysis model based on the correlation analysis results, and use the trained abnormal fever analysis model to predict and analyze new abnormal fever areas; S2.55: Analyze the cause of abnormal fever based on the output results of the abnormal fever analysis model and combined with domain knowledge; The specific steps of identifying abnormal temperature points in S2.6 include: S2.61: Obtain temperature change data X={x1,...,x N }, where x N represents the temperature change data in the Nth abnormal heating area, where N represents the number of temperature change data in the abnormal heating area; S2.62: Use the improved outlier detection algorithm to detect X = {x1, ..., x N } for identification, the formula of the improved outlier detection algorithm is: Among them, u i represents the recognition result of the ith temperature change data in the abnormal heating area, Z(x i ) represents the median of X, Φ -1 (·) represents the inverse cumulative distribution function of the standard normal distribution, α represents the quantile position index, |*| represents the absolute deviation, Z(|x i -Z(x i )|) means |x i -Z(x i )|median; S2.63: Set the temperature threshold h2, If u i ≤h2, then x i It is a normal point in the abnormally hot area; If u i >h2, then x i It is an abnormal point in the abnormal heating area.

2. A method for detecting a silicon carbide power device with self-detection of hot spots as claimed in claim 1, characterized in that: The specific steps of step S3 include: S3.1: Set risk assessment dimensions and evaluation indicators, use machine learning methods, and combine domain knowledge to build a multi-dimensional risk assessment model; S3.2: The identified hotspot data is input into the multidimensional risk assessment model. The multidimensional risk assessment model calculates the quantitative score of the hotspot based on the input data. The formula is: Among them, core wen Indicates the quantitative score of the hot spot, W max Indicates the upper limit of normal temperature, W min Indicates the lower limit of normal temperature, K up represents the temperature rise rate, t represents the length of time the temperature remains at the current level or above the temperature threshold, δ, ε, η represents the weight coefficient, q W It represents the number of times the temperature exceeds the temperature threshold within a period of time Δt, and m represents the exponential factor; S3.3: Generate a preliminary test report based on the quantitative assessment results, and review and revise the generated report.

3. A method for detecting a silicon carbide power device with self-detection of hot spots as claimed in claim 2, characterized in that: The color mapping scheme in S2.1 is represented by red and blue tones, with red representing high temperature and blue representing low temperature.

4. A method for detecting a silicon carbide power device with self-detection of hot spots as claimed in claim 3, characterized in that: The risk assessment dimensions in S3.1 include: temperature, current, voltage, power fluctuation, and historical temperature anomaly records; the assessment indicators include: temperature rise rate, current overload multiple, and voltage fluctuation range.

5. A method for detecting a silicon carbide power device with self-detection of hot spots as claimed in claim 4, characterized in that: The control measures in step S4 include adjusting operating parameters, optimizing heat dissipation design, early warning notification or automatic shutdown.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for detecting a silicon carbide power device with self-detection of hot spots as described in any one of claims 1 to 5 are implemented.

Images