A multi-chip IGBT module health state monitoring method based on shell temperature change

Abstract

The present application relates to the technical field of device health monitoring, and discloses a multi-chip IGBT module health state monitoring method based on shell temperature change, comprising: arranging transient temperature sensors to form a sensor array; injecting a narrow pulse electric heat excitation signal to make heat propagate in the form of heat waves inside the module; recording the time point at which the temperature rising rate of each sensor first reaches a peak value, and calculating the heat wave arrival time difference value to form a time difference vector; comparing the time difference vector with a reference time difference vector under a healthy state, calculating a time difference offset vector, and judging the overall health degree of the module according to the module value; calculating the projection position of the aging point on the module shell plane to locate the chip area where aging occurs; and outputting a health grade according to the comparison result of the health degree and a preset health degree threshold, and simultaneously outputting aging chip position information. The present application realizes accurate positioning of the aging chip of the multi-chip IGBT module, does not require high-voltage isolation, and is sensitive to early aging.

Description

Technical Field

[0001] This invention relates to the field of device health monitoring technology, and in particular to a method for monitoring the health status of a multi-chip IGBT module based on changes in housing temperature. Background Technology

[0002] Multi-chip IGBT modules are widely used in new energy power generation, electric vehicle traction, and rail transportation. Power module failure is primarily induced by temperature factors.

[0003] Currently, health status monitoring methods for multi-chip IGBT modules are mainly divided into two categories: one is based on electrical parameters, which judges the degree of aging by monitoring changes in on-state voltage drop or threshold voltage. However, this method requires high-voltage isolation and is difficult to distinguish the aging differences between different chips. The other is based on the casing temperature rise, which assesses the thermal state of the module by measuring the casing temperature. However, most existing methods treat the module as a whole, ignoring the uneven temperature distribution caused by thermal coupling between chips inside the multi-chip module, and cannot locate the specific chip that has experienced aging online.

[0004] Therefore, this invention proposes a health status monitoring method for multi-chip IGBT modules based on changes in housing temperature. Summary of the Invention

[0005] The purpose of this invention is to solve the problems of inability to locate online, thermal coupling interference, and the need for high-voltage isolation in the prior art. To this end, a health status monitoring method for multi-chip IGBT modules based on shell temperature changes is proposed.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature, comprising the following steps: Step S1: Arrange multiple non-collinearly distributed transient temperature sensors on the housing surface of the multi-chip IGBT module to form a sensor array. Step S2: During the standby state, a narrow pulse electrothermal excitation signal is injected into the module to cause heat to propagate inside the module in the form of thermal waves, so that the responses of each sensor are distinguishable in time. Step S3: Record the time when the temperature rise rate of each sensor first reaches its peak after narrow pulse excitation, and calculate the time difference of the arrival of the heat wave between any two sensors to form a time difference vector. Step S4: Compare the time difference vector with the pre-calibrated baseline time difference vector under healthy conditions, calculate the time difference offset vector, and determine the overall health of the module based on the magnitude of the time difference offset vector. Step S5: Calculate the projection position of the aging point on the module housing plane based on the direction of the time difference offset vector and the geometric arrangement of the sensor array, and locate the chip area where aging has occurred. Step S6: Output the health level based on the comparison result between the health level and the preset health level threshold, and output the location information of the aging chip at the same time.

[0007] The beneficial effects of the technical solution provided by this invention include at least the following: This invention injects a narrow-pulse electrothermal excitation signal into a multi-chip IGBT module, records the time point when the temperature rise rate of each sensor first reaches its peak, and calculates the time difference vector. The magnitude of the time difference offset vector is used to determine the overall health of the module. At the same time, the aging chip area is located based on the direction of the time difference offset vector and the geometric arrangement of the sensor array. This enables non-invasive online health monitoring of multi-chip IGBT modules without the need to measure junction temperature or on-state voltage drop, thus avoiding high-voltage isolation problems.

[0008] This invention utilizes the time difference of heat wave arrival as a health characteristic to naturally decouple the thermal coupling effect between multiple chips, effectively distinguish the aging degree of different chips, and achieve precise positioning of aging chips, thus solving the technical problem that existing technologies cannot locate specific aging chips online.

[0009] This invention eliminates the interference of cooling condition changes on monitoring results by comparing the time difference of the shell temperature response with a pre-calibrated benchmark, thereby improving the robustness and accuracy of the monitoring results and making it suitable for complex field conditions.

[0010] This invention extracts the dynamic feature of the peak time of temperature rise rate, which has high sensitivity to early aging and can provide effective early warning information before module failure, thus providing decision support for predictive maintenance. Attached Figure Description

[0011] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention. Detailed Implementation

[0013] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a multi-chip IGBT module health status monitoring method based on shell temperature changes proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0014] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0015] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0016] The following description, in conjunction with the accompanying drawings, details a specific scheme for a health status monitoring method for a multi-chip IGBT module based on casing temperature changes provided by the present invention.

[0017] Please see Figure 1 The diagram illustrates a flowchart of a method for monitoring the health status of a multi-chip IGBT module based on changes in housing temperature, according to an embodiment of the present invention. The method includes the following steps: Step S1: Arrange multiple non-collinearly distributed transient temperature sensors on the housing surface of the multi-chip IGBT module to form a sensor array. Step S2: During the standby state, a narrow pulse electrothermal excitation signal is injected into the module to cause heat to propagate inside the module in the form of thermal waves, so that the responses of each sensor are distinguishable in time. Step S3: Record the time when the temperature rise rate of each sensor first reaches its peak after narrow pulse excitation, and calculate the time difference of the arrival of the heat wave between any two sensors to form a time difference vector. Step S4: Compare the time difference vector with the pre-calibrated baseline time difference vector under healthy conditions, calculate the time difference offset vector, and determine the overall health of the module based on the magnitude of the time difference offset vector. Step S5: Calculate the projection position of the aging point on the module housing plane based on the direction of the time difference offset vector and the geometric arrangement of the sensor array, and locate the chip area where aging has occurred. Step S6: Output the health level based on the comparison result between the health level and the preset health level threshold, and output the location information of the aging chip at the same time.

[0018] It should be noted that a multi-chip IGBT module refers to a power module that integrates multiple insulated gate bipolar transistor chips, with each chip connected in parallel or series to carry a large current.

[0019] The housing surface refers to the outer surface of the IGBT module package, located on the heat conduction path between the chip and the heat sink, and is a location that can be touched and its temperature measured.

[0020] Non-collinear distribution means that any three sensors in the sensor array are not on the same straight line, ensuring that the sensors can provide the orientation information required for two-dimensional spatial positioning.

[0021] Transient temperature sensors are temperature sensors with a response time of less than 1 millisecond, capable of capturing rapid changes in temperature signals, including NTC thermistors, K-type thermocouples, or infrared temperature sensors.

[0022] A sensor array is a collection of multiple temperature sensors arranged in a predetermined geometric position to collect temperature change information at different spatial locations on the surface of a housing.

[0023] Narrow pulse electrothermal excitation signal refers to an electrical pulse signal with a pulse width less than one-tenth of the module's thermal time constant, applied between the collector and emitter of the IGBT module, causing instantaneous power loss inside the module and thus generating heat.

[0024] Thermal wave refers to the wave-like propagation of heat within the module encapsulation layer at a finite speed, distinct from steady-state thermal diffusion, and manifested as the propagation of a temperature rise front.

[0025] A time-distinguishable response means that there is a measurable difference in the time points at which different sensors receive the heat wave. This difference is greater than the inherent resolution of the time measurement system, allowing the responses of each sensor to be distinguished.

[0026] The rate of temperature rise refers to the first derivative of temperature with respect to time, which is the amount of temperature change per unit time. It is used to locate the moment when the front of a heat wave arrives at the sensor.

[0027] The moment when the temperature rise rate first reaches its peak is the moment when the temperature rise rate reaches its first local maximum after rising from zero. This moment marks the arrival of the heat wave front at the sensor.

[0028] The time to arrival of a heat wave refers to the time elapsed from the start of narrow pulse excitation to the peak of the sensor's temperature rise rate, characterizing the propagation time required for a heat wave to travel from the heat source to the sensor.

[0029] The heat wave arrival time difference refers to the difference in the arrival time of heat waves between any two sensors, reflecting the order and time interval of the arrival of heat waves at these two sensors.

[0030] The time difference vector is a one-dimensional vector formed by arranging the arrival time differences of the thermal waves from all sensor pairs in a fixed order, and serves as a characteristic representation of the module's thermal state.

[0031] The reference time difference vector refers to the statistical characteristic value of the time difference vector obtained through multiple measurements during the module's factory calibration phase or under known health conditions. It is usually the median or average value.

[0032] The time difference offset vector refers to the difference between the measured time difference vector and the reference time difference vector, reflecting the change in the module's current thermal state relative to its healthy state.

[0033] The modulus refers to the Euclidean norm of a vector, which is the square root of the sum of the squares of its components. It is used to quantify the overall size of the time difference offset vector.

[0034] Health status is a quantitative indicator that comprehensively reflects the degree of aging of a module. It is inversely proportional to the magnitude of the time difference offset vector. The larger the magnitude, the more severe the aging.

[0035] The direction of the time difference offset vector refers to the direction of the time difference offset vector in multidimensional space, reflecting the distribution pattern of the time difference offset of each sensor, and is related to the spatial location of the aging point.

[0036] The geometric arrangement of a sensor array refers to the two-dimensional coordinates of each sensor on the plane of the housing and the relative positional relationship between them.

[0037] The projection position of the aging point on the module shell plane refers to the coordinate position of the actual aging point inside the module projected vertically onto the shell surface along the heat propagation path.

[0038] The aging chip area refers to the spatial range where the aging chip is located inside the module. The location result can be a single chip or an area composed of multiple adjacent chips.

[0039] Health level refers to the level classified according to the comparison between health status and preset threshold, including health level, attention level and failure level.

[0040] The preset health threshold refers to one or more numerical limits pre-set to classify health levels. In this invention, it specifically refers to the first health threshold and the second health threshold, used to divide the comprehensive health index into three levels: healthy, attentive, and malfunctioning. The acquisition rules are as follows: determined through statistical analysis of a large amount of measured comprehensive health data from healthy modules and modules with different aging levels. First, the comprehensive health value of healthy modules is collected, and its upper boundary (e.g., 0.5) is taken as the first threshold; the comprehensive health value of malfunctioning modules is collected, and its lower boundary (e.g., 1.0) is taken as the second threshold. The thresholds can be calibrated and adjusted according to different module models and application scenarios.

[0041] In one specific implementation, a six-chip parallel IGBT module is used as an example. This module integrates six IGBT chips arranged in a two-row, three-column rectangular pattern. Each chip measures 8 mm × 8 mm, and the overall module housing dimensions are 320 mm × 280 mm. This module is installed in a 55 kW frequency converter to drive an industrial fan. The frequency converter operates continuously for 16 hours daily and is shut down for 8 hours overnight.

[0042] Step S1 implementation: First, the main sensor group is arranged. The module contains six chips, therefore six main sensors are set up. Each main sensor uses a type K thermocouple with a response time constant of 0.5 milliseconds, and is attached to the corresponding chip's housing surface directly below it using thermal grease. After finite element simulation optimization, the planar coordinates (unit: millimeters) of the six main sensors are set as follows: Main sensor 1 (corresponding chip 1): (82, 82) Main sensor 2 (corresponding chip 2): (122, 82) Main sensor 3 (corresponding chip 3): (162, 82) Main sensor 4 (corresponding chip 4): (82, 122) Main sensor 5 (corresponding chip 5): (122,122) Main sensor 6 (corresponding chip 6): (162, 122) These points are not collinear on the shell plane, forming a rectangular array of two rows and three columns. The spacing between adjacent main sensors is 40 mm in the X direction and 40 mm in the Y direction. The chip's projection area on the shell surface is a square with a side length of 8 mm, and its minimum circumscribed circle diameter is 11.3 mm (8 x √2). Therefore, the spacing of 40 mm between adjacent main sensors is much greater than 11.3 mm, satisfying the minimum spacing constraint. The distance from the sensor to the left boundary of the shell is 82 mm, the right boundary distance is 320 - 162 = 158 mm, the lower boundary distance is 82 mm, and the upper boundary distance is 280 - 122 = 158 mm. Half of each boundary distance is 41 mm, 79 mm, 41 mm, and 79 mm, respectively. The maximum spacing between adjacent main sensors is 40 mm, which is less than half of each boundary distance (minimum is 41 mm), satisfying the maximum spacing constraint.

[0043] Next, the auxiliary sensor group was arranged. One auxiliary sensor was placed in the triangular gap area formed by every three adjacent chips, for a total of four auxiliary sensors. The planar coordinates of the four auxiliary sensors were determined through geometric calculations as follows: Auxiliary sensor A: (102, 102) Auxiliary sensor B: (142, 102) Auxiliary sensor C: (102, 142) Auxiliary sensor D: (142, 142) Taking auxiliary sensor A (102, 102) as an example, its two adjacent main sensors are (82, 82) and (122, 82). The calculated vectors are (20, 20) and (-20, 20), their dot product is -400 plus 400 equals 0, and the included angle is 90 degrees. The bisector of this angle is vertically upward (positive Y-axis direction). The main direction of heat diffusion in the module encapsulation layer, calibrated by finite element simulation, is also vertically upward, consistent with the above. The other auxiliary sensors are verified similarly.

[0044] All signal lines from the main and auxiliary sensors are connected to the same multi-channel synchronous data acquisition card, with a sampling frequency set to 10 kHz for subsequent temperature data acquisition.

[0045] Step S2 is implemented as follows: Excitation injection is performed in the standby state before the inverter starts each day. The temperature of the chip gap area is read by an auxiliary sensor. The current measured temperature of the chip gap area is 26.3 degrees Celsius, the preset reference temperature is 25 degrees Celsius, and the temperature difference is 1.3 degrees Celsius. The preset temperature difference threshold is 2 degrees Celsius. The current temperature difference is less than the threshold, so it is determined that the inverter is in standby state.

[0046] Next, the pulse width of the narrow pulse is calculated. The module's junction-to-case thermal resistance is 0.1 Kelvin per watt, its heat capacity is 5 Joules per Kelvin, and its thermal time constant is 0.5 seconds. One-tenth of this is 0.05 seconds, or 50 milliseconds. The distance between the nearest neighbor main sensors is 40 mm. The thermal wave velocity of the module's encapsulation material, silicone, is calibrated to 2000 mm / s. The theoretical thermal wave propagation time difference is 40 divided by 2000, which equals 0.02 seconds, or 20 milliseconds. Half of this is 10 milliseconds. Taking the smaller of the two upper limits, 10 milliseconds, as the upper limit of the pulse width, the narrow pulse width is set to 8 milliseconds, which is less than both 50 milliseconds and 10 milliseconds, satisfying both constraints.

[0047] Then, pulse excitation is performed. The gate drive circuit applies a steep-front single pulse with a rise time of 0.5 microseconds and a pulse width of 8 milliseconds to the gate of the IGBT module. At the same time, a test current is applied between the collector and emitter, which is set to 30% of the module's rated current of 200 amps, i.e., 60 amps. The pulse is automatically turned off after 8 milliseconds, and the module enters the cooling phase.

[0048] The specific implementation of step S3: The data acquisition card continuously acquires the temperature data of each sensor at a sampling rate of 10 kHz, with an acquisition time of 200 milliseconds, covering the entire process of pulse injection and cooling.

[0049] First, signal preprocessing and peak detection are performed. The temperature time series acquired by each sensor is decomposed into four levels using the Daubechies 4 wavelet. The first-level detail coefficients are set to zero to remove high-frequency noise, and then the temperature series is reconstructed. The temperature rise rate is calculated using a three-point difference formula. Taking the main sensor 1 as an example, its root mean square value of the temperature rise rate sequence is 0.08 degrees Celsius per millisecond, and the adaptive threshold is set to three times the root mean square value, i.e., 0.24 degrees Celsius per millisecond. Scanning begins after excitation; when the temperature rise rate of three consecutive sampling points exceeds the threshold and then decreases, the last maximum point before the decrease is determined as the peak value.

[0050] Health status calibration (module at factory): In a healthy state without any aging, due to the symmetrical layout, all main sensors have the same thermal arrival time. By taking the median of 10 repeated measurements, the original thermal arrival time of each main sensor in the healthy state is found to be 12.0 milliseconds (actually calibrated). The thermal arrival time of the auxiliary sensors is also symmetrical; for example, the auxiliary sensor (102,102) has a thermal arrival time of 20.0 milliseconds.

[0051] Actual aging test (after 3 years of module operation, assuming bonding wire aging occurred in chip 2): The measured arrival time of the original heat waves of each main sensor is as follows: Main sensor 1: 12.1 milliseconds Main sensor 2: 13.8 milliseconds (significantly increased due to increased local thermal resistance caused by bonding wire aging). Main sensor 3: 12.3 milliseconds Main sensor 4: 12.2 milliseconds Main sensor 5: 12.4 milliseconds Main sensor 6: 12.3 milliseconds The arrival time of the auxiliary sensor's thermal wave is less affected by aging and remains at approximately 20.0 milliseconds.

[0052] Then, the thermal coupling contribution weight is calculated. Taking the auxiliary sensor (102, 102) as an example, its adjacent main sensors are (82, 82) and (122, 82), with a geometric distance of 28.28 mm (√20² + 20²). For isotropic media, each sensor has a weight of 0.5. Coupling component: The coupling amount of each main sensor is equal to the arrival time of the auxiliary sensor multiplied by 0.5, i.e., 20.0 multiplied by 0.5 equals 10.0 milliseconds. Therefore, the arrival time of the decoupled pure thermal wave is equal to the original arrival time minus 10.0 milliseconds. Main sensor 1: 12.1 minus 10.0 equals 2.1 milliseconds Main sensor 2: 13.8 minus 10.0 equals 3.8 milliseconds Main sensor 3: 12.3 minus 10.0 equals 2.3 milliseconds. Main sensor 4: 12.2 minus 10.0 equals 2.2 milliseconds Main sensor 5: 12.4 minus 10.0 equals 2.4 milliseconds. Main sensor 6: 12.3 minus 10.0 equals 2.3 milliseconds. The clean heat wave arrival time under healthy conditions (also decoupled) should be 12.0 minus 10.0 equals 2.0 milliseconds (consistent across all chips). Therefore, aging caused the clean time of chip 2 to increase from 2.0 to 3.8 milliseconds, while the other chips remained largely unchanged.

[0053] Finally, a time difference vector is constructed. The pairwise Euclidean distances between the six main sensors are calculated. Sort by distance from smallest to largest, the first few sensor pairs (adjacent horizontally or vertically, 40 mm apart) are: Main sensor 1 and main sensor 2, Main sensor 2 and main sensor 3, Main sensor 4 and main sensor 5, Main sensor 5 and main sensor 6, Main sensor 1 and main sensor 4, Main sensor 2 and main sensor 5, Main sensor 3 and main sensor 6. The absolute values ​​of the arrival time differences of the pure thermal waves for these sensor pairs are calculated: The difference between main sensor 1 and main sensor 2 is |2.1−3.8| = 1.7 milliseconds. The difference between main sensor 2 and main sensor 3 is |3.8−2.3|=1.5 milliseconds. The difference between main sensor 4 and main sensor 5 is |2.2−2.4|=0.2 milliseconds. The difference between main sensor 5 and main sensor 6 is |2.4−2.3|=0.1 milliseconds. The difference between main sensor 1 and main sensor 4 is |2.1−2.2|=0.1 milliseconds. The difference between main sensor 2 and main sensor 5 is |3.8−2.4|=1.4 milliseconds. The difference between main sensor 3 and main sensor 6 is |2.3−2.3|=0.0 milliseconds. The differences between the remaining, more distant sensor pairs are smaller. They are calculated in the same way and arranged sequentially to form a 15-dimensional time difference vector. At the same time, the sensor pair index corresponding to each difference is recorded to form a mapping table.

[0054] The specific implementation of step S4: Reference data has been established during the module's factory calibration phase. Under healthy conditions, the arrival time of the reference pure thermal wave is 2.0 milliseconds; therefore, the reference time difference vector (the time difference between all sensor pairs) is a zero vector. The reference peak amplitude vector is obtained through repeated measurements: the peak temperature rise rate of each main sensor is 0.32 degrees Celsius per millisecond.

[0055] In actual monitoring, the reference temperature at the substrate edge is first read, currently measured at 45 degrees Celsius. Based on the pre-calibrated velocity-temperature relationship: the thermal wave velocity (mm / s) equals 2000 minus 2.5 multiplied by (temperature minus 20). The correction factor is calculated: at 20 degrees Celsius, the velocity is 2000 mm / s; at 45 degrees Celsius, the velocity is 2000 minus 2.5 multiplied by 25, equaling 1937.5 mm / s. The correction factor is approximately 2000 divided by 1937.5, which is approximately 1.032. Each element of the measured time difference vector is multiplied by 1.032 for velocity correction. After correction, the above differences become approximately 1.75, 1.55, 0.21, 0.10, 0.10, 1.45, and 0.00 milliseconds, etc.

[0056] Then, redundancy verification is performed. Taking the triangle formed by main sensor 1, main sensor 2, and main sensor 4 as an example, the theoretical relationship should be that the time difference between main sensor 1 and main sensor 2 plus the time difference between main sensor 1 and main sensor 4 is approximately equal to the time difference between main sensor 2 and main sensor 4. After correction, the time difference between main sensor 1 and main sensor 2 is 1.75 milliseconds, the time difference between main sensor 1 and main sensor 4 is 0.10 milliseconds, and the time difference between main sensor 2 and main sensor 4 is 1.45 milliseconds. 1.75 plus 0.10 equals 1.85 milliseconds, which differs from 1.45 milliseconds by 0.40 milliseconds. Considering the changes in the heat wave propagation path caused by actual aging, this difference is within the allowable range (preset allowable deviation 0.5 milliseconds), therefore all sensor pairs pass the redundancy verification without discarding any data.

[0057] Calculate the time difference offset vector, which is the measured time difference vector minus the reference time difference vector. Since the reference is zero, the offset vector is the measured time difference vector. Calculate the Euclidean norm of this vector: the square root of the sum of the squares of each component. The squares of each component are: 1.75 squared approximately 3.06, 1.55 squared approximately 2.40, 0.21 squared approximately 0.04, 0.10 squared approximately 0.01, 0.10 squared approximately 0.01, 1.45 squared approximately 2.10, and the remaining smaller contributions are ignored. The sum of squares is approximately 3.06 + 2.40 + 0.04 + 0.01 + 0.01 + 2.10 = 7.62, and the square root is approximately 2.76 milliseconds. Since the reference is zero, the time difference health component is set to 2.76.

[0058] Simultaneously calculate the amplitude health component. The measured peak amplitude vector is (unit: degrees Celsius per millisecond): Main sensor 1: 0.30 Main sensor 2: 0.27 Main sensor 3: 0.31 Main sensor 4: 0.31 Main sensor 5: 0.32 Main sensor 6: 0.31 The baseline peak amplitude is 0.32. The amplitude attenuation vector components are calculated by dividing the measured value by the baseline value: 0.30 divided by 0.32 equals 0.938, 0.27 divided by 0.32 equals 0.844, 0.31 divided by 0.32 equals 0.969, 0.31 divided by 0.32 equals 0.969, 0.32 divided by 0.32 equals 1.000, and 0.31 divided by 0.32 equals 0.969. The magnitude of this vector is calculated by taking the square root of the sum of the squares of each component. The squares are 0.880, 0.712, 0.939, 0.939, 1.000, and 0.939, respectively, with a total sum of approximately 5.409. The square root is approximately 2.326. The amplitude health component is set to 2.326.

[0059] Based on the application scenario of this module (high risk of bonding wire aging), the weight of the time difference health component is set to 0.4, and the weight of the amplitude health component is set to 0.6. The overall health index is equal to 0.4 multiplied by 2.76 plus 0.6 multiplied by 2.326, which is 1.104 plus 1.396 equals 2.50.

[0060] The specific implementation of step S5 is as follows: First, the basic positioning process is executed. The time difference offset vector is normalized by dividing each component by the modulus of the vector, i.e., obtaining the unit direction vector. The mapping matrix established during the factory calibration stage is queried. This matrix is ​​a 2-row, 15-column matrix obtained through finite element simulation. The unit direction vector is multiplied by the mapping matrix to obtain the projected coordinates of the aging point on the shell plane. Based on the simulation data, the projected coordinates are calculated to be (122, 84) mm.

[0061] Match the projected coordinates with the projection area of ​​each chip's casing (the chip projection area is a square with a side length of 8 mm centered on the chip's center): Chip 1: X from 78 to 86, Y from 78 to 86 Chip 2: X increases from 118 to 126, Y increases from 78 to 86 Chip 3: X increases from 158 to 166, Y increases from 78 to 86 Chip 4: X from 78 to 86, Y from 118 to 126 Chip 5: X from 118 to 126, Y from 118 to 126 Chip 6: X increases from 158 to 166, Y increases from 118 to 126 The projection coordinates (122, 84) have an X of 122 (between 118 and 126) and a Y of 84 (between 78 and 86), which fall within the projection area of ​​chip 2. Therefore, chip 2 is determined to be an aging chip.

[0062] To verify the robustness of the localization, a multi-combination voting fusion method was employed. All possible combinations of three non-collinear sensors from the six main sensors were selected, resulting in 20 possible subarray combinations. After removing collinear combinations, 16 effective combinations remained. Candidate projected coordinates were calculated independently for each group. Of the 16 candidate coordinates, 12 were concentrated in the range of 121–123 mm in the X direction and 83–85 mm in the Y direction; 3 were in the range of 124–126 mm in the X direction and 85–87 mm in the Y direction; and 1 was in the range of 115–117 mm in the X direction and 80–82 mm in the Y direction. After cluster analysis, the cluster containing the most coordinates had 12 coordinates, with its center coordinates at (122, 84) mm, consistent with the basic localization.

[0063] Finally, a location reliability assessment is performed. The center of the projection area of ​​chip 2 is (122, 82) mm, and the radius of its circumscribed circle is 5.66 mm (half the diagonal of the chip). The Euclidean distance between the projection coordinates (122, 84) and the center is |84 - 82|, which equals 2 mm. The normalized positioning error is equal to the distance divided by the circumscribed circle radius, i.e., 2 divided by 5.66, approximately equal to 0.353. The location reliability is equal to 1 minus the normalized error, i.e., 1 minus 0.353, which equals 0.647. The preset lower threshold is 0.6. The location reliability of 0.647 is greater than 0.6, indicating reliable positioning. Output chip 2 is an aged chip with a confidence level of 0.647.

[0064] Step S6 is implemented as follows: First, a health level determination is performed. The overall health index is 2.50, the preset first health threshold is 0.5, and the second health threshold is 1.0. Since 2.50 is greater than 1.0, it is determined to be at a failure level.

[0065] Since the health level is deemed failed, the failure handling procedure is executed. The location of the aging chip determined in step S5 is read as chip 2. The location of the sensor with the highest temperature rise rate in step S2 is read: among all main sensors, the peak amplitude of main sensor 2 corresponding to chip 2 drops from the baseline of 0.32 to 0.27, the largest decrease (0.05), and its absolute peak value of 0.27 is lower than other sensors. Therefore, the location of the sensor with the highest temperature rise rate coincides with the location of the aging chip, a failure alarm is output, and replacement of chip 2 is recommended.

[0066] Data such as health level (failure), location of aging chip (chip 2), and positioning confidence level (0.647) are uploaded to the inverter main control unit via the CAN communication interface. Simultaneously, a temperature cloud map is generated in the host computer monitoring software. Based on the real-time housing temperature data collected in step S1 (temperature values ​​measured by the six main sensors are 51°C, 58°C, 53°C, 49°C, 55°C, and 52°C respectively), a color distribution map of the entire housing surface is generated using a bilinear interpolation algorithm. The highest temperature is at the location of chip 2, indicated by red, with the surrounding area appearing yellow and green. The location of the aging chip is marked on the cloud map with a flashing red circle.

[0067] The above detailed implementation fully demonstrates the application process of the present invention's method on a six-chip parallel IGBT module. From the sensor array arrangement in step S1 (satisfying all geometric constraints), the narrow pulse excitation injection in step S2, the thermal wave arrival time detection and decoupling processing in step S3, the comprehensive health calculation in step S4, the aging chip location in step S5 (including multi-combination voting fusion and confidence assessment), to the output and visualization in step S6, each step is logically coherent and the data is self-consistent. The final location result shows that chip 2 is an aging chip, with a comprehensive health index of 2.50 reaching the failure level, and a location confidence score of 0.647. Maintenance personnel arranged maintenance based on the failure alarm and temperature cloud map. After disassembly, they found a bond line crack in chip 2, consistent with the monitoring results, verifying the effectiveness of the present invention. The entire monitoring process is automatically completed in the inverter's standby state, with a single monitoring time of approximately 200 milliseconds, without affecting the normal operation of the inverter.

[0068] Step S1 further includes the following sub-steps: S1-1, multiple transient temperature sensors are divided into a main sensor group and an auxiliary sensor group. Each sensor in the main sensor group is arranged in a position directly opposite to the surface of the housing directly below each chip, and all main sensors are non-collinearly distributed on the housing plane. Each sensor in the auxiliary sensor group is arranged in the middle position between the chips. The main sensor group and the auxiliary sensor group together form a non-collinearly distributed sensor array. S1-2, Set the minimum spacing between adjacent main sensors to be greater than the maximum value of the minimum outer circle diameter of the two adjacent chips projected on the surface of the housing, and the maximum spacing to be less than half the distance from the sensor to the boundary of the module housing; S1-3, Set the number of main sensors to be the same as the number of chips inside the module, with each main sensor corresponding to one chip; S1-4, an auxiliary sensor is set at the center of each chip gap area, and the angle between the auxiliary sensor and the line connecting the two adjacent main sensors is 90 degrees. The bisector of this angle is consistent with the main direction of thermal diffusion of the module packaging layer.

[0069] It should be noted that the main sensor group refers to a collection of temperature sensors, the same number as the number of chips inside the module. Each main sensor is positioned directly opposite the corresponding chip on the surface of the housing, and is used to collect the arrival time of the heat wave generated by that chip.

[0070] The auxiliary sensor group refers to a collection of temperature sensors arranged in the center of the gap area between chips, used to collect thermal coupling interference signals generated by adjacent chips.

[0071] The position directly opposite the chip on the surface of the housing refers to the position on the surface of the module housing that is vertically aligned with the geometric center of the chip, i.e., the vertical projection point of the chip center onto the housing plane.

[0072] Non-collinear distribution means that all the main sensors are not on the same straight line on the housing plane, ensuring that the sensor array can provide the orientation information required for two-dimensional spatial positioning.

[0073] The minimum spacing constraint between adjacent main sensors refers to ensuring that the heat wave propagation path is not blocked by adjacent chips. This requires the spacing between adjacent main sensors to be greater than the maximum of the minimum circumscribed circle diameters of the projection areas of the two adjacent chips. The rule for obtaining this constraint is: each chip's projection area is a square with a side length of 8 mm, and its minimum circumscribed circle diameter is 11.3 mm. The larger of the projection area diameters of two adjacent chips is used; if the chip sizes are the same, then 11.3 mm is the acceptable value. The method of application is to arrange the sensors so that the actual spacing between adjacent main sensors is greater than this value. In this embodiment, the actual spacing is 40 mm, which meets the requirement.

[0074] The maximum spacing refers to the requirement that the distance between adjacent main sensors should be less than half the distance from the sensor to the module housing boundary to avoid interference from boundary reflected waves directly reaching the heat wave. The rule for obtaining this distance is: measure the shortest distance from each main sensor to the four boundaries of the housing, take the minimum value, and then divide it by 2. For example, if the sensor is 82 mm from the left boundary, half is 41 mm; if it's 158 mm from the right boundary, half is 79 mm, etc., and the minimum half is 41 mm. Usage: The actual distance between adjacent main sensors must be less than this value. In this embodiment, the actual distance is 40 mm, which is less than 41 mm, thus satisfying the constraint.

[0075] The middle position of the chip gap refers to the geometric center point of the area enclosed by the projections of multiple adjacent chips on the housing plane. This point is approximately equidistant from the projection edges of each adjacent chip.

[0076] A sensor array is a set of temperature sensors consisting of a main sensor group and an auxiliary sensor group, with each sensor arranged on the surface of the housing according to a preset geometric position.

[0077] Adjacent master sensors refer to the two master sensors that are spatially closest in a sensor array, usually corresponding to two adjacent chips.

[0078] The surface projection area of ​​the casing refers to the area covered by the vertical projection of the chip onto the casing plane, and its shape and size are consistent with the actual size of the chip.

[0079] The minimum circumscribed circle diameter refers to the diameter of the smallest circle that can completely cover the projected area of ​​the chip casing surface, and is used to quantify the size of the chip's projected area.

[0080] The distance from the sensor to the module housing boundary refers to the shortest straight-line distance from the sensor location along the housing plane to the edge of the module housing.

[0081] The main direction of heat diffusion refers to the direction in which heat diffuses laterally fastest within the module encapsulation layer. It is determined by the anisotropy of the thermal conductivity of the encapsulation material and is usually consistent with the orientation of the copper busbar in the DBC layer.

[0082] The angle bisector is the angle bisector formed by the lines connecting the auxiliary sensor and the two adjacent main sensors. This line divides the included angle into two equal parts.

[0083] In step S1, this application solves the problem of difficult differentiation of thermal coupling interference in multi-chip IGBT modules by dividing the sensors into main sensor groups and auxiliary sensor groups and optimizing their spatial layout. The main sensors face the center of each chip to ensure the purity of the direct heat wave; the auxiliary sensors are located between the chips to specifically collect the thermal coupling signals of adjacent chips. The minimum spacing between adjacent main sensors is set by geometric constraints to ensure that the heat wave propagation path is not blocked by other chips; at the same time, the maximum spacing is set to avoid boundary reflection interference. The angle between the auxiliary sensor and the line connecting the two adjacent main sensors is 90 degrees, and its bisector is consistent with the main direction of heat diffusion, so that the coupling sensitivity of the auxiliary sensor to the chips on both sides is balanced, providing a hardware basis for subsequent linear decoupling. This layout breaks the limitations of traditional single-type sensors and achieves the separation and acquisition of direct waves and coupled waves for the first time.

[0084] Step S2 further includes the following sub-steps: S2-1 monitors the temperature of the chip gap area through an auxiliary sensor and compares it with a preset reference temperature. When the measured temperature difference is less than the preset temperature difference threshold, it is determined that the converter is in standby mode. S2-2, Based on the module thermal time constant and the theoretical thermal wave propagation time difference between the nearest neighbor sensors in the sensor array, the pulse width of the narrow pulse is set to be both less than one-tenth of the module thermal time constant and less than half of the theoretical thermal wave propagation time difference. S2-3, in standby mode, applies a steep-front single-pulse excitation with a rise time of less than 1 microsecond to the IGBT module through the gate drive circuit, and at the same time applies a test current of preset amplitude to the collector.

[0085] It should be noted that the auxiliary sensor refers to the temperature sensor located at the center of the chip gap area, used to monitor the temperature of the chip gap area as a basis for judging the thermal steady state of the converter.

[0086] The chip gap region refers to the projection of the gap region between adjacent chips in a multi-chip IGBT module onto the surface of the housing. This region does not directly correspond to any chip, and its temperature is affected by the combined influence of the adjacent chips.

[0087] The preset reference temperature refers to the base temperature value determined during the module's factory calibration stage. It represents the temperature when the converter is in a fully cooled state and is usually set to the ambient temperature or the radiator temperature.

[0088] The preset temperature difference threshold refers to a pre-set allowable temperature deviation value used to determine the upper limit of the temperature difference for whether the converter is in standby mode. When the difference between the chip gap area temperature and the reference temperature is less than this threshold, the module is considered to have reached thermal steady state. Acquisition rule: During the module's factory calibration phase, this is determined by repeatedly measuring the temperature fluctuation range of the module in a fully cooled state. It is typically taken as 2 to 3 times the ambient temperature variation, for example, 0.5 degrees Celsius to 2.0 degrees Celsius. In this embodiment, 2.0 degrees Celsius is used. Usage method: The chip gap temperature measured by the auxiliary sensor is read in real time, and the absolute value temperature difference is calculated with the pre-set reference temperature (e.g., 25 degrees Celsius). If the temperature difference is less than this threshold, the converter is determined to be in standby mode, allowing pulse excitation; otherwise, the module continues to wait for cooling.

[0089] The inverter standby state refers to the state where the main circuit of the inverter is disconnected, no large current flows through the power module, and the module temperature is basically the same as the ambient temperature. At this time, it is suitable to inject test pulses without affecting normal operation.

[0090] The module thermal time constant is a characteristic parameter that describes the response speed of the IGBT module thermal system to a step thermal excitation. It is equal to the module's thermal resistance multiplied by its thermal capacity, and is measured in seconds. It represents the time required for heat to travel from the chip to the casing.

[0091] Nearest neighbor sensors refer to the two main sensors in a sensor array that are spatially closest to each other, with the smallest distance between them and the smallest time difference in heat wave propagation.

[0092] The theoretical thermal wave propagation time difference refers to the theoretical time difference calculated based on the distance between the nearest neighbor sensors and the equivalent thermal wave velocity of the module packaging material. The calculation formula is distance divided by equivalent thermal wave velocity.

[0093] The pulse width of a narrow pulse refers to the duration of the electrothermal excitation signal, measured in microseconds or milliseconds, and is a key parameter that determines the spectral characteristics of the excitation signal.

[0094] One-tenth of the module thermal time constant refers to the value obtained by multiplying the module thermal time constant by 0.1. This serves as one of the upper limits for the narrow pulse width, ensuring the pulse is narrow enough to excite a heat wave rather than allowing slow heat diffusion. The rule for obtaining this value is: the module thermal time constant equals the junction-to-case thermal resistance multiplied by the thermal capacity, which can be obtained from the datasheet or measured experimentally. One-tenth of this value is used as a constraint. For example, if the thermal time constant is 0.5 seconds, then one-tenth is 50 milliseconds. Usage: When setting the narrow pulse width, the pulse width must be less than this value. This constraint takes precedence over other constraints to ensure the transient characteristics of the excitation.

[0095] Half of the theoretical thermal wave propagation time difference refers to the theoretical thermal wave propagation time difference between the nearest neighbor sensors divided by 2. This serves as another upper limit for the narrow pulse width, ensuring that the responses of adjacent sensors do not overlap in time. The acquisition rule is: measure the spatial distance between the two nearest neighbor master sensors in the sensor array, divide it by the equivalent thermal wave velocity of the module packaging material (calibrated experimentally), obtain the theoretical propagation time difference, and then take half of it. For example, a distance of 40 mm, a velocity of 2000 mm / s, and a time difference of 20 ms would result in a half of 10 ms. Usage: Set the narrow pulse width to be less than both this value and one-tenth of the thermal time constant, and take the smaller of the two as the upper limit.

[0096] Rise time refers to the time required for a pulse signal to rise from 10% to 90% of its amplitude, measured in microseconds, and reflects the steepness of the pulse's leading edge.

[0097] Steep-front single-pulse excitation refers to a single electrical pulse signal with an extremely short rise time. A steep front means that the high-frequency components are abundant, which is conducive to generating a clear thermal wave front.

[0098] The gate drive circuit is a circuit that provides a drive signal to the gate of the IGBT module to control the IGBT's turn-on and turn-off. In this scheme, it is used to apply test pulses.

[0099] The collector refers to the power output terminal of the IGBT module, and the test current is applied between the collector and the emitter.

[0100] The preset amplitude refers to the amplitude of the test current applied between the collector and emitter of the IGBT module during narrow pulse excitation, measured in amperes. This amplitude is a pre-set fixed value used to generate repeatable electrothermal excitation. The acquisition rule is based on the rated current of the IGBT module, typically set to 30% to 50% of the rated current. The selection principle is to generate sufficient thermal wave signal without damaging the module, while avoiding overheating. For example, with a rated current of 200 amperes, the preset amplitude is 60 amperes (30%). Usage: During pulse excitation, while the gate drive circuit applies a steep-fronted single pulse, a test current of the preset amplitude is injected into the collector through a current source. This current value, together with the pulse width, determines the total energy injected into the module (current × voltage × pulse width), ensuring consistent excitation intensity for each measurement.

[0101] In step S2, this application solves the problems of consistency and distinguishability of thermal wave excitation through standby state judgment and narrow pulse parameter optimization. By using an auxiliary sensor to monitor the temperature difference between the chip gap and the reference temperature, it accurately determines whether the converter is in thermal steady state, ensuring consistent starting conditions for each test. The narrow pulse width is constrained by both the module's thermal time constant and the theoretical thermal wave propagation time difference of the nearest neighbor sensor: less than one-tenth of the thermal time constant ensures that the excitation is a true thermal wave rather than thermal diffusion; less than half of the theoretical time difference ensures that the responses of neighboring sensors do not overlap in time. A steep leading edge (rise time less than 1 microsecond) further improves the temporal resolution of the thermal wave leading edge. This series of designs makes the excitation signal both sufficiently "narrow" and sufficiently "steep," creating conditions for accurate measurement of the time difference.

[0102] Step S3 further includes the following sub-steps: S3-1, Wavelet denoising is performed on the temperature time series collected by each main sensor and auxiliary sensor to extract the temperature rise rate curve, and the adaptive threshold method is used to detect the time point when each curve first reaches the peak value, which are recorded as the arrival time of the heat wave of the main sensor and the arrival time of the heat wave of the auxiliary sensor, respectively. S3-2, Based on the geometric distance between the auxiliary sensor and its adjacent main sensor and the thermal diffusion anisotropy coefficient of the module packaging layer, calculate the thermal coupling contribution weight from different adjacent chips in the arrival time of the heat wave measured by each auxiliary sensor. S3-3, based on the thermal coupling contribution weight, decompose the thermal wave arrival time of each auxiliary sensor into coupling components corresponding to different adjacent chips, and subtract these coupling components from the thermal wave arrival time of the corresponding main sensor according to the weight to obtain the pure thermal wave arrival time after decoupling. S3-4 Calculate the time difference of arrival of the pure thermal waves from any two main sensors to form a time difference vector.

[0103] Furthermore, in sub-steps S3-4, constructing the time difference vector specifically includes: Based on the two-dimensional coordinates of each main sensor on the housing plane, calculate the Euclidean distance between every two main sensors, and sort the sensor pairs in ascending order of distance. For sensor pairs that are equidistant from each other, they are sorted in ascending order of the angle between the line connecting the sensor pairs and the preset reference direction. Based on the above sorting results, calculate the arrival time difference of the pure thermal wave of the two main sensors in each sensor pair in turn, and arrange these differences into a vector in order as the time difference vector. Record the sensor pair index corresponding to each difference to form a mapping table, and match the elements of the time difference vector with the sensor pair.

[0104] It should be noted that a temperature time series refers to a data sequence formed by arranging temperature values ​​collected by a sensor at consecutive time points in chronological order, reflecting the process of temperature change over time.

[0105] Wavelet denoising refers to using wavelet transform to decompose a temperature time series, breaking the signal down into different frequency components, removing high-frequency noise components, and then reconstructing the signal to improve the signal-to-noise ratio.

[0106] The temperature rise rate curve is the first derivative curve of temperature with respect to time, that is, the change in temperature per unit time as a function of time, and is used to locate the arrival time of the heat wave front.

[0107] The adaptive threshold method is a peak detection method that dynamically adjusts the detection threshold based on the statistical characteristics of the signal itself. The threshold is a multiple of the root mean square of the signal and can adapt to peak detection under different signal-to-noise ratio conditions.

[0108] The time point at which the first peak value is reached refers to the time corresponding to the first local maximum value reached after the temperature rise rate curve starts to rise from zero. This moment marks the arrival of the thermal wave front at the sensor.

[0109] The arrival time of the heat wave from the main sensor refers to the time elapsed from the start of the narrow pulse excitation to the peak temperature rise rate of the main sensor, representing the propagation time of the direct heat wave from the heat source to the main sensor.

[0110] The arrival time of the thermal wave from the auxiliary sensor refers to the time elapsed from the start of the narrow pulse excitation to the peak temperature rise rate of the auxiliary sensor, representing the propagation time of the thermally coupled signal to the auxiliary sensor.

[0111] Geometric distance refers to the Euclidean distance between the auxiliary sensor and its adjacent main sensor on the plane of the housing, that is, the straight-line distance between the two points.

[0112] The thermal diffusion anisotropy coefficient is a parameter that describes the difference in thermal diffusion capability of a module encapsulation layer in different directions. Its value is equal to the ratio of thermal conductivity in different directions.

[0113] Thermal coupling contribution weight refers to the parameter that quantifies the proportion of the thermal wave arrival time measured by the auxiliary sensor from a specific adjacent chip. The weight value is related to the geometric distance and the anisotropy coefficient.

[0114] The coupling component refers to the portion of the arrival time of the thermal wave from the auxiliary sensor contributed by the thermal effect of a specific adjacent chip. The coupling components of different chips can be superimposed.

[0115] Weighted subtraction refers to subtracting the arrival time of the auxiliary sensor's heat wave multiplied by its corresponding weight from the arrival time of the heat wave from the arrival time of the heat wave from the main sensor, which is used to eliminate thermal coupling interference.

[0116] Pure heat wave arrival time refers to the heat wave arrival time of the main sensor obtained after decoupling processing, which eliminates thermal coupling interference between adjacent chips and only reflects the heat wave propagation time of the corresponding chip itself.

[0117] Euclidean distance refers to the straight-line distance between the two-dimensional coordinates of the sensor on the plane of the housing. The calculation formula is the square root of the sum of the squares of the difference between the horizontal coordinates and the squares of the difference between the vertical coordinates of the two points.

[0118] The preset reference direction refers to a fixed direction that is pre-set on the plane of the housing. It is usually set to the horizontal edge direction of the module and is used to define the angle of the sensor relative to the connecting line.

[0119] Secondary sorting refers to sorting based on a second criterion, namely the angle between the lines, when the distances are the same, to ensure the uniqueness of the sorting result.

[0120] Sequential arrangement refers to placing the time difference values ​​of each sensor pair into a vector in order according to the sorting results, forming an ordered time difference vector.

[0121] A mapping table is a data structure that records the correspondence between the position of each element in the time difference vector and the corresponding sensor pair index, and is used for fast positioning and mapping.

[0122] In step S3, this application solves the problem of extracting the pure heat wave arrival time under noise interference and thermal coupling contamination by employing wavelet denoising, adaptive threshold peak detection, and thermal coupling decoupling. Wavelet denoising effectively filters out electromagnetic and environmental noise, while the adaptive threshold adapts to different signal-to-noise ratios, ensuring the robustness of peak detection. Using the geometric distance and anisotropy coefficient between the auxiliary sensor and its adjacent main sensors, the contribution weight of each adjacent chip to the auxiliary sensor signal is quantitatively calculated. The arrival time of the auxiliary sensor is decomposed into coupling components corresponding to different chips. These coupling components are then weighted and subtracted from the arrival time of the main sensors to obtain the pure time reflecting only the heat wave propagation of the chip itself. This decoupling method avoids the complex calculation of inverting the traditional thermal network matrix, achieving high-precision decoupling with only simple subtraction. Finally, a time difference vector is constructed by sorting the sensor pairs by distance and angle, ensuring the consistency and reproducibility of the feature vectors.

[0123] Step S4 further includes the following sub-steps: S4-1, During the module's factory calibration phase or under known healthy conditions, repeat steps S2 and S3 a preset number of times to obtain the reference time difference vector and the reference peak amplitude vector, and calibrate the relationship curve between thermal wave velocity and temperature. S4-2, monitors the reference temperature at the edge of the module substrate in real time, and performs velocity correction on the measured time difference vector according to the calibrated velocity-temperature relationship curve; S4-3, based on the geometric constraints of the sensor array, perform redundancy verification on the time difference vector after velocity correction and eliminate the elements corresponding to abnormal sensor pairs; S4-4, Subtract the measured time difference vector after removing anomalies from the baseline time difference vector to obtain the time difference offset vector, and calculate its Euclidean norm as the time difference health component. S4-5, compare the measured peak amplitude of the temperature rise rate of each main sensor with the reference peak amplitude, calculate the amplitude attenuation vector, and calculate its magnitude to obtain the amplitude health component. S4-6, the time difference health component and the amplitude health component are weighted and summed according to preset weights to obtain the comprehensive health index.

[0124] It should be noted that the module factory calibration stage refers to the parameter calibration process carried out on the IGBT module after it is manufactured and before it is put into use. During this stage, the module is in a known healthy state and is used to establish baseline data.

[0125] Known health status refers to the state in which the module has been confirmed by factory testing to have no aging damage, and at this time, the module's various performance indicators meet the design specifications.

[0126] The preset number of times refers to the number of times the measurement is repeated in advance, such as 10 or 20 times, which is used to obtain statistical characteristic values ​​through multiple measurements to improve the accuracy of the benchmark.

[0127] The baseline time difference vector refers to the statistical characteristic value of the time difference vector obtained through multiple measurements under the module's health status. It is usually the median or average of each element, which is used as a reference benchmark for subsequent health status judgment.

[0128] The reference peak amplitude vector refers to the statistical characteristic value of the peak amplitude of the temperature rise rate of each main sensor obtained through multiple measurements under the module's healthy state. It is used to compare with the measured amplitude to evaluate the amplitude decay.

[0129] The curve describing the relationship between thermal wave velocity and temperature describes how the thermal wave propagation velocity in the module encapsulation layer changes with temperature. It is obtained through experimental calibration and reflects the temperature dependence of the material's thermal diffusivity.

[0130] The substrate edge reference temperature refers to the temperature value measured by a temperature sensor installed on the edge of the module substrate in an area not covered by chips. It represents the reference temperature of the module and is used to correct the influence of temperature on the thermal wave velocity.

[0131] The velocity-temperature relationship curve refers to the functional relationship curve between thermal wave velocity and temperature. It is usually a linear or polynomial fitting curve, determined by calibration experimental data.

[0132] Velocity correction refers to multiplying each element in the measured time difference vector by a correction coefficient based on the current measured temperature and the calibrated velocity-temperature relationship curve, thereby eliminating the influence of temperature changes on the propagation speed of heat waves.

[0133] Geometric constraints refer to mathematical constraints determined by the spatial relationship between the sensors in a sensor array, such as the geometric identity that the sum of the time differences between any two of the three sensors should be zero.

[0134] Redundancy verification refers to using the geometric redundancy characteristics of sensor arrays to check the consistency of measurement data, detect and mark abnormal measurement values ​​that exceed the allowable deviation of geometric constraints.

[0135] An abnormal sensor pair refers to a sensor pair whose time difference measurement value violates geometric constraints or deviates from the statistical distribution by more than a preset threshold, which may be caused by sensor failure or local interference.

[0136] The time difference offset vector is the difference vector between the measured time difference vector and the reference time difference vector, reflecting the change in the module's current heat propagation characteristics relative to its health state.

[0137] The Euclidean norm is the square root of the sum of the squares of the components of a vector. It is used to quantify the overall size of the time difference offset vector. The larger the norm, the more serious the deviation from the healthy state.

[0138] The time difference health component is a single value calculated from the Euclidean norm of the time difference offset vector, reflecting the health of the module in the time dimension of heat wave propagation.

[0139] The peak amplitude of the temperature rise rate refers to the local maximum value of the sensor's temperature rise rate curve, that is, the maximum rise rate of the heat wave front, and its magnitude is related to the energy attenuation of the heat wave.

[0140] The amplitude attenuation vector is the ratio vector of the measured peak amplitude to the reference peak amplitude, reflecting the degree of attenuation of the thermal wave amplitude at each sensor location.

[0141] The magnitude refers to the Euclidean norm of a vector, which is used to quantify the overall size of the amplitude decay vector. The larger the magnitude, the more severe the amplitude decay.

[0142] The amplitude health component is a single value calculated from the magnitude of the amplitude attenuation vector, reflecting the health of the module in the thermal wave amplitude dimension.

[0143] The preset weights refer to the proportional coefficients used to balance the contributions of the time difference health component and the amplitude health component in the calculation of the comprehensive health index. The acquisition rules are determined based on historical failure mode analysis of the module or the application scenario. If the solder fatigue risk is high, the weight of the time difference component is increased; if the bond wire aging risk is high, the weight of the amplitude component is increased. The sum of the two weights is 1. For example, when the bond wire aging risk is high, the time difference weight is 0.4 and the amplitude weight is 0.6. Usage: Multiply each health component by its corresponding weight and then add them together to obtain the comprehensive health index.

[0144] Weighted summation refers to the calculation method of multiplying two health components by their respective preset weights and then adding them together to obtain the comprehensive health index.

[0145] The comprehensive health index is a single quantitative indicator that integrates time difference information and amplitude information to comprehensively reflect the overall health status of the module.

[0146] In step S4, this application addresses the issues of environmental changes, sensor failures, and insufficient single-feature information through benchmark calibration, velocity correction, redundancy verification, and dual-feature fusion. Factory calibration establishes a benchmark time difference vector and a benchmark peak amplitude vector, eliminating individual differences. Real-time temperature correction eliminates the influence of heat wave velocity variations with temperature, making measurements comparable under different operating conditions. Geometric redundancy verification automatically identifies and eliminates abnormal sensor pairs using triangle closure conditions, improving the system's fault tolerance. The time difference health component reflects changes in heat wave propagation time and is sensitive to solder layer fatigue; the amplitude health component reflects the attenuation of heat wave amplitude and is sensitive to bond line aging. The weighted fusion of these two components achieves complementary perception of different failure modes, providing a more comprehensive and robust approach than using a single feature.

[0147] Step S5 further includes the following sub-steps: S5-1, Normalize the time difference offset vector obtained in step S4 to obtain a unit direction vector. Each component of the unit direction vector represents the time difference offset trend of the corresponding sensor in the direction of the connection. S5-2, based on the geometric arrangement of each main sensor in the sensor array, establishes an offline mapping matrix from the direction vector space to the shell plane coordinate system during the module factory calibration stage; S5-3, multiply the unit direction vector by the mapping matrix to obtain the projected coordinates of the aging point on the shell plane; S5-4 Match the projected coordinates with the projection area of ​​the shell of each chip inside the module. If the projected coordinates fall within the projection area of ​​a chip, the chip is identified as an aging chip. If the projected coordinates fall within the chip gap area, the adjacent chips that make up the gap are all marked as suspected aging chips.

[0148] Furthermore, in sub-step S5-3, the projection coordinates of the aging points on the shell plane are obtained using a multi-combination voting fusion method instead of matrix multiplication, specifically including: Select all possible combinations of three non-collinear sensors from the sensor array; For each subarray combination, the time difference offset vector is normalized to obtain the unit direction vector, and then multiplied with the mapping matrix of sub-step S5-2 to independently calculate a candidate projection coordinate. Cluster analysis is performed on all candidate projection coordinates to identify the cluster containing the most candidate coordinates, and the center coordinates of the cluster are used as the final projection coordinates.

[0149] Furthermore, after determining the correspondence between the projected coordinates and the chip in step S5, a location reliability assessment is also performed: Calculate the Euclidean distance between the projected coordinates and the center of the projected area of ​​the identified aging chip or suspected aging chip, and divide the distance by the radius of the circumcircle of the chip's projected area to obtain the normalized positioning deviation. The location reliability is calculated based on the normalized positioning error. The location reliability is equal to 1 minus the normalized positioning error. When the positioning confidence level is lower than the preset lower threshold, an unreliable positioning flag is output, along with the projected coordinates and the corresponding chip information.

[0150] It should be noted that the time difference offset vector is the difference vector between the measured time difference vector and the reference time difference vector, which reflects the change in the current heat propagation characteristics of the module relative to its healthy state, and its direction is related to the spatial location of the aging point.

[0151] Normalization refers to dividing each component of a vector by the vector's magnitude, making the vector's magnitude 1, thus preserving direction information while eliminating amplitude information.

[0152] The unit direction vector is a vector with a magnitude of 1 obtained after normalization. The sign and relative magnitude of each component reflect the distribution pattern of time difference offset among different sensor pairs.

[0153] The corresponding sensor pair connection direction refers to the spatial connection direction between the two main sensors corresponding to each component in the time difference offset vector. Different components correspond to different spatial directions.

[0154] The time difference offset trend refers to the distribution pattern of the time difference offset in different spatial directions reflected by the sign and magnitude of each component of the unit direction vector, which is related to the position of the aging point relative to the sensor array.

[0155] The direction vector space refers to a multi-dimensional space composed of the direction information of the time difference offset vector. Its dimension is equal to the number of sensor pairs, and each dimension corresponds to the direction of the connection between sensor pairs.

[0156] The mapping matrix is ​​a linear transformation matrix from the direction vector space to the shell plane coordinate system. It is established in advance through simulation or experimental calibration, realizing the transformation from direction information to spatial coordinates.

[0157] The shell plane coordinate system refers to a two-dimensional rectangular coordinate system established with the module shell plane as the coordinate plane, which is used to describe the projection position of the aging point on the shell.

[0158] Projected coordinates refer to the position coordinates of the aging point projected perpendicularly onto the shell plane along the heat propagation path, expressed in millimeters.

[0159] The housing projection area refers to the area covered by the vertical projection of the chip onto the housing plane, and its shape is a rectangle or square with the same size as the chip.

[0160] An aged chip is a chip whose projected coordinates fall within the projection area of ​​its casing; such chips are considered to have aged.

[0161] The chip gap region refers to the empty area between the projection areas of adjacent chips. When the projection coordinates fall in this area, the aging chip cannot be uniquely identified.

[0162] Suspected aging chips refer to adjacent chips that form the gap when the projected coordinates fall in the chip gap area, which need further confirmation or can be considered as candidate outputs.

[0163] Non-collinear sensors are three sensors that are not on the same straight line, which can form a triangle to determine their position in a plane.

[0164] A subarray combination refers to a subset of three non-collinear sensors selected from a sensor array. Each subarray can independently calculate a candidate projection coordinate.

[0165] Candidate projection coordinates refer to the projection coordinates calculated independently by combining individual subarrays; different subarrays may yield different results.

[0166] Cluster analysis refers to grouping multiple candidate projected coordinates according to spatial distance, with coordinates that are close to each other belonging to the same cluster.

[0167] A cluster is a set of candidate projected coordinates that are spatially close in cluster analysis. The cluster containing the most coordinates represents the consensus result of the majority of subarrays.

[0168] The center coordinates of a cluster are the coordinates obtained by averaging the coordinates of all candidate projected coordinates within the cluster, and are used as the final projected coordinates.

[0169] Euclidean distance refers to the straight-line distance between the projected coordinates and the center coordinates of the chip's projection area. The calculation formula is the square root of the sum of the squares of the differences in the horizontal and vertical coordinates of the two points.

[0170] The circumcircle radius refers to the radius of the smallest circumcircle that can cover the chip's projected area. For a square chip, it is equal to half the length of the diagonal.

[0171] Normalized positioning deviation is the ratio of Euclidean distance to the radius of the circumscribed circle, ranging from 0 to 1, reflecting the size of the positioning deviation relative to the chip size.

[0172] Location reliability is an index that quantifies the reliability of location results. It is equal to 1 minus the normalized location error, and its value ranges from 0 to 1. The higher the value, the more reliable the location.

[0173] The preset lower threshold is the minimum confidence level for determining the reliability of the positioning result. If the value is below this threshold, an unreliable positioning flag is output. The acquisition rule is set according to the application's requirements for positioning accuracy. A larger value (e.g., 0.7) is used when high accuracy is required, and a smaller value (e.g., 0.5) is used when a certain error is acceptable. In this implementation, 0.6 is used. The usage method is as follows: Calculate the positioning confidence level (1 minus the normalized positioning deviation). If it is less than this threshold, an unreliable flag and multiple suspected chips are output.

[0174] The location unreliability flag is a flag signal output when the location reliability is lower than a preset lower threshold, indicating to the user that the current location result has low reliability.

[0175] In step S5, this application addresses the issues of positioning accuracy, robustness, and reliability through direction vector mapping, multi-combination voting fusion, and confidence assessment. The time difference offset vector is normalized to a unit direction vector, retaining only direction information and eliminating the impact of aging on positioning. The offline-established mapping matrix achieves a linear transformation from the direction vector space to the shell plane coordinate system; online positioning requires only one matrix multiplication, resulting in extremely high computational efficiency. The multi-combination voting fusion method independently calculates candidate projected coordinates using all three possible sensor subarrays, then selects the largest cluster center through clustering. Even if some sensor data is abnormal, it will not corrupt the final result, greatly enhancing the system's fault tolerance. The confidence assessment provides a quantitative reliability index based on the normalized deviation between the projected coordinates and the chip center. When the confidence level is below a threshold, multiple suspected chips are output, avoiding blindly trusting unreliable results.

[0176] Step S6 further includes the following sub-steps: S6-1, Based on the comprehensive health index determined in step S4, compare it with the preset first health threshold and second health threshold. The first health threshold is less than the second health threshold. When the comprehensive health index is less than the first health threshold, it is determined to be a healthy level. When it is greater than or equal to the first health threshold and less than the second health threshold, it is determined to be a warning level. When it is greater than or equal to the second health threshold, it is determined to be a failure level. S6-2, when the health level is determined to be "attention", output encrypted monitoring suggestions and shorten the monitoring interval to half of the original interval; S6-3, when the health level is determined to be failure, and the location of the aging chip determined in step S5 is consistent with the location of the sensor with the highest temperature rise rate in step S2, output a failure alarm and suggest replacing the chip; when the two locations are inconsistent, output a failure alarm and suggest a comprehensive inspection. S6-4 outputs the health level, aging chip location, and positioning confidence through the communication interface, and visualizes the temperature distribution on the shell surface and the aging chip location in the form of a temperature cloud map.

[0177] It should be noted that the comprehensive health index is a single quantitative indicator obtained by weighting and summing the time difference health component and the amplitude health component according to preset weights, which comprehensively reflects the overall health status of the module.

[0178] The preset first and second health thresholds are two boundary values ​​used to classify health levels, with the first health threshold being less than the second health threshold. A comprehensive health level below the first threshold indicates a healthy level, between these two levels indicates a warning level, and greater than or equal to the second threshold indicates a failure level. The acquisition rule is determined through statistical analysis of a large amount of measured data from healthy and aging modules. The comprehensive health level of a healthy module is typically below 0.5, while that of a severely aging module is above 1.0. In this implementation, the first threshold is set to 0.5, and the second threshold to 1.0. The usage method is to compare the calculated comprehensive health index with the two thresholds and output the corresponding health level.

[0179] The health level refers to the judgment result when the comprehensive health index is less than the first threshold, indicating that the module is in normal operation, all performance indicators are normal, and no maintenance measures are required.

[0180] The "attention level" refers to the judgment result when the comprehensive health index is greater than or equal to the first threshold and less than the second threshold. It indicates that the module has experienced slight aging or performance degradation, requiring enhanced monitoring but not immediate shutdown.

[0181] The failure level refers to the judgment result when the comprehensive health index is greater than or equal to the second threshold, indicating that the module is severely aged or has failed and needs to be shut down immediately.

[0182] Encrypted monitoring recommendations are maintenance suggestions output when the health level is at the "attention" level, prompting users to increase the monitoring frequency to track the aging trend.

[0183] The monitoring interval refers to the time interval between two online health status monitoring sessions. The interval is longer under normal conditions and shortened to half the original interval under the attention status.

[0184] Half the original interval means that when the health level is "attention," the normal monitoring interval is shortened to half of the original time for encrypted monitoring. Acquisition rules: determined by the system's preset normal monitoring interval. The normal interval can be set according to the module's usage frequency (e.g., 30 days), then it is shortened to 15 days under the "attention" status. Usage: When the health level is determined to be "attention," the time of the next monitoring is automatically adjusted to the current time plus half of the original interval.

[0185] The sensor with the highest temperature rise rate is the one with the largest peak temperature rise rate among all sensors after the narrow pulse excitation in step S2. The location of this sensor corresponds to the region where the thermal effect is most concentrated.

[0186] A communication interface refers to a data transmission interface used to output monitoring results to a host computer or monitoring system. It can be a CAN bus, RS485, Ethernet, or a wireless communication module.

[0187] A temperature cloud map is a color distribution map generated by interpolating the temperature measurements from various sensors on the surface of the housing. Different colors represent different temperature ranges and are used to visually display the temperature distribution of the module.

[0188] Visualization refers to displaying information such as health level, location of aging chips, location reliability, and temperature cloud map on a human-machine interface in a graphical way, so that maintenance personnel can quickly understand the health status of the module.

[0189] In step S6, this application addresses the issues of effective transmission of health information and decision support through hierarchical output, cross-validation, and visualization. Based on a comparison of the comprehensive health index with preset thresholds, the module status is divided into three levels: healthy, alert, and failed, corresponding to normal, encrypted monitoring, and immediate handling strategies, respectively. When a failure is detected, the location of the aging chip is further cross-validated with the location of the sensor with the highest temperature rise rate. If they match, the chip is directly located and replaced; otherwise, a comprehensive inspection is recommended, improving the reliability of the alarm. The temperature cloud map visually displays the temperature distribution of the casing and the location of the aging chip, enabling maintenance personnel to quickly locate the fault point. The entire output process transforms complex monitoring data into actionable maintenance instructions, achieving a closed loop from monitoring to decision-making.

[0190] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature, characterized in that, Includes the following steps: Step S1: Arrange multiple non-collinearly distributed transient temperature sensors on the housing surface of the multi-chip IGBT module to form a sensor array. Step S2: During the standby state, a narrow pulse electrothermal excitation signal is injected into the module to cause heat to propagate inside the module in the form of thermal waves, so that the responses of each sensor are distinguishable in time. Step S3: Record the time when the temperature rise rate of each sensor first reaches its peak after narrow pulse excitation, and calculate the time difference of the arrival of the heat wave between any two sensors to form a time difference vector. Step S4: Compare the time difference vector with the pre-calibrated baseline time difference vector under healthy conditions, calculate the time difference offset vector, and determine the overall health of the module based on the magnitude of the time difference offset vector. Step S5: Calculate the projection position of the aging point on the module housing plane based on the direction of the time difference offset vector and the geometric arrangement of the sensor array, and locate the chip area where aging has occurred. Step S6: Output the health level based on the comparison result between the health level and the preset health level threshold, and output the location information of the aging chip at the same time.

2. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 1, characterized in that: Step S1 further includes the following sub-steps: S1-1, multiple transient temperature sensors are divided into a main sensor group and an auxiliary sensor group. Each sensor in the main sensor group is arranged in a position directly opposite to the surface of the housing directly below each chip, and all main sensors are non-collinearly distributed on the housing plane. Each sensor in the auxiliary sensor group is arranged in the middle position between the chips. The main sensor group and the auxiliary sensor group together constitute a non-collinearly distributed sensor array. S1-2, Set the minimum spacing between adjacent main sensors to be greater than the maximum value of the minimum outer circle diameter of the two adjacent chips projected on the surface of the housing, and the maximum spacing to be less than half the distance from the sensor to the boundary of the module housing; S1-3, Set the number of main sensors to be the same as the number of chips inside the module, with each main sensor corresponding to one chip; S1-4, an auxiliary sensor is set at the center of each chip gap area, and the angle between the auxiliary sensor and the line connecting the two adjacent main sensors is 90 degrees. The bisector of this angle is consistent with the main direction of thermal diffusion of the module packaging layer.

3. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 1, characterized in that: Step S2 further includes the following sub-steps: S2-1 monitors the temperature of the chip gap area through an auxiliary sensor and compares it with a preset reference temperature. When the measured temperature difference is less than the preset temperature difference threshold, it is determined that the converter is in standby mode. S2-2, Based on the module thermal time constant and the theoretical thermal wave propagation time difference between the nearest neighbor sensors in the sensor array, the pulse width of the narrow pulse is set to be both less than one-tenth of the module thermal time constant and less than half of the theoretical thermal wave propagation time difference. S2-3, in standby mode, applies a steep-front single-pulse excitation with a rise time of less than 1 microsecond to the IGBT module through the gate drive circuit, and at the same time applies a test current of preset amplitude to the collector.

4. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 1, characterized in that: Step S3 further includes the following sub-steps: S3-1, Wavelet denoising is performed on the temperature time series collected by each main sensor and auxiliary sensor to extract the temperature rise rate curve, and the adaptive threshold method is used to detect the time point when each curve first reaches the peak value, which are recorded as the arrival time of the heat wave of the main sensor and the arrival time of the heat wave of the auxiliary sensor, respectively. S3-2, Based on the geometric distance between the auxiliary sensor and its adjacent main sensor and the thermal diffusion anisotropy coefficient of the module packaging layer, calculate the thermal coupling contribution weight from different adjacent chips in the arrival time of the heat wave measured by each auxiliary sensor. S3-3, based on the thermal coupling contribution weight, decompose the thermal wave arrival time of each auxiliary sensor into coupling components corresponding to different adjacent chips, and subtract these coupling components from the thermal wave arrival time of the corresponding main sensor according to the weight to obtain the pure thermal wave arrival time after decoupling. S3-4 Calculate the time difference of arrival of the pure thermal waves from any two main sensors to form a time difference vector.

5. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 4, characterized in that, In sub-steps S3-4, the construction of the time difference vector specifically includes: Based on the two-dimensional coordinates of each main sensor on the housing plane, calculate the Euclidean distance between every two main sensors, and sort the sensor pairs in ascending order of distance. For sensor pairs that are equidistant from each other, they are sorted in ascending order of the angle between the line connecting the sensor pairs and the preset reference direction. Based on the above sorting results, calculate the arrival time difference of the pure thermal wave of the two main sensors in each sensor pair in turn, and arrange these differences into a vector in order as the time difference vector. Record the sensor pair index corresponding to each difference to form a mapping table, and associate the elements of the time difference vector with the sensor pair.

6. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 1, characterized in that: Step S4 further includes the following sub-steps: S4-1, During the module's factory calibration phase or under known healthy conditions, repeat steps S2 and S3 a preset number of times to obtain the reference time difference vector and the reference peak amplitude vector, and calibrate the relationship curve between thermal wave velocity and temperature. S4-2, monitors the reference temperature at the edge of the module substrate in real time, and performs velocity correction on the measured time difference vector according to the calibrated velocity-temperature relationship curve; S4-3, based on the geometric constraints of the sensor array, perform redundancy verification on the time difference vector after velocity correction and eliminate the elements corresponding to abnormal sensor pairs; S4-4, Subtract the measured time difference vector after removing anomalies from the baseline time difference vector to obtain the time difference offset vector, and calculate its Euclidean norm as the time difference health component. S4-5, compare the measured peak amplitude of the temperature rise rate of each main sensor with the reference peak amplitude, calculate the amplitude attenuation vector, and calculate its magnitude to obtain the amplitude health component. S4-6, the time difference health component and the amplitude health component are weighted and summed according to preset weights to obtain the comprehensive health index.

7. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 1, characterized in that: Step S5 further includes the following sub-steps: S5-1, Normalize the time difference offset vector obtained in step S4 to obtain a unit direction vector. Each component of the unit direction vector represents the time difference offset trend of the corresponding sensor in the direction of the connection. S5-2, based on the geometric arrangement of each main sensor in the sensor array, establishes an offline mapping matrix from the direction vector space to the shell plane coordinate system during the module factory calibration stage; S5-3, multiply the unit direction vector by the mapping matrix to obtain the projected coordinates of the aging point on the shell plane; S5-4 Match the projected coordinates with the projection area of ​​the shell of each chip inside the module. If the projected coordinates fall within the projection area of ​​a chip, the chip is identified as an aging chip. If the projected coordinates fall within the chip gap area, the adjacent chips that make up the gap are all marked as suspected aging chips.

8. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 7, characterized in that, In sub-step S5-3, the projection coordinates of the aging points on the shell plane are obtained by replacing matrix multiplication with a multi-combination voting fusion method, specifically including: Select all possible combinations of three non-collinear sensors from the sensor array; For each subarray combination, the time difference offset vector is normalized to obtain the unit direction vector, and then multiplied with the mapping matrix of sub-step S5-2 to independently calculate a candidate projection coordinate. Cluster analysis is performed on all candidate projection coordinates to identify the cluster containing the most candidate coordinates, and the center coordinates of the cluster are used as the final projection coordinates.

9. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 7, characterized in that, After determining the correspondence between the projected coordinates and the chip in step S5, a positional reliability assessment is also performed: Calculate the Euclidean distance between the projected coordinates and the center of the projected area of ​​the identified aged chip or suspected aged chip, and divide the distance by the radius of the circumcircle of the chip's projected area to obtain the normalized positioning deviation. The location reliability is calculated based on the normalized positioning error. The location reliability is equal to 1 minus the normalized positioning error. When the positioning confidence level is lower than the preset lower threshold, an unreliable positioning flag is output, along with the projected coordinates and the corresponding chip information.

10. The method for monitoring the health status of a multi-chip IGBT module based on changes in casing temperature according to claim 1, characterized in that: Step S6 further includes the following sub-steps: S6-1, Based on the comprehensive health index determined in step S4, compare it with the preset first health threshold and second health threshold. The first health threshold is less than the second health threshold. When the comprehensive health index is less than the first health threshold, it is determined to be a healthy level. When it is greater than or equal to the first health threshold and less than the second health threshold, it is determined to be a warning level. When it is greater than or equal to the second health threshold, it is determined to be a failure level. S6-2, when the health level is determined to be "attention", output encrypted monitoring suggestions and shorten the monitoring interval to half of the original interval; S6-3, when the health level is determined to be failure, and the location of the aging chip determined in step S5 is consistent with the location of the sensor with the highest temperature rise rate in step S2, output a failure alarm and suggest replacing the chip; when the two locations are inconsistent, output a failure alarm and suggest a comprehensive inspection. S6-4 outputs the health level, aging chip location, and positioning confidence through the communication interface, and visualizes the temperature distribution on the shell surface and the aging chip location in the form of a temperature cloud map.

Images