A method and system for early warning of drift-mapped shielded gate MOS

Abstract

This invention relates to the field of power semiconductor device health management technology, and discloses a method and system for early warning of a shielded gate MOS based on drift mapping. Addressing the problems of existing technologies lacking operating condition awareness and relying on fixed thresholds, this invention includes: responding to the validity of a steady-state flag bit in response to operating conditions; collecting electrical parameters and generating an identifier code indexing expected parameter value group based on the operating conditions; constructing a normalized drift fingerprint vector and calculating the drift rate and overall confidence level; obtaining the group health level by indexing a group prior mapping table with the drift fingerprint vector; calculating the individual health level based on the drift fingerprint vector and individual offset parameters; outputting a fused health level by adding 1 to the product of the overall confidence level and the individual health level and subtracting the product of the overall confidence level and the group health level; updating the individual offset parameters and expected parameter values ​​respectively in response to deviation rate exceeding limits; and outputting an early warning level based on the larger value between the drift amount and the rate trigger level. This invention achieves adaptive operating condition assessment and dual-track feedback correction.

Description

Technical Field

[0001] This invention relates to the field of power semiconductor device health management and predictive maintenance technology, specifically to a drift-mapped shielded gate MOS early warning method and system for real-time monitoring, health assessment and graded early warning of electrical parameter drift in shielded gate MOS devices. Background Technology

[0002] Shielded-gate MOSFETs (MOSFETs) are advanced power semiconductor devices widely used in power electronics systems, electric drive systems for new energy vehicles, industrial frequency converters, and photovoltaic inverters due to their low on-resistance and excellent switching characteristics. During long-term operation, MOSFETs are affected by factors such as thermal stress, electrical stress, and mechanical stress, causing their key electrical parameters (such as threshold voltage, on-resistance, and drain-source current) to gradually drift away from their initial values. When this drift accumulates to a certain extent, it will lead to performance degradation or even failure of the device. Therefore, effective monitoring and early warning of electrical parameter drift in MOSFETs is a critical technical requirement for ensuring reliable system operation.

[0003] Existing power device health status early warning methods suffer from the following technical shortcomings. First, they lack operational condition awareness. Traditional methods use fixed reference standards to determine parameter anomalies, failing to distinguish between normal parameter changes and abnormal drifts under different operating conditions (such as different load currents and ambient temperatures), easily leading to false alarms or missed alarms. Second, relying on fixed thresholds makes it difficult to adapt to individual device differences and long-term parameter drift patterns. Devices in the same batch may have initial parameter differences due to manufacturing process fluctuations, and device parameters will drift collectively over time; fixed thresholds cannot track these changes. Third, they lack the incorporation of group prior knowledge. Existing methods rely solely on historical data from individual devices for evaluation, failing to leverage statistical patterns within the same batch to improve accuracy. Finally, the feedback correction mechanism is simplistic. Even with parameter adaptive mechanisms, existing methods typically only have one correction path, unable to simultaneously handle two different types of deviations: mapping errors and early warning errors.

[0004] Therefore, there is an urgent need for a method and system for early warning of parameter drift in shielded gate MOS devices that can achieve adaptive operation, integrate group priors and individual characteristics, and have dual-track feedback correction capabilities. Summary of the Invention

[0005] To address the problems of insufficient working condition perception, difficulty in adapting fixed thresholds, lack of group prior fusion, and single feedback correction in existing technologies, this invention provides a method and system for early warning of a shielded gate MOS based on drift mapping.

[0006] This invention provides a drift-mapping-based early warning method for shielded gate MOS devices, comprising the following steps: responding to the validity of a steady-state determination flag, acquiring electrical parameters of the shielded gate MOS device through a data acquisition interface, and generating a condition identification code based on a combination of load current range and ambient temperature range; using the condition identification code to index a condition-parameter expected value mapping table to obtain a group of expected parameter values, constructing a normalized drift fingerprint vector based on the electrical parameters and the expected parameter value group, and calculating the drift rate and overall confidence level; using the drift fingerprint vector to index a preset population prior mapping table to obtain the population health level, and based on the drift fingerprint vector... The individual health level is calculated using the texture vector and individual offset parameters. The fused health level is obtained by adding 1 to the product of the comprehensive confidence level and the individual health level, and subtracting the product of the comprehensive confidence level and the group health level. The learning rate is adjusted and the individual offset parameters are updated in response to the mapping deviation rate exceeding the limit trigger signal. The expected parameter values ​​of the working condition-parameter expected value mapping table are corrected in response to the warning deviation rate exceeding the limit trigger signal. The drift amount trigger level is obtained by comparing the fused health level with a preset threshold, and the rate trigger level is obtained by comparing the drift rate with a preset threshold. The larger of the two values ​​is taken as the final warning level.

[0007] Furthermore, the electrical parameters include threshold voltage, on-resistance, and drain-source current; the load current range includes a light load range, a rated load range, and a heavy load range; and the ambient temperature range includes a low temperature range, a normal temperature range, and a high temperature range.

[0008] Furthermore, the overall confidence level is equal to the product of the data accumulation confidence level and the stability confidence level. The data accumulation confidence level is equal to the ratio of the cumulative number of samples to the preset saturation threshold and has an upper limit of 1. The stability confidence level is calculated by multiplying the preset adjustment coefficient by the mean of the drift variance within the preset statistical window, adding 1 as the denominator, and using 1 as the numerator to find the quotient.

[0009] Furthermore, the mapping deviation rate is equal to the absolute value of the difference between the actual change and the predicted change in the fused health level, divided by the larger of the absolute value of the predicted change and the preset minimum value for preventing zero. The predicted change is equal to the difference between the fused health level of the previous period and the fused health level of the previous two periods. The early warning deviation rate is equal to the absolute value of the difference between the final early warning level of the previous period and the final early warning level of the previous two periods, divided by the larger of the absolute value of the final early warning level of the previous two periods and the preset minimum value for preventing zero.

[0010] Furthermore, the learning rate remains unchanged when the mapping deviation rate is lower than a preset first threshold, increases by a preset percentage when the mapping deviation rate is between a preset first threshold and a preset second threshold, and decreases by a preset percentage when the mapping deviation rate is higher than a preset second threshold; the expected parameter value remains unchanged when the warning deviation rate is lower than a preset first deviation threshold, is adjusted by a preset percentage step when the warning deviation rate is between a preset first deviation threshold and a preset second deviation threshold, and is updated by overwriting the measured parameter value of the current sampling period when the warning deviation rate is higher than a preset second deviation threshold.

[0011] Furthermore, both the individual offset parameter update and the expected parameter value correction are set with a preset lag interval and a preset cooling time.

[0012] This invention also provides a drift-mapping shielded gate MOS early warning system, comprising: a condition-sensing electrical parameter sampling module, configured to collect electrical parameters of the shielded gate MOS device when a steady-state determination flag is valid, and generate a condition identification code based on a combination of a load current range and an ambient temperature range; a confidence-fusion drift fingerprint construction module, configured to index a condition-parameter expected value mapping table using the condition identification code to obtain a group of expected parameter values, construct a normalized drift fingerprint vector based on the electrical parameters and the expected parameter value group, and calculate the drift rate and overall confidence; and a drift mapping early warning arbitration module, configured to index a preset group prior mapping table using the drift fingerprint vector to obtain group health. The system calculates an individual health level based on the drift fingerprint vector and individual offset parameters. A fused health level is obtained by adding 1 to the product of the overall confidence level and the individual health level, and subtracting the product of the overall confidence level and the group health level. A drift trigger level is obtained by comparing the fused health level with a preset threshold, and a rate trigger level is obtained by comparing the drift rate with a preset threshold. The larger of the two values ​​is output as the final warning level. A dual-track mapping feedback correction module is configured to adjust the learning rate and update the individual offset parameters in response to a mapping deviation rate exceeding the limit trigger signal, and to correct the expected parameter values ​​in the operating condition-parameter expected value mapping table in response to a warning deviation rate exceeding the limit trigger signal.

[0013] Furthermore, the system is also configured to implement the electrical parameter types, operating condition interval division, comprehensive confidence calculation, deviation rate calculation, learning rate adjustment strategy, expected parameter value correction strategy, and lag interval and cooling time settings in the above method.

[0014] The present invention has the following beneficial effects: First, it achieves adaptive parameter expectation value matching under operating conditions. By combining the load current range and the ambient temperature range to generate an operating condition identification code, and using this code to index the operating condition-parameter expectation value mapping table to obtain the expected parameter value under the corresponding operating condition, it avoids false alarms caused by changes in operating conditions and improves the accuracy of early warning.

[0015] Second, it integrates group prior knowledge with individual historical characteristics. By establishing a group prior mapping table, it introduces the statistical regularities of devices in the same batch, while retaining individual offset parameters to record individual device characteristics. Using comprehensive confidence as the weight, it integrates the group health level and the individual health level, so that the evaluation results have both group reference value and individual relevance.

[0016] Third, the confidence-weighted fusion mechanism ensures that the evaluation results gradually stabilize as data accumulates. The overall confidence level is obtained by multiplying the data accumulation confidence level and the stability confidence level. In the initial operation stage of the device, the focus is on the group's prior knowledge, and as data accumulates, the weight of individual characteristics is gradually increased to achieve a smooth transition.

[0017] Fourth, dual-track feedback correction enables independent optimization of mapping parameters and early warning parameters. Independent correction paths are designed for mapping deviation rate and early warning deviation rate respectively. When the mapping deviation rate exceeds the limit, the learning rate is adjusted and the individual offset parameters are updated. When the early warning deviation rate exceeds the limit, the expected parameter value is corrected. The two paths do not interfere with each other and each performs its own function.

[0018] Fifth, the dual-trigger arbitration strategy takes into account both the cumulative drift amount and the mutation rate as risk dimensions. It simultaneously evaluates the drift amount trigger level corresponding to the fusion health level and the rate trigger level corresponding to the drift rate, and takes the larger value as the final warning level. This can both capture the risk of gradual parameter degradation and respond to sudden parameter anomalies. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of a drift-mapped shielded gate MOS early warning method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the principle of drift fingerprint vector construction and confidence calculation provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the principle of group-individual health level fusion provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the principle of the dual-track mapping feedback correction mechanism provided in an embodiment of the present invention. Figure 5 This is a module structure diagram of a drift-mapped shielded gate MOS early warning system provided in an embodiment of the present invention.

[0020] Explanation of reference numerals in the attached figures S1 - Sampling steps for electrical parameters of operating condition sensing; S2 - Confidence fusion and drift fingerprint construction steps; S3 - Drift mapping early warning arbitration steps; S4 - Dual-track mapping feedback correction steps; S5 - Dual-trigger early warning level output steps; 10-Operating condition sensing electrical parameter sampling module; 20-Confidence fusion drift fingerprint construction module; 21-Operating condition index unit; 22-Drift fingerprint construction unit; 23-Confidence calculation unit; 30-Drift mapping early warning arbitration module; 31-Group health level assessment unit; 32-Individual health level assessment unit; 33-Confidence fusion unit; 34-Dual trigger arbitration unit; 40-Dual track mapping feedback correction module; 41-Mapping deviation correction unit; 42-Early warning deviation correction unit; 101-Working Condition-Parameter Expected Value Mapping Table; 102-Population Prior Mapping Table; 103-Individual Offset Parameter Storage Area. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of protection of the present invention.

[0022] Example 1: See Figure 1 This embodiment provides a drift-mapped early warning method for shielded gate MOS devices. This method is used for real-time monitoring, health assessment, and graded early warning of electrical parameter drift in shielded gate MOS devices. This embodiment focuses on describing the underlying processing logic of the technical solution and does not limit specific business application scenarios.

[0023] In step S1, operating condition sensing electrical parameter sampling is performed. Specifically, the processor first checks the status of the steady-state determination flag. When the flag is valid, it indicates that the shielded gate MOS device is in a stable operating state. At this time, the electrical parameters of the shielded gate MOS device are collected through a pre-configured data acquisition interface. The electrical parameters include key parameters characterizing the device's performance status, such as threshold voltage, on-resistance, and drain-source current. These parameters are periodically collected and stored in the processor's data buffer through an analog-to-digital conversion interface. Simultaneously with acquiring the electrical parameters, the processor performs interval division determination based on the current load current value and ambient temperature value. The load current interval is divided according to a preset current threshold, and the ambient temperature interval is divided according to a preset temperature threshold. The processor combines the load current interval code and the ambient temperature interval code to generate a unique operating condition identifier. This operating condition identifier uses a multi-bit encoding format, where the high-order bits represent the load current interval level and the low-order bits represent the ambient temperature interval level. Through the above-mentioned working condition identification code generation mechanism, the expected values ​​of parameters under different working conditions are distinguished and indexed, providing a benchmark reference for subsequent parameter drift assessment based on working condition perception.

[0024] See Figure 2In step S2, confidence fusion drift fingerprint construction is performed. Specifically, the processor uses the operating condition identifier code generated in step S1 as the index key to query the pre-stored operating condition-parameter expected value mapping table 101, and obtains the expected parameter value group corresponding to the current operating condition identifier code from the mapping table. The expected parameter value group contains the standard expected values ​​of each electrical parameter under the current operating condition. These expected values ​​are pre-calibrated and stored according to the statistical regularity of devices in the same batch under the same operating condition. After obtaining the expected parameter value group, the processor performs a step-by-step comparison operation between the measured electrical parameters collected in step S1 and the expected parameter value group. For each electrical parameter, the processor calculates the deviation between the measured value and the expected value, and divides the deviation by the corresponding expected value for normalization processing to obtain the normalized drift amount of the parameter. When the expected value is close to zero, the processor uses a preset minimum value for zero protection as the lower limit constraint of the divisor to ensure the numerical stability of the normalization operation. The processor combines the normalized drift amounts of each electrical parameter in a predetermined order to construct a normalized drift fingerprint vector. The dimension of the drift fingerprint vector is equal to the number of electrical parameters. Each component in the vector represents the normalized drift degree of the corresponding electrical parameter; a positive value indicates that the parameter value is higher than the expected value, and a negative value indicates that the parameter value is lower than the expected value. Simultaneously with constructing the drift fingerprint vector, the processor calculates the drift rate. The drift rate is calculated as follows: the processor obtains the drift fingerprint vectors of the current sampling period and the previous sampling period, calculates the norm of the difference between each component of the two vectors, and then divides it by the sampling period duration to obtain the drift change rate per unit time. This drift rate characterizes the speed of change of device parameters. In the first sampling period, since there is no drift fingerprint vector from the previous sampling period, the processor initializes the drift rate to zero. The processor further calculates the overall confidence score to assess the credibility of the current evaluation result. The overall confidence score is equal to the product of the data accumulation confidence score and the stability confidence score. The data accumulation confidence score is calculated by comparing the cumulative number of samples with a preset saturation threshold; when the ratio is greater than or equal to 1, the upper limit of 1 is taken. The stability confidence score is calculated as follows: the processor acquires the drift fingerprint vector sequence within a preset statistical window, calculates the mean variance of the sequence, multiplies the mean variance by a preset adjustment coefficient, adds 1 as the denominator, and calculates the quotient with 1 as the numerator to obtain the stability confidence score. Through this calculation mechanism, the data accumulation confidence score is low when the initial sampling frequency is low; the stability confidence score is also low when drift data fluctuations are large. The overall confidence score comprehensively reflects the reliability of both the sufficiency and stability of the data.

[0025] See Figure 3In step S3, drift mapping early warning arbitration is performed. Specifically, the processor uses the normalized drift fingerprint vector output in step S2 as input and queries a preset group prior mapping table to obtain the group health level. The group prior mapping table 102 is a statistical mapping relationship pre-established based on aging test data of a large number of devices in the same batch. This mapping table maps drift fingerprint vectors with different value ranges to corresponding health level ranges. The processor matches the current drift fingerprint vector with the group prior mapping table and outputs the corresponding group health level value. The group health level reflects the health assessment conclusion of the current device under the group statistical law. While obtaining the group health level, the processor calculates the individual health level based on the drift fingerprint vector and the individual offset parameters pre-stored in the individual offset parameter storage area 103. The individual offset parameters are personalized deviation records accumulated by the device during historical operation, used to characterize the offset of the device relative to the average level of the group. The processor performs vector operations on the current drift fingerprint vector and the individual offset parameters, and calculates the individual health level through a preset individual evaluation function. The individual health level reflects the personalized health assessment conclusion based on the historical characteristics of the device. After acquiring the group health level and individual health level separately, the processor uses the overall confidence level calculated in step S2 as a weighting coefficient to perform a weighted fusion operation on the group health level and individual health level. The fusion formula is: the fused health level equals the overall confidence level multiplied by the individual health level plus (1 minus the overall confidence level) multiplied by the group health level. Through this fusion mechanism, when the overall confidence level is low, the fused health level is more biased towards the group health level, making full use of prior knowledge of the group for evaluation; when the overall confidence level is high, the fused health level is more biased towards the individual health level, reflecting personalized evaluation characteristics. As device operating data accumulates and confidence level increases, the evaluation result gradually transitions from group bias to individual bias, achieving a smooth transition.

[0026] See Figure 4In step S4, dual-track mapping feedback correction is performed. Specifically, this step includes two independent feedback correction paths, which independently correct mapping deviation and warning deviation respectively. During the first two sampling periods after system startup, since there are insufficient historical fusion health levels and historical final warning levels for calculating the mapping deviation rate and warning deviation rate, the processor initializes both the mapping deviation rate and warning deviation rate to zero and does not trigger any feedback correction action until normal calculation begins in the third sampling period. The first feedback path adjusts the learning rate and updates individual offset parameters in response to the mapping deviation rate exceeding the limit trigger signal. The processor first calculates the mapping deviation rate, which is equal to the absolute value of the difference between the actual change and the predicted change in the fusion health level, divided by the larger of the absolute value of the predicted change and the preset minimum value for preventing zero. The predicted change is equal to the difference between the fusion health level of the previous period and the fusion health level of the previous two periods. The processor compares the calculated mapping deviation rate with a preset first threshold and a second threshold. When the mapping deviation rate is lower than the first threshold, it indicates that the prediction accuracy of the current mapping model is within an acceptable range, and the processor maintains the learning rate unchanged and does not update the individual offset parameters. When the mapping deviation rate is between the first and second thresholds, it indicates a moderate deviation in the mapping model. The processor increases the learning rate by a preset percentage to accelerate the update speed of individual offset parameters, allowing the model to adapt to device changes more quickly. When the mapping deviation rate is higher than the second threshold, it indicates a significant deviation in the mapping model, potentially indicating abnormal fluctuations. The processor decreases the learning rate by a preset percentage to reduce the impact of abnormal data on the model. After adjusting the learning rate, the processor incrementally updates the individual offset parameters based on the current drift fingerprint vector and the fused health level. The second feedback path is the expected parameter value in the parameter expectation value mapping table, which is used to correct the trigger signal for the warning deviation rate exceeding the limit. The processor first calculates the warning deviation rate, which is equal to the absolute value of the difference between the final warning level of the previous cycle and the final warning level of the two previous cycles, divided by the larger of the absolute value of the final warning level of the two previous cycles and the preset minimum value for zero prevention. The processor compares the calculated warning deviation rate with the preset first and second deviation thresholds. When the warning deviation rate is lower than the first deviation threshold, the processor maintains the expected parameter value unchanged. When the warning deviation rate is between the first and second deviation thresholds, the processor fine-tunes the expected parameter value according to a preset proportional step size. When the warning deviation rate is higher than the second deviation threshold, it indicates that the current expected parameter value deviates significantly from the actual state. The processor then overwrites the expected parameter value of the corresponding operating condition in the updated operating condition-expected parameter value mapping table with the measured parameter value of the current sampling period. To avoid system oscillations caused by frequent corrections, preset lag intervals and preset cooldown times are set for both individual offset parameter updates and expected parameter value corrections.After triggering a correction action, the processor enters a cooldown time lock state, during which it does not respond to new over-limit trigger signals, ensuring that the correction effect is stable before proceeding with the next correction judgment. Through a dual-track feedback correction mechanism, mapping deviation optimizes the accuracy of individual models by updating individual offset parameters, while warning deviation optimizes the benchmark reference by correcting the expected parameter values. The two paths do not interfere with each other and each performs its own function, achieving independent optimization of mapping parameters and warning parameters.

[0027] In step S5, a dual-trigger warning level output is performed. Specifically, the processor compares the fusion health level output in step S3 with a preset drift threshold to obtain the drift trigger level. The drift threshold is divided into multiple level intervals according to severity, and the processor uses the interval number into which the fusion health level falls as the drift trigger level. Simultaneously, the processor compares the drift rate calculated in step S2 with a preset rate threshold to obtain the rate trigger level. The rate threshold is divided into multiple level intervals according to the rate of change, and the processor uses the interval number into which the drift rate falls as the rate trigger level. After obtaining the drift trigger level and the rate trigger level, the processor performs a larger value operation on the two levels and outputs the final warning level. This dual-trigger arbitration strategy can capture both the gradual parameter degradation risk caused by accumulated drift and the sudden parameter anomalies caused by abrupt changes in drift rate, achieving comprehensive coverage of both risk dimensions. The final warning level is sent to the upper-level system or display device through a preset output interface for maintenance personnel to refer to and handle.

[0028] Example 2: Based on the general method provided in Embodiment 1 above, this embodiment provides a specific application scenario: a health monitoring and early warning scenario for shielded gate MOS power transistors in the electric drive system of new energy vehicles. In this scenario, the aforementioned shielded gate MOS device is specifically embodied in the shielded gate MOS power transistor in the electric vehicle motor driver, which is used to drive a permanent magnet synchronous motor to achieve vehicle power output. The main control unit of the electric drive system is deployed in the vehicle domain controller, equipped with a 32-bit microprocessor and 8GB of storage space, and is used to execute the early warning method of this embodiment.

[0029] In step S1, the electrical parameters of the operating condition are sampled. Specifically, after the electric drive system completes its startup self-test, the main control unit detects the steady-state judgment flag. This flag is output by the motor speed stability detection module. When the motor speed fluctuation is less than 2% of the rated speed for 10 consecutive sampling cycles, the flag is valid. Under the condition that the flag is valid, the main control unit collects the electrical parameters of the shielded gate MOS power transistor through a 12-bit precision analog-to-digital converter interface. The electrical parameters include the threshold voltage Vth, the on-resistance Rds(on), and the drain-source current Ids. In this embodiment, the threshold voltage is obtained through the threshold detection channel of the gate drive circuit, with a typical sampling value of 3.5V; the on-resistance is calculated by the ratio of the drain-source voltage to the drain-source current, with a typical sampling value of 8mΩ; the drain-source current is collected by a Hall current sensor, with a typical sampling value of 120A. The sampling period is set to 100ms, and the sampled data is stored in the loop buffer of the main control unit. While acquiring the electrical parameters, the main control unit obtains the current load current value and the ambient temperature value through the vehicle sensors. The load current range is divided into three levels: light load range (less than 50A), rated load range (50A to 150A), and heavy load range (greater than 150A). The ambient temperature range is also divided into three levels: low temperature range (below -20°C), normal temperature range (between -20°C and 60°C), and high temperature range (above 60°C). Assuming the current load current is 100A and the ambient temperature is 35°C, the load current falling into the rated load range is coded as 2, and the ambient temperature falling into the normal temperature range is also coded as 2. The main control unit combines these two codes to generate the operating condition identifier code 22. This operating condition identifier code uniquely identifies the rated load plus normal temperature operating conditions.

[0030] See Figure 2In step S2, confidence fusion drift fingerprint construction is performed. Specifically, the main control unit uses operating condition identifier 22 as the index key to query the pre-calibrated operating condition-parameter expected value mapping table 101. This mapping table was established before the vehicle left the factory using parameter test data of 500 shielded gate MOS power transistors in the same batch under different operating conditions, and contains expected parameter value groups for 9 operating condition combinations. For the rated load plus normal temperature operating condition corresponding to operating condition identifier 22, the expected parameter value group returned by the mapping table is: threshold voltage expected value 3.5V, on-resistance expected value 8mΩ, and drain-source current expected value 120A. Assume that the currently measured electrical parameters are: threshold voltage 3.605V, on-resistance 8.48mΩ, and drain-source current 117.6A. The main control unit calculates the normalized drift of each parameter: the normalized drift of the threshold voltage equals (3.605 - 3.5) divided by 3.5, which equals 0.03; the normalized drift of the on-resistance equals (8.48 - 8) divided by 8, which equals 0.06; and the normalized drift of the drain-source current equals (117.6 - 120) divided by 120, which equals -0.02. The main control unit combines the three normalized drift values ​​in sequence to construct a normalized drift fingerprint vector of [0.03, 0.06, -0.02]. This vector indicates that the threshold voltage drifts upward by 3%, the on-resistance drifts upward by 6%, and the drain-source current drifts downward by 2%, reflecting the parameter drift characteristics of the current power transistor. Simultaneously with constructing the drift fingerprint vector, the main control unit calculates the drift rate. In the first sampling period, since there is no drift fingerprint vector from the previous period, the drift rate is initialized to zero. Assuming the current operation is stable, the drift fingerprint vector from the previous sampling period is [0.025, 0.055, -0.018], and the sampling period is 100ms. Therefore, the drift rate equals the vector difference norm divided by 0.1 seconds, which is approximately 0.08 seconds. The main control unit further calculates the overall confidence level. Assuming the power transistor has accumulated 1200 samples, and the preset saturation threshold is 1000 samples, the data accumulation confidence level equals 1200 divided by 1000, with an upper limit of 1. Assuming the preset statistical window is 20 periods, the mean variance of the drift fingerprint vector within the window is 0.0005, and the preset adjustment coefficient is 100, the stability confidence level equals 1 divided by (1 plus 100 multiplied by 0.0005), which is approximately 0.95. The overall confidence level equals 1 multiplied by 0.95, which equals 0.95.

[0031] See Figure 3In step S3, drift mapping early warning arbitration is performed. Specifically, the main control unit uses the normalized drift fingerprint vector [0.03, 0.06, -0.02] as input and queries the population prior mapping table 102. This mapping table is established based on the parameter change curves of 500 power transistors in the same batch during 2000 hours of accelerated aging testing, mapping the drift fingerprint vector to a health level range of 0 to 100. For the current drift fingerprint vector, the population prior mapping table outputs a population health level of 85, indicating that the power transistor is in good health according to the population statistical law. At the same time, the main control unit reads the individual offset parameter of the power transistor from the individual offset parameter storage area 103, assuming it is [0.005, -0.01, 0.008]. The main control unit performs vector operation on the current drift fingerprint vector and the individual offset parameter, and calculates the individual health level as 82 through the preset individual evaluation function. Because this power transistor exhibited a slightly higher-than-average on-resistance drift trend in its historical operation, its individual health level was slightly lower than the group health level. After obtaining a group health level of 85 and an individual health level of 82, the main control unit performed a fusion calculation using a comprehensive confidence level of 0.95 as a weighting coefficient: the fused health level equals 0.95 multiplied by 82 plus (1 minus 0.95) multiplied by 85 equals 77.9 plus 4.25 equals 82.15. Due to the higher comprehensive confidence level, the fused health level is more biased towards the individual health level of 82.

[0032] See Figure 4In step S4, dual-track mapping feedback correction is performed. During the first two sampling periods after system startup (corresponding to the first 200ms after vehicle startup), since sufficient historical fusion health levels and historical final warning levels have not yet been accumulated, the main control unit initializes both the mapping deviation rate and the warning deviation rate to zero, without triggering any feedback correction action. From the third sampling period onwards, the main control unit begins to calculate the deviation rate normally. Specifically, the main control unit first calculates the mapping deviation rate. Assuming the fusion health level of the previous period was 82.8 points and the fusion health level of the previous two periods was 83.5 points, the predicted change is equal to 82.8 minus 83.5, which equals -0.7 points. The current period's fusion health level is 82.15 points, and the actual change is equal to 82.15 minus 82.8, which equals -0.65 points. The mapping deviation rate is equal to the absolute value of -0.65 minus (-0.7) divided by the absolute value of 0.7, which equals 0.05 divided by 0.7, which equals approximately 7.1%. Assuming the first threshold is set to 10% and the second threshold to 30%, since the mapping deviation rate of 7.1% is lower than the first threshold of 10%, the main control unit maintains the learning rate at 0.1 and does not trigger the individual offset parameter update operation. The main control unit further calculates the warning deviation rate. Assuming the final warning level of the previous cycle is level 1, and the final warning level of the previous two cycles is also level 1, then the warning deviation rate equals the absolute value of 1 minus 1 divided by 1, which equals 0%. Assuming the first deviation threshold is set to 15%, since the warning deviation rate of 0% is lower than the first deviation threshold, the main control unit maintains the expected parameter value in the operating condition-parameter expected value mapping table unchanged. During this sampling period, the dual-track feedback correction mechanism does not trigger any correction action, and the system is in a stable operating state. Assuming that in a subsequent cycle, due to the power transistor experiencing a high load condition and the parameter changes significantly, the mapping deviation rate rises to 25% (between 10% and 30%), the main control unit increases the learning rate by 10% to 0.11 and updates the individual offset parameters based on the current drift fingerprint vector. If the warning deviation rate simultaneously rises to 50% (above the second deviation threshold of 30%), the main control unit will overwrite the expected parameter value of the corresponding operating condition identifier 22 in the updated operating condition-parameter expected value mapping table with the current measured parameter value. Through this dual-track feedback correction mechanism, the mapping deviation optimizes the personalized evaluation model of the power transistor by updating the individual offset parameters, while the warning deviation optimizes the parameter benchmark under the operating condition by correcting the expected parameter value. The two correction paths do not interfere with each other and each performs its own function, enabling the system to maintain warning accuracy continuously during the long-term operation and parameter aging of the power transistor.

[0033] In step S5, a dual-trigger warning level output is executed. Specifically, the main control unit compares the fusion health level of 82.15 with a preset drift threshold. The drift threshold is divided into four levels: Level 0 corresponds to a fusion health level greater than or equal to 90 (normal state), Level 1 corresponds to a fusion health level between 80 and 90 (attention state), Level 2 corresponds to a fusion health level between 60 and 80 (warning state), and Level 3 corresponds to a fusion health level less than 60 (emergency state). Since the current fusion health level of 82.15 falls within the 80-90 range, the drift trigger level is Level 1. Simultaneously, the main control unit compares the drift rate of 0.08 seconds with a preset rate threshold. The drift rate threshold is divided into four levels: Level 0 corresponds to a drift rate less than 0.05 s / s, indicating a stable state; Level 1 corresponds to a drift rate between 0.05 s / s and 0.1 s / s, indicating a slowly changing state; Level 2 corresponds to a drift rate between 0.1 s / s and 0.5 s / s, indicating a rapidly changing state; and Level 3 corresponds to a drift rate greater than 0.5 s / s, indicating a sudden change. Since the current drift rate of 0.08 s / s falls within the 0.05 to 0.1 s / s range, the rate trigger level is Level 1. The main control unit calculates the larger value between the drift trigger level (Level 1) and the rate trigger level (Level 1), outputting a final warning level of Level 1. This warning level is sent to the vehicle controller via the vehicle's CAN bus, displaying a yellow warning icon on the instrument panel, reminding the driver that this power transistor needs to be checked during the next maintenance.

[0034] Based on the above embodiments, the specific types of electrical parameters and the division of operating ranges are further as follows. In this embodiment, the electrical parameters are selected as threshold voltage Vth, on-resistance Rds(on), and drain-source current Ids. The technical basis for selecting these three parameters is that: threshold voltage Vth can characterize the aging degree of the shielding gate oxide layer; when trap charges accumulate in the oxide layer, the threshold voltage will drift; on-resistance Rds(on) can characterize the deterioration degree of the conductive channel; when carrier mobility decreases or doping concentration changes, the on-resistance will increase; the change of drain-source current Ids under the same driving conditions can reflect the degradation of the overall conductivity of the device. The three parameters are independent yet interrelated, comprehensively reflecting the health status of the shielded gate MOS device. The load current range is divided into three levels: the light load range corresponds to a load current less than 50A, the rated load range corresponds to a load current between 50A and 150A, and the heavy load range corresponds to a load current greater than 150A. The ambient temperature range is also divided into three levels: the low temperature range corresponds to an ambient temperature below -20 degrees Celsius, the normal temperature range corresponds to an ambient temperature between -20 degrees Celsius and 60 degrees Celsius, and the high temperature range corresponds to an ambient temperature above 60 degrees Celsius. This range division creates a total of nine operating condition combinations, covering the main operating conditions of the electric drive system of new energy vehicles.

[0035] Further, in a preferred embodiment, the calculation method of the overall confidence level is as follows: The overall confidence level is equal to the product of the data accumulation confidence level and the stability confidence level. The calculation formula for the data accumulation confidence level is: the data accumulation confidence level is equal to the ratio of the cumulative number of samples to a preset saturation threshold, and the upper limit of 1 is taken when the ratio is greater than or equal to 1. In this embodiment, the preset saturation threshold is set to 1000 times, indicating that the data accumulation confidence level reaches the full value of 1 after the cumulative number of samples reaches 1000 times. The calculation formula for the stability confidence level is: the stability confidence level is equal to 1 divided by (1 plus the adjustment coefficient multiplied by the mean variance). Wherein, the mean variance is the arithmetic mean of the variances of each component of the normalized drift fingerprint vector within the preset statistical window. In this embodiment, the preset statistical window is set to 20 sampling periods, and the adjustment coefficient is set to 100. When the drift data fluctuation within the window is large, the mean variance increases, and the stability confidence level decreases; when the drift data fluctuation is small, the mean variance decreases, and the stability confidence level increases. Through this calculation mechanism, the overall confidence level can simultaneously reflect the sufficiency and stability of the data, providing a reliable weighting basis for group-individual fusion.

[0036] Further, the specific calculation formulas for the mapping deviation rate and the early warning deviation rate are as follows. The mapping deviation rate is calculated as follows: the absolute value of the difference between the actual change and the predicted change in the fused health level is divided by the larger of the absolute value of the predicted change and the preset minimum value for zero prevention. Wherein, the predicted change is equal to the difference between the fused health level of the previous cycle and the fused health level of the two previous cycles. In this embodiment, the preset minimum value for zero prevention is set to 0.01 to avoid division-by-zero anomalies. The early warning deviation rate is calculated as follows: the early warning deviation rate is equal to the absolute value of the difference between the final early warning level of the previous cycle and the final early warning level of the two previous cycles is divided by the larger of the absolute value of the final early warning level of the two previous cycles and the preset minimum value for zero prevention. When the early warning level remains stable, the early warning deviation rate is 0; when the early warning level changes abruptly, the early warning deviation rate will increase significantly, triggering the expected parameter value correction mechanism.

[0037] Furthermore, the learning rate adjustment strategy and the expected parameter value correction strategy are as follows. For learning rate adjustment, a three-threshold control strategy is adopted. When the mapping deviation rate is below the first threshold of 10%, it indicates that the mapping model has good prediction accuracy, and the learning rate remains unchanged. When the mapping deviation rate is between the first threshold of 10% and the second threshold of 30%, it indicates that the mapping model has a moderate degree of deviation, and the learning rate is increased by 10% to accelerate the update speed of individual offset parameters. When the mapping deviation rate is above the second threshold of 30%, it indicates that there may be abnormal fluctuations, and the learning rate is decreased by 20% to reduce the impact of abnormal data on the model. In this embodiment, the initial value of the learning rate is set to 0.1, with a range of 0.05 to 0.2. For the correction of the expected parameter value, the same three-threshold control strategy is adopted. When the warning deviation rate is below the first deviation threshold of 15%, the expected parameter value remains unchanged. When the warning deviation rate is between the first deviation threshold of 15% and the second deviation threshold of 30%, the expected parameter value is fine-tuned and corrected in a step size of 5%. When the warning deviation rate is higher than the second deviation threshold of 30%, the expected parameter value of the corresponding working condition is updated by overwriting the measured parameter value of the current sampling period.

[0038] Furthermore, to avoid system oscillations caused by frequent corrections, both the individual offset parameter update and the expected parameter value correction are configured with lag intervals and cooling times. In this embodiment, the lag interval for the individual offset parameter update is set to 5% of the mapping deviation rate, meaning that when the mapping deviation rate fluctuates within 5% of the threshold, a state switch is not triggered. The cooling time for the individual offset parameter update is set to 10 sampling periods, meaning that within 10 periods after an update is triggered, no new over-limit trigger signals will be responded to. The lag interval for the expected parameter value correction is set to 3% of the warning deviation rate, and the cooling time is set to 20 sampling periods. By setting the lag interval and cooling time, it is ensured that the correction action is performed only after stabilization, avoiding frequent switching near the critical state.

[0039] In other embodiments, the selection of the aforementioned electrical parameters can be adjusted according to specific application scenarios. For example, in industrial frequency converter applications, the electrical parameters can be expanded to four parameters: threshold voltage, on-resistance, drain-source current, and gate charge, to obtain a more comprehensive characterization of the device status. In photovoltaic inverter applications, the electrical parameters can be simplified to two parameters: on-resistance and reverse recovery charge, to reduce computational complexity. Furthermore, the granularity of the operating condition range division can also be adjusted according to actual needs; for example, the load current range can be expanded to 5 or 7 levels to achieve finer operating condition sensing.

[0040] Example 3: See Figure 5This embodiment also provides a drift-mapped shielded gate MOS early warning system for executing the aforementioned drift-mapped shielded gate MOS early warning method. The system includes a condition-sensing electrical parameter sampling module 10, a confidence-based drift fingerprint construction module 20, a drift-mapped early warning arbitration module 30, and a dual-track mapping feedback correction module 40. These modules interact through a preset data channel.

[0041] The operating condition sensing electrical parameter sampling module 10 is used to perform the function corresponding to step S1 in the above method. Specifically, the operating condition sensing electrical parameter sampling module 10 is configured to: respond to the validity of the operating condition steady-state determination flag bit, collect the electrical parameters of the shielded gate MOS device through the analog-to-digital conversion interface, the electrical parameters including threshold voltage, on-resistance, and drain-source current; divide the load current value and ambient temperature value into intervals, and combine the load current interval code and the ambient temperature interval code to generate an operating condition identification code. The operating condition sensing electrical parameter sampling module 10 outputs the collected electrical parameters and the generated operating condition identification code to the confidence fusion drift fingerprint construction module 20.

[0042] The confidence-based drift fingerprint construction module 20 is used to perform the function corresponding to step S2 in the above method. In one implementation, the confidence-based drift fingerprint construction module 20 includes a working condition index unit 21, a drift fingerprint construction unit 22, and a confidence-based calculation unit 23. The working condition index unit 21 is used to query the working condition-parameter expected value mapping table 101 using the working condition identifier code as the index key to obtain the expected parameter value group corresponding to the current working condition. The drift fingerprint construction unit 22 is used to compare the measured electrical parameters with the expected parameter value group item by item, calculate the normalized drift amount of each parameter, and combine the normalized drift amounts in a predetermined order to construct a normalized drift fingerprint vector, while calculating the drift rate. The confidence-based calculation unit 23 is used to calculate the comprehensive confidence, which is equal to the product of the data accumulation confidence and the stability confidence. The confidence-based drift fingerprint construction module 20 outputs the normalized drift fingerprint vector, the drift rate, and the comprehensive confidence to the drift mapping early warning arbitration module 30.

[0043] The drift mapping early warning arbitration module 30 is used to perform the functions corresponding to steps S3 and S5 in the above method. In one implementation, the drift mapping early warning arbitration module 30 includes a group health level assessment unit 31, an individual health level assessment unit 32, a confidence fusion unit 33, and a dual-trigger arbitration unit 34. The group health level assessment unit 31 is used to query the group prior mapping table 102 with the normalized drift fingerprint vector as input and output the group health level. The individual health level assessment unit 32 is used to read the individual offset parameters from the individual offset parameter storage area 103 and calculate the individual health level by combining the normalized drift fingerprint vector. The confidence fusion unit 33 is used to calculate the fused health level by adding 1 to the product of the comprehensive confidence level and the individual health level and subtracting the product of the comprehensive confidence level and the group health level. The dual-trigger arbitration unit 34 is used to obtain the drift amount trigger level based on the fused health level, obtain the rate trigger level based on the drift rate, and perform a calculation on the larger value of the two levels to output the final early warning level. The drift mapping early warning arbitration module 30 will integrate the health level and the final early warning level and output them to the dual-track mapping feedback correction module 40.

[0044] The dual-track mapping feedback correction module 40 performs the function corresponding to step S4 in the above method. In one implementation, the dual-track mapping feedback correction module 40 includes a mapping deviation correction unit 41 and a warning deviation correction unit 42. The mapping deviation correction unit 41 calculates the mapping deviation rate, adjusts the learning rate in response to the mapping deviation rate exceeding the limit trigger signal, and updates the individual offset parameters in the individual offset parameter storage area 103 based on the current drift fingerprint vector. The warning deviation correction unit 42 calculates the warning deviation rate and corrects the expected parameter values ​​in the working condition-parameter expected value mapping table 101 in response to the warning deviation rate exceeding the limit trigger signal. The two correction units operate independently to achieve dual-track feedback correction.

[0045] In practical implementation, the aforementioned functional modules can be implemented as software programs, with the corresponding functions completed by the microprocessor in the vehicle domain controller executing the program code in the memory. Each module can also be implemented using dedicated hardware circuits, such as FPGA devices for high-speed data processing, or a combination of software programs and hardware circuits. The operating condition-parameter expected value mapping table 101, the group prior mapping table 102, and the individual offset parameter storage area 103 are all stored in non-volatile memory, ensuring data retention even after vehicle power is lost.

[0046] In summary, the drift-mapped shielded gate MOS early warning method and system provided by the embodiments of the present invention can achieve the following technical effects: By employing a condition-aware electrical parameter sampling and condition identification code generation mechanism, this solution can distinguish the expected parameter values ​​under different load and temperature conditions, avoiding misjudging normal operating condition changes as parameter drift and improving early warning accuracy. Compared to the fixed threshold detection method in the background technology, which cannot adapt to varying operating conditions, this solution achieves dynamic benchmark reference through an operating condition-expected parameter value mapping table.

[0047] By employing a normalized drift fingerprint vector construction mechanism, the drift quantities of electrical parameters with different dimensions can be converted into dimensionless vectors of a uniform scale, facilitating subsequent health mapping and fusion evaluation. Compared to the problem of single-parameter threshold monitoring in the background technology, this scheme achieves a comprehensive characterization of the device health status through multi-parameter vectorization.

[0048] By employing a group-individual health level fusion mechanism with comprehensive confidence weighting, this approach fully leverages prior knowledge of the group in the early stages of device development and fully reflects individual characteristics in the later stages, achieving a smooth transition in the evaluation strategy. Compared to background technologies that rely solely on group statistical patterns or individual historical data, this solution combines the advantages of both through confidence weighted fusion.

[0049] Because it employs a dual-track mapping feedback correction mechanism, it can independently optimize mapping deviation and early warning deviation. Mapping deviation is addressed by updating individual offset parameters to improve individual model accuracy, while early warning deviation is addressed by correcting expected parameter values ​​to optimize the benchmark. These two paths operate independently, ensuring the system maintains consistent early warning accuracy over long-term operation. Compared to the lack of adaptive correction mechanisms in previous technologies, this solution achieves online optimization of model and benchmark parameters through dual-track feedback.

[0050] By employing a dual-trigger arbitration strategy involving both drift magnitude and rate trigger levels, this approach can capture both the gradual parameter degradation risk caused by accumulated drift and the sudden parameter anomalies caused by abrupt changes in drift rate, achieving comprehensive coverage of both risk dimensions. Compared to the single-dimensional early warning strategies in the background technology, this solution improves the comprehensiveness and timeliness of early warning through dual-trigger arbitration.

[0051] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning of drift mapping in a shielded gate MOS, characterized in that, Includes the following steps: When the steady-state judgment flag of the response condition is valid, the electrical parameters of the shielded gate MOS device are collected through the data acquisition interface, and the operating condition identification code is generated based on the combination of the load current range and the ambient temperature range. The expected parameter value group is obtained by indexing the working condition-parameter expected value mapping table with the working condition identification code, and a normalized drift fingerprint vector is constructed based on the electrical parameters and the expected parameter value group, and the drift rate and comprehensive confidence are calculated. The group health level is obtained by indexing the preset group prior mapping table with the drift fingerprint vector, the individual health level is calculated based on the drift fingerprint vector and the individual offset parameter, and the fused health level is obtained by adding 1 to the product of the comprehensive confidence and the individual health level and subtracting the product of the comprehensive confidence and the group health level. The learning rate is adjusted and the individual offset parameter is updated in response to the trigger signal that the mapping deviation rate exceeds the limit; the expected parameter value of the working condition-parameter expected value mapping table is corrected in response to the trigger signal that the early warning deviation rate exceeds the limit. The drift trigger level is obtained by comparing the fused health level with a preset threshold, and the rate trigger level is obtained by comparing the drift rate with a preset threshold. The larger of the two values ​​is then used to output the final warning level.

2. The method for early warning of drift mapping using a shielded gate MOS according to claim 1, characterized in that, The electrical parameters include threshold voltage, on-resistance, and drain-source current; the load current range includes light load range, rated load range, and heavy load range; and the ambient temperature range includes low temperature range, normal temperature range, and high temperature range.

3. The method for early warning of drift mapping shielded gate MOS according to claim 1, characterized in that, The overall confidence level is equal to the product of the data accumulation confidence level and the stability confidence level. The data accumulation confidence level is equal to the ratio of the cumulative number of samples to the preset saturation threshold and has an upper limit of 1. The stability confidence level is calculated by multiplying the preset adjustment coefficient by the mean of the drift variance within the preset statistical window, adding 1 as the denominator, and using 1 as the numerator to find the quotient.

4. The method for early warning of drift mapping shielded gate MOS according to claim 1, characterized in that, The mapping deviation rate is equal to the absolute value of the difference between the actual change and the predicted change in the fused health level, divided by the larger of the absolute value of the predicted change and the preset minimum value for zero prevention. The predicted change is equal to the difference between the fused health level of the previous period and the fused health level of the previous two periods. The early warning deviation rate is equal to the absolute value of the difference between the final early warning level of the previous period and the final early warning level of the previous two periods, divided by the larger of the absolute value of the final early warning level of the previous two periods and the preset minimum value for zero prevention.

5. The method for early warning of drift mapping shielded gate MOS according to claim 1, characterized in that, The learning rate remains unchanged when the mapping deviation rate is lower than a preset first threshold, increases by a preset ratio when the mapping deviation rate is between a preset first threshold and a preset second threshold, and decreases by a preset ratio when the mapping deviation rate is higher than a preset second threshold. The expected parameter value remains unchanged when the warning deviation rate is lower than the preset first deviation threshold, is adjusted according to a preset proportional step size when the warning deviation rate is between the preset first deviation threshold and the second deviation threshold, and is updated by overwriting the measured parameter value of the current sampling period when the warning deviation rate is higher than the preset second deviation threshold.

6. The method for early warning of drift mapping using a shielded gate MOS according to claim 1, characterized in that, Both the individual offset parameter update and the expected parameter value correction are set with a preset lag interval and a preset cooling time.

7. A drift-mapped shielded gate MOS early warning system, characterized in that, include: The operating condition sensing electrical parameter sampling module is configured to collect the electrical parameters of the shielded gate MOS device when the operating condition steady state determination flag is valid, and generate an operating condition identification code based on the combination of the load current range and the ambient temperature range. The confidence fusion drift fingerprint construction module is configured to obtain the expected parameter value group by indexing the working condition-parameter expected value mapping table with the working condition identifier code, construct a normalized drift fingerprint vector based on the electrical parameters and the expected parameter value group, and calculate the drift rate and comprehensive confidence. The drift mapping early warning arbitration module is configured to obtain the group health level by indexing a preset group prior mapping table with the drift fingerprint vector, calculate the individual health level based on the drift fingerprint vector and the individual offset parameter, add 1 to the product of the comprehensive confidence level and the individual health level and subtract the product of the comprehensive confidence level and the group health level as the fusion health level, obtain the drift amount trigger level by comparing the fusion health level with a preset threshold, obtain the rate trigger level by comparing the drift rate with a preset threshold, and output the final early warning level by taking the larger value of the two. The dual-track mapping feedback correction module is configured to adjust the learning rate and update the individual offset parameters in response to the mapping deviation rate exceeding the limit trigger signal, and to correct the expected parameter values ​​of the working condition-parameter expected value mapping table in response to the warning deviation rate exceeding the limit trigger signal.

8. The drift-mapped shielded gate MOS early warning system according to claim 7, characterized in that, The system is also configured to implement the method of any one of claims 2 to 6.

Images