Battery protection board life prediction and fault diagnosis system based on deep learning

Abstract

The invention relates to the technical field of electrical equipment test and evaluation, and particularly discloses a battery protection board life prediction and fault diagnosis system based on deep learning, which deeply integrates functions of intelligent test driving, real-time state evaluation, autonomous decision judgment, accurate analysis and diagnosis and the like. A complete closed-loop management system for predicting the service life of the battery protection board is constructed, the limitation of dependence on a fixed test process and post analysis traditionally is broken through, the system can simulate the judgment process of a qualified expert, the dynamic response characteristic of an element is sensed in real time in the test process, and the test strategy and parameters are adaptively adjusted according to the dynamic response characteristic. The intelligent normal form not only realizes full-process autonomous operation from health state screening, life prediction to fault diagnosis, but also obviously improves the automation level and intelligent decision-making capability of the whole test and evaluation process, and establishes a new technical standard for reliability evaluation of the battery protection board.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment testing and evaluation technology, specifically a battery protection board life prediction and fault diagnosis system based on deep learning. Background Technology

[0002] In the field of electrical equipment testing and evaluation technology, battery protection board life prediction and fault diagnosis are being carried out. Currently, in this field, technological development is evolving from traditional threshold-based simple diagnosis to a more intelligent approach, combining multimodal data, artificial intelligence, and physical models to achieve more accurate prediction and health management. In terms of life prediction, to address the challenge of missing early data, feature extraction combined with deep learning models such as convolutional neural networks is used to achieve high-precision life status assessment. In terms of fault diagnosis, advanced methods such as multimodal data fusion and artificial intelligence algorithms are widely used. Hybrid neural networks are used to extract the spatiotemporal features of battery data, and image recognition and sensor data analysis are used to analyze the structural health status, initially achieving quantitative assessment of the remaining lifespan of the battery protection board and early warning of potential faults.

[0003] For example, the invention patent announcement CN116593902B discloses a fault diagnosis method based on the fusion of performance parameters of lithium iron phosphate batteries into an AI model. This method includes: establishing an AI model, which includes the operating environment, performance data of the ladder-type battery, and performance data of individual cells; establishing an alarm module, which alarms for defects in the operating environment, ladder-type battery performance data, and individual cell performance data; obtaining the cell characteristic values ​​of the ladder-type battery, which are obtained through the charge-discharge curve of the ladder-type battery; plotting the life prediction curve of the ladder-type battery; obtaining the operating time of the ladder-type battery under alarm conditions based on the alarm data of the operating environment and ladder-type battery performance data of the alarm module, and calculating the performance loss value based on the time; and re-plotting the life prediction curve of the ladder-type battery based on the performance loss value to obtain a new life prediction time.

[0004] For example, the invention patent announcement CN113608134B discloses a method for predicting the cycle life and remaining life of a lithium-ion battery; the first step is to take the battery to be tested, and under the set working environment, perform a cyclic charge-discharge test on the battery to be tested with a set current and voltage, and collect the number of cycles and the battery capacity data after each cycle; the second step is to establish the relationship curve between electrolyte content and battery capacity retention rate; the third step is to determine the electrolyte consumption rate after short-term cycling of the battery to be tested; the fourth step is to calculate the remaining life of the battery and the battery life.

[0005] However, in the process of implementing the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems: Existing methods for predicting the lifespan and diagnosing faults of battery protection boards usually rely on experimental data from accelerated aging tests. Methods based on aging tests can often only provide effective diagnosis when the battery protection board has undergone significant degradation or is even close to failure. They lack the ability to warn of early potential faults, which makes the maintenance strategy still passive. At the same time, the data obtained is usually collected under preset fixed conditions, which is difficult to cover the varied and complex operating conditions in actual operation. This results in insufficient generalization ability of the trained model, and its performance drops significantly when facing novel fault modes or different operating environments. In addition, most of these methods are still limited to a single data analysis dimension and fail to deeply integrate multimodal information such as the electrical characteristics, thermal behavior and structural state of the protection board, thereby limiting the accuracy and reliability of diagnosis and prediction. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a deep learning-based battery protection board life prediction and fault diagnosis system, which can effectively solve the problems mentioned in the background technology.

[0007] To achieve the above objectives, this invention employs the following technical solution: a deep learning-based battery protection board lifespan prediction and fault diagnosis system, comprising a test initialization module for marking the MOSFET components of the battery protection board as target components, progressively performing temperature rise tests on the target components, and initializing the test parameters for the protection test of the target components based on the final results of the temperature rise tests; a test optimization module for performing protection tests on the target components, acquiring and analyzing the transient stress parameters of the target components when the protection action is triggered during the protection test, thereby matching the impact risk level of the target components and optimizing the test parameters for the next protection test of the target components; and a judgment prediction module for acquiring the protection test results of the target components, determining whether to perform lifespan prediction on the target components, and if it is determined that lifespan prediction should be performed on the target components, then lifespan prediction of the battery protection board is performed based on deep learning; if it is determined that lifespan prediction should not be performed on the target components, then fault diagnosis of the battery protection board is performed based on deep learning.

[0008] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0009] (1) This invention provides a battery protection board life prediction and fault diagnosis system based on deep learning, which deeply integrates functions such as intelligent test driving, real-time status assessment, autonomous decision-making and accurate analysis and diagnosis, and constructs a complete closed-loop management system for predicting the life of battery protection boards. It breaks through the limitations of traditional reliance on fixed test processes and post-analysis, enabling the system to simulate the judgment process of senior experts, perceive the dynamic response characteristics of components in real time during the test, and adaptively adjust test strategies and parameters accordingly. This intelligent paradigm not only realizes the autonomous operation of the entire process from health status screening and life prediction to fault diagnosis, but also significantly improves the automation level and intelligent decision-making ability of the entire test evaluation process, and establishes a new technical standard for the reliability assessment of battery protection boards.

[0010] (2) By adopting an adaptive progressive temperature rise test algorithm, the system can safely and reliably detect the performance boundary parameters of components, effectively avoiding the instantaneous overload damage that may be caused by traditional fixed high current test methods. This not only protects the safety of the tested components and test equipment, but also improves the reliability of the test process. It automatically generates a multi-dimensional and clearly labeled feature dataset containing the complete evolution process of the component from normal state to critical state, providing training and verification materials for subsequent deep learning models. This fundamentally solves the core technical problem of artificial intelligence models relying on scarce failure data, and significantly improves the safety and efficiency of the test process.

[0011] (3) By introducing the thermal shock index as a comprehensive quantitative indicator, the system can perform multi-parameter coupled calculation of the transient electrical stress and thermal stress that the component is subjected to during the test, thereby realizing the early perception and warning of potential faults. Based on the dynamic risk level assessment system constructed by the index, the system can accurately identify the risk state in the early stage when the component performance has slightly degraded but has not yet failed, and automatically trigger the test parameter optimization or fault diagnosis process, significantly enhancing the system's early warning capability and reliability management level, and realizing early, quantitative and real-time accurate assessment of the health status of the battery protection board.

[0012] (4) This invention uses multi-dimensional feature data generated during the adaptive testing process, including thermal shock index, dynamic filtering time constant and temperature rise rate curve, as input features of the deep learning model. The system makes full use of key parameters that reflect the operating status and stress response of the battery protection board, implements an intelligent current diversion processing mechanism based on real-time test results, performs accurate remaining life prediction for components that pass the test, and initiates multi-modal fault diagnosis and location analysis based on complete process data for failed components. The dual protection mechanism effectively enhances the accuracy, reliability and engineering practical value of the output results, and provides solid technical support for the status assessment and health management of the battery protection board. Attached Figure Description

[0013] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the system module connections of the present invention.

[0015] Figure 2 This is a schematic diagram of the parameter initialization process for the step-by-step temperature rise test and protection test of the target component in this invention.

[0016] Figure 3 This is a schematic diagram of the target component protection test and dynamic control process for thermal shock risk in this invention.

[0017] Figure 4 This is a schematic diagram of the iterative optimization and risk decision-making process for the thermal shock index of the target element of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 As shown, this embodiment of the invention provides a technical solution: a battery protection board life prediction and fault diagnosis system based on deep learning, including a test initialization module, a test optimization module, a judgment and prediction module, and a test database.

[0020] The test initialization module is connected to the test optimization module, the test optimization module is connected to the test prediction module, and the test database is connected to the test initialization module, the test optimization module, and the decision prediction module, respectively.

[0021] A test database is used to store parameters for a deep learning-based battery protection board life prediction and fault diagnosis system.

[0022] The test initialization module is used to mark the MOSFET components of the battery protection board as target components. The target components undergo temperature rise tests step by step, and the test parameters for the protection test of the target components are initialized based on the final results of the temperature rise tests.

[0023] Reference Figure 2As shown, the target component undergoes temperature rise testing step by step. Subsequently, the test status determination conditions of the target component are analyzed. The test status determination conditions are divided into three categories: if the further test conditions are met, the temperature rise test continues with the next test parameters; if the temperature rise test termination conditions are met, the current step is marked as the final test; if the temperature rise test termination conditions are met, the previous test is marked as the final test. Finally, regardless of how the temperature rise test ends, the process will initialize the protection test parameters of the target component.

[0024] Specifically, the target component undergoes a temperature rise test step by step. The specific test process is as follows: the target component is subjected to a temperature rise test under preset temperature rise test conditions, and the temperature and voltage of the target component are recorded in real time at the end of this step of the temperature rise test. The test status judgment conditions of the target component are analyzed.

[0025] The aforementioned preset temperature rise test conditions refer to the preset temperature rise test conditions of the circuit protection board in the test database, including the temperature rise test applied current, the temperature rise test ambient temperature, and the temperature rise test data acquisition frequency.

[0026] The test status determination conditions for the aforementioned target components include further temperature rise test conditions, temperature rise test termination conditions, and temperature rise test termination conditions.

[0027] The above-mentioned further temperature rise test conditions refer to the target component's temperature rise rate being less than the upper limit of the temperature rise rate and the target component's maximum temperature being less than the upper limit of the maximum temperature; the above-mentioned temperature rise test termination conditions refer to the target component's temperature rise rate being equal to the upper limit of the temperature rise rate and the target component's maximum temperature being equal to the upper limit of the maximum temperature; the above-mentioned temperature rise test termination conditions refer to the target component's temperature rise rate being greater than the upper limit of the temperature rise rate or the target component's maximum temperature being greater than the upper limit of the maximum temperature.

[0028] If there are further temperature rise test conditions at the end of this temperature rise test step, the target component will undergo temperature rise test under the preset next temperature rise test conditions until there are temperature rise test end conditions or temperature rise test termination conditions at the end of this temperature rise test step; if there are temperature rise test end conditions at the end of this temperature rise test step, this temperature rise test step is marked as the final test; if there are temperature rise test termination conditions at the end of this temperature rise test step, the temperature rise test performed under the preset first temperature rise test conditions is marked as the final test.

[0029] The above-mentioned upper limit of temperature rise rate is a preset value in the test database, referring to the maximum allowable temperature rise rate.

[0030] The above maximum temperature limit is a preset value in the test database, referring to the maximum allowable value of the maximum temperature.

[0031] In one specific embodiment, the testing system first retrieves the initial temperature rise test conditions for the circuit protection board from a preset database. Based on this, the system applies current to the target component and monitors its temperature and voltage data in real time. By calculating the temperature rise rate and maximum temperature, the system compares them in real time with the preset upper limits of the temperature rise rate and maximum temperature in the database. Based on the following judgment conditions, the system automatically decides the test progress: If the further test conditions are met (temperature rise rate is less than the upper limit and maximum temperature is less than the upper limit), the system automatically applies the next test conditions (such as increasing the current to 100% of the rated current) and continues iterative testing; if the test termination conditions are met (temperature rise rate tends to stabilize and maximum temperature reaches thermal equilibrium), the system marks the test result of this step as the final data and terminates the test normally; if the forced termination conditions are met (temperature rise rate exceeds the upper limit or maximum temperature exceeds the upper limit), the system immediately cuts off the current, stops the test urgently, and backtracks to mark the first step test result as valid data, ensuring test safety and data integrity.

[0032] Furthermore, the test parameters for the protection test of the target component are initialized based on the final results of the temperature rise test. The specific initialization process is as follows: the final results of the temperature rise test are the final test results, including the voltage data set of the final test, the temperature data set of the final test, the average temperature rise rate of the final test, and the duration of the end of the temperature rise of the final test.

[0033] S11. First, calculate the on-resistance of the target component based on the final result of the temperature rise test and the preset temperature rise test conditions, and establish the characteristic curve of the on-resistance changing with temperature based on this data; S12. Initialize the starting ambient temperature and load current range of the protection test based on the maximum temperature and average temperature rise rate obtained from the temperature rise test; S13. Initialize the preset value of the protection delay time based on the temperature rise duration; S14. Set the maximum allowable junction temperature rise threshold during the protection test based on the maximum temperature in the temperature rise test.

[0034] It needs to be explained that the on-resistance of the target component is calculated, and a characteristic curve of the on-resistance changing with temperature is established based on this data. The voltage data in the final test voltage data set is obtained and divided by the value of the applied current in the preset test conditions to calculate the on-resistance of the target component corresponding to each recorded voltage data. The temperature data in the final test temperature data set is obtained, and the on-resistance of the target component corresponding to each voltage data and the temperature data corresponding to each voltage data are plotted on a coordinate system with temperature as the horizontal axis and on-resistance as the vertical axis. The coordinate points are connected to establish the characteristic curve of the on-resistance changing with temperature.

[0035] It needs to be explained that, based on the maximum temperature and average temperature rise rate obtained from the temperature rise test, the initial ambient temperature and load current range of the protection test are initialized. The maximum temperature obtained from the temperature rise test is used as the initial ambient temperature of the protection test, so that the protection test can start under the known thermal boundary conditions of the component, ensuring that the test environment is consistent with the thermal limit state previously reached by the component. The temperature rise test applied current in the preset temperature rise test conditions is used as the lower limit value of the protection test load current. The ratio of the average temperature rise rate to the maximum temperature is used as the thermal load coefficient, which is then multiplied by the preset thermal coupling coefficient of the target component in the test database to calculate the allowable increment of the protection test load current. The allowable increment of the protection test load current is added to the lower limit value of the protection test load current to obtain the upper limit value of the protection test load current, thereby determining the upper limit of the load current that can effectively trigger the protection mechanism without exceeding the thermal tolerance limit of the component.

[0036] The thermal coupling coefficient of the target component preset in the aforementioned test database refers to a dimensionless proportionality coefficient determined by the test database through historical test data, used to convert the temperature rise rate increment into the corresponding allowable load current increment.

[0037] It should be explained that, based on the duration of the temperature rise, the preset value of the protection delay time is initialized. The duration of the temperature rise is multiplied by the preset protection safety factor in the test database to obtain the preset value of the protection delay time. This ensures that the protection action can be performed in time before the component is damaged by overheating, while retaining a response margin to avoid false triggering caused by instantaneous temperature fluctuations.

[0038] The preset protection safety factor in the aforementioned test database refers to the time multiplier coefficient determined by the test database based on historical test data, which is used to convert the duration of temperature rise into a protection delay time with a safety margin.

[0039] It should be explained that, based on the maximum temperature in the temperature rise test, the maximum allowable junction temperature rise threshold during the protection test is set. The maximum temperature in the temperature rise test is used as the maximum allowable junction temperature rise threshold during the protection test, thereby ensuring the test safety of the component.

[0040] In one specific embodiment, the system first acquires the final results of the temperature rise test, including voltage data set, maximum temperature, average temperature rise rate, and duration of temperature rise termination. Based on the average voltage drop of the final test and the known applied current value in the preset temperature rise test conditions, the system calculates the on-resistance of the component in its current state. It then establishes a characteristic curve of the on-resistance versus temperature by analyzing the resistance values ​​at different temperatures. Based on the measured maximum temperature and temperature rise rate, the system initializes the starting ambient temperature of the protection test to a value slightly higher than the maximum test temperature (e.g., increasing it by 10°C). Simultaneously, based on the current level that brings the component temperature close to the critical value, the system sets the test range of the load current (e.g., 90% of the rated current). Based on the thermal accumulation characteristics reflected by the end of the temperature rise test (70%-150%), the overcurrent protection delay time is initialized (e.g., 70% of the thermal equilibrium time is used as a benchmark). Based on the current withstand characteristics exhibited by the component in the temperature rise test, the precise current threshold range for the short circuit protection test is set (e.g., 800A-1200A). With the maximum temperature value of the temperature rise test as a reference, and in combination with the safety specifications of the component material, the maximum junction temperature rise threshold for the protection test is set (e.g., 20°C lower than the measured maximum temperature as a safety margin). Through this series of data-driven initialization processes, it is ensured that the subsequent protection test can effectively verify the component performance while fully guaranteeing the safety of the test process.

[0041] The test optimization module is used to perform protection tests on target components, acquire and analyze the transient stress parameters of the target components when the protection action is triggered during the protection test, so as to match the impact risk level of the target components and optimize the test parameters for the next protection test of the target components.

[0042] The aforementioned protection test refers to the overcurrent protection test, which is a reliability verification process that verifies whether the battery protection board can reliably cut off the circuit within a limited time by applying a preset fault current.

[0043] It should be explained that the protection action is triggered when the current passing through the target component exceeds the preset current safety threshold in the test database and the duration of the exceedance is greater than or equal to the preset safety duration in the test database.

[0044] The aforementioned current safety threshold refers to the upper limit of the current passing through the target component preset in the test database.

[0045] The aforementioned security duration refers to the preset protection delay duration in the test database.

[0046] Reference Figure 3As shown, the system initializes the test parameters for the protection test of the target component, then executes the protection test and calculates the thermal shock index. The system matches the corresponding impact risk level based on the thermal shock index. If it is a safe impact risk level, the current protection test of the target component is marked as a pass, and the test continues with the current test parameters. Finally, the system obtains the longest protection trigger delay time of the component and inputs the number of successful protections into the deep learning model. If it is a warning impact risk level, the test parameters for the next test are optimized, and the thermal shock index of the target component is recalculated to match the impact risk level again. If it is a dangerous impact risk level, the current protection test of the target component is marked as a fail. The system then performs fault diagnosis on the battery protection board based on deep learning, thus forming a complete closed loop of testing, evaluation, optimization, and fault handling.

[0047] Specifically, the impact risk level of the target component is matched. The specific matching process is as follows: the transient stress parameters of the target component, including the on-resistance, power and voltage of the target component when the protection is triggered.

[0048] By introducing weighted influence coefficients, the influence of the ratio of the on-resistance at protection trigger to the on-resistance at the defined protection trigger, the ratio of the power at protection trigger to the power at the defined protection trigger, and the ratio of the maximum drain-source voltage to the voltage at the defined protection trigger on the thermal shock index of the target component are quantified. These influence levels are then coupled to obtain the thermal shock index of the target component, specifically expressed as:

[0049]

[0050] In the formula, I is the thermal shock index of the target component, r_ds is the on-resistance of the target component when the protection is triggered, p_tr is the power of the target component when the protection is triggered, v_br is the voltage of the target component when the protection is triggered, r_max is the preset limit on-resistance of the protection trigger in the test database, which is the critical reference maximum value of the on-resistance of the protection trigger, p_max is the preset limit power of the protection trigger in the test database, which is the critical reference maximum value of the power of the protection trigger, v_max is the preset limit voltage of the protection trigger in the test database, which is the critical reference maximum value of the voltage of the protection trigger, w1 is the corresponding weight influence coefficient of the preset on-resistance of the protection trigger in the test database, w2 is the corresponding weight influence coefficient of the preset power of the protection trigger in the test database, and w3 is the corresponding weight influence coefficient of the preset voltage of the protection trigger in the test database.

[0051] The weighted influence coefficient corresponding to the on-resistance when the protection is triggered represents the degree of influence of the ratio between the on-resistance when the protection is triggered and the on-resistance when the protection is triggered on the thermal shock index of the target component, and the value range is (0, 1]. The weighted influence coefficient corresponding to the power when the protection is triggered represents the degree of influence of the ratio between the power when the protection is triggered and the power when the protection is triggered on the thermal shock index of the target component, and the value range is (0, 1). The weighted influence coefficient corresponding to the voltage when the protection is triggered represents the degree of influence of the ratio between the maximum drain-source voltage of the target component and the voltage when the protection is triggered on the thermal shock index of the target component, and the value range is (0, 1).

[0052] It should be explained that if the on-resistance (r_ds) is significantly higher than its critical reference maximum value (r_max) when the protection is triggered, it indicates that the component may have deteriorated resistance characteristics due to overheating or damage. If the power (p_tr) is much higher than its critical reference maximum value (p_max) when the protection is triggered, it reflects that the instantaneous thermal load of the component is extremely large, and the impact of power overload on the index is more prominent. If the voltage (v_br) ​​exceeds its critical reference maximum value (v_max) when the protection is triggered, it means that the component is subjected to overvoltage stress, which may cause dielectric breakdown. The danger of voltage over-limit is more significantly reflected in the index. Finally, the thermal shock index (I) is the result of the weighted coupling of the three factors: any abnormal exceedance of any parameter will lead to an increase in the index, and the weighting influence coefficient determines the degree of influence of each parameter. If multiple parameters exceed the limit at the same time, their effect will be superimposed through weighting, further increasing the index value, thereby comprehensively characterizing the severity of the thermal shock risk to the component. The higher the index I, the closer the state of the component is to or beyond its safety boundary when the protection is triggered, and the higher the overall risk.

[0053] If the thermal shock index of the target component falls within the thermal shock index range corresponding to the preset safety shock risk level in the test database, then the shock risk level of the target component is matched as the safety shock risk level, and the current protection test of the target component is marked as a pass. The target component continues to undergo protection testing with the current test parameters.

[0054] If the thermal shock index of the target component falls within the thermal shock index range corresponding to the preset warning impact risk level in the test database, then the impact risk level of the target component is matched as the warning impact risk level, and the test parameters for the next protection test of the target component are optimized.

[0055] If the thermal shock index of the target component falls within the thermal shock index range corresponding to the preset dangerous impact risk level in the test database, then the impact risk level of the target component is matched as the dangerous impact risk level, and the current protection test of the target component is marked as a test failure state. Fault diagnosis of the battery protection board is performed based on deep learning.

[0056] It should be explained that the three risk ranges of the thermal shock index together constitute a graded response risk assessment system. If the impact risk level of the target component is the safe impact risk level, it means that the stress on the component is far below its tolerance limit, the test parameters are conservative and reliable, so the test is continued to accumulate data. If the impact risk level of the target component is the warning impact risk level, it means that the stress has approached the safety boundary. Although it has not failed immediately, the risk is significant. Therefore, subsequent test parameters need to be dynamically optimized to accurately find the performance critical point. If the impact risk level of the target component is the dangerous impact risk level, it means that the stress has exceeded the tolerance limit and may cause damage. Therefore, the test is immediately terminated and in-depth diagnosis is initiated to analyze the root cause of failure. These three ranges are logically connected and correspond to three test states, forming a closed-loop decision-making mechanism from risk monitoring and early warning to final disposal.

[0057] In a specific embodiment, taking a certain type of MOSFET as the target component for overcurrent protection testing as an example, its thermal shock index safe range is preset to [0, 0.7), warning range to [0.7, 1.0), and danger range to [1.0, ∞). If the thermal shock index I = 0.5 is calculated after a protection test, which belongs to the safe shock risk level, the system determines that the component performs robustly under the test parameters of 20A current and 100 cycles, and automatically enters the next test with the same parameters to continuously collect durability data; if the thermal shock index I = 0.8 falls into the warning shock risk level, the system will adjust the thermal shock index according to the optimization algorithm. The next test current is slightly reduced from 20A to 18A, thereby accurately determining the actual safety boundary of the component while avoiding damage. If the thermal shock index reaches I=1.2 and enters the dangerous shock risk level, the component may suffer gate damage. The system immediately terminates the test and calls the deep learning model for analysis. Specific numerical values ​​and operational cases show that the three intervals not only reflect the differences in risk levels, but also represent a progressive test strategy from verification optimization to diagnosis. Among them, the parameter optimization for the early warning shock risk level avoids unexpected failures that jump directly from safety to danger during the test process through adaptive adjustment, demonstrating the intelligence and preventiveness of the test system.

[0058] Furthermore, the target component is continuously subjected to protection testing with the current test parameters. The specific test process is as follows: the target component is continuously subjected to protection testing with the current test parameters, and the number of successful protections of the target component is recorded; the longest protection trigger delay of the target component is obtained, and the number of successful protections of the target component is input into the deep learning model.

[0059] The number of successful protections mentioned above refers to the cumulative number of times when the protection trigger delay time of the target element is less than or equal to the upper limit of the preset protection trigger delay time in the test database.

[0060] Reference Figure 4 As shown, the system first matches the thermal shock risk index of the optimized target component with the shock risk level. If it is determined to be at a safe shock risk level, the system proceeds to the test pass state; if it is at a dangerous shock risk level, the system directly enters the fault diagnosis process; if it is at a warning shock risk level, the system will further compare the changes in the thermal shock index before and after optimization: if the thermal shock index does not decrease, the optimization is considered invalid, and the system proceeds to the test failure and fault diagnosis process; if the thermal shock index decreases effectively, the test parameters are further adjusted, and the thermal shock index is reacquired for a final judgment. If the system is marked as a test pass state, the battery protection board life prediction is initiated; otherwise, the system is marked as a test failure state, and fault diagnosis is performed. This forms a multi-level iterative, progressive intelligent decision-making and risk control closed loop.

[0061] Furthermore, the test parameters for the next protection test of the target component are optimized. The specific optimization process is as follows: based on the thermal shock index of the target component, the initial filter time constant reduction constant is matched from the test database, and the filter time constant of the protection test is initially optimized and adjusted, that is, the test parameters for the next protection test of the target component are optimized.

[0062] It needs to be explained that the process involves matching the initial filter time constant reduction constant, calling the preset initial filter time constant reduction mapping table in the test database, inputting the thermal shock index of the target component and assigning it to the corresponding thermal shock index range in the initial filter time constant reduction mapping table, and mapping and matching the corresponding initial filter time constant reduction constant.

[0063] It should be explained that the initial optimization adjustment of the filter time constant for the protection test involves subtracting the initial filter time constant reduction constant from the filter time constant for the protection test. This initial optimization adjustment of the filter time constant for the protection test reduces the protection trigger delay time, enabling the protection circuit to detect overcurrent faults and trigger actions more quickly. This shortens the duration for which components are subjected to excessive stress, helping to reduce their thermal shock risk. This quantitative adjustment method based on the thermal shock index ensures the accuracy and directionality of parameter optimization, laying a scientific foundation for further iterative optimizations that may be needed in the future.

[0064] Obtain the optimized thermal shock index of the target component and rematch it with the impact risk level of the target component; if the impact risk level of the target component is matched again as a safe impact risk level, mark the current protection test of the target component as a pass state, and the target component continues to undergo protection testing with optimized test parameters.

[0065] It should be explained that marking the target component's current protection test as passed, with its thermal shock index falling from the warning range to the safe range, proves that adjusting parameters such as the filter time constant has successfully reduced the electrical and thermal stresses experienced by the component at the moment of protection triggering, bringing the component's operating state back away from the damage risk boundary. This not only means that the optimization strategy is effective, but more importantly, it signifies that the component exhibits good tolerance under the optimized test parameters. The component can now operate stably and reliably under the current optimized test conditions. Therefore, the system can determine that the current test cycle is complete and record this parameter combination as the safe operating configuration of the component, providing a reliable benchmark for subsequent testing or applications.

[0066] If the impact risk level of the target component is matched again to the warning impact risk level, the thermal shock index of the optimized target component is compared with the thermal shock index of the original target component. If the thermal shock index of the optimized target component is less than the thermal shock index of the original target component, the test parameters of the next protection test of the target component are further adjusted.

[0067] If the thermal shock index of the optimized target component is greater than or equal to the thermal shock index of the unoptimized target component, then the current protection test of the target component is marked as a test failure, and fault diagnosis of the battery protection board is performed based on deep learning.

[0068] It should be explained that marking the current protection test of the target component as a test failure state means that if the thermal shock index does not decrease after optimization but increases or remains the same, it indicates that the parameter optimization operation has not produced the expected effect, the stress state of the component has not been improved or has even deteriorated further, the component may have hidden damage or performance degradation, and its ability to withstand stress continues to decline. At the same time, it means that under the current test conditions, the component cannot be brought back to a safe state by conventional parameter adjustment, and continuing the test will face the risk of complete component damage. The test process must be interrupted to prevent the failure from escalating. The failure state indicates that the test objective has changed from performance optimization to failure analysis, and deep learning diagnosis needs to be initiated to explore the root cause of the failure.

[0069] Furthermore, the test parameters for the next protection test of the target component are further adjusted. The specific adjustment process is as follows: based on the optimized thermal shock index of the target component, the filter time constant reduction coefficient is matched from the test database, and the filter time constant of the protection test is further adjusted, that is, the test parameters for the next protection test of the target component are further adjusted.

[0070] It needs to be explained that the filter time constant reduction coefficient is matched, the preset filter time constant reduction coefficient mapping table in the test database is called, the thermal shock index of the optimized target component is input and assigned to the corresponding thermal shock index range in the filter time constant reduction coefficient mapping table, and the corresponding filter time constant reduction coefficient is mapped out.

[0071] It should be explained that further adjustments are made to the filtering time constant of the protection test. The filtering time constant reduction factor is multiplied by the filtering time constant of the protection test to further adjust the filtering time constant of the protection test, thereby further improving the protection response speed of the target component.

[0072] The thermal shock index of the target component after further adjustment is obtained and compared with the thermal shock index range corresponding to the preset safety impact risk level in the test database. If the thermal shock index of the target component after further adjustment belongs to the thermal shock index range corresponding to the preset safety impact risk level in the test database, the current protection test of the target component is marked as a pass, and the target component continues to undergo protection testing with the further adjusted test parameters. If the thermal shock index of the target component after further adjustment does not belong to the thermal shock index range corresponding to the preset safety impact risk level in the test database, the current protection test of the target component is marked as a fail, and fault diagnosis of the battery protection board is performed based on deep learning.

[0073] The prediction module is used to obtain the protection test results of the target component and determine whether to perform life prediction on the target component. If it is determined that life prediction should be performed on the target component, the life prediction of the battery protection board is performed based on deep learning. If it is determined that life prediction should not be performed on the target component, the fault diagnosis of the battery protection board is performed based on deep learning.

[0074] Specifically, the process for determining whether to perform life prediction on the target component is as follows: if the current protection test of the target component is in a pass state, then life prediction on the target component is determined, and life prediction on the battery protection board is performed based on a deep learning model; if the current protection test of the target component is in a fail state, then life prediction on the target component is determined, and fault diagnosis on the battery protection board is performed based on a deep learning model.

[0075] Furthermore, the battery protection board is predicted based on a deep learning model. The specific prediction process is as follows: if the current protection test of the target element is in a pass state and the number of successful protections is used as the input number of protections, then the thermal shock index of the target element, the number of successful protections of the target element, the longest protection trigger delay of the target element, the filtering time constant of the target element, and the maximum temperature and temperature rise rate in the final result of the temperature rise test are used as input features and input into the deep learning model.

[0076] The aforementioned deep learning model, including the deep learning lifespan prediction model, employs a long short-term memory network. It calculates the input features using pre-set correlation coefficients to obtain associated features, analyzes the nonlinear mapping relationship between the input features and associated features and the battery protection board lifespan, and calculates the predicted remaining lifespan value of the target component.

[0077] It's important to explain that the deep learning lifetime prediction model is an intelligent prediction system based on a long short-term memory network, built and trained using a systematic data-driven approach. This model is trained using a large amount of historical test data from the same type of component. This historical test data contains complete test cycle records, and each sample includes a sequence of time-series features such as thermal shock index, number of successful protection attempts (updated only when the test passes), longest protection trigger delay, filter time constant variation, maximum temperature during temperature rise testing, and temperature rise rate. In the data preprocessing stage, the system performs missing value imputation, outlier handling, and feature normalization. Based on the training data, it estimates a pre-defined correlation coefficient matrix using statistical methods to quantify the coupling relationship between features. The model employs a dual-path architecture to process time-series features and feature relationships. The model employs a two-layer LSTM network to extract temporal patterns from feature sequences, while the associated feature branches utilize a pre-defined correlation coefficient matrix to construct weighted cross-features, explicitly capturing the coupling effect between features. The outputs of the two paths are fused and output as the predicted remaining lifespan through an MLP regression layer. During training, Huber loss is used as the main loss function, with monotonicity constraints to ensure the predicted lifespan decreases over time and temporal smoothness constraints are added. The Adam optimizer is used for parameter optimization. Training, validation, and test sets are divided according to component batches to avoid data leakage, and an early stopping strategy is employed to prevent overfitting. The trained model accurately learns the complex nonlinear mapping relationship between input features and remaining lifespan, particularly enhancing its ability to represent the coupling effect between features through correlation coefficient-weighted associated features. Finally, the model's prediction accuracy is validated on the test set, and a prediction interval can be output to quantify uncertainty, thus providing a reliable basis for predicting the remaining lifespan of battery protection board components.

[0078] It should be explained that the system performs missing value imputation, outlier handling, and feature normalization. First, to address potential data missing issues, a forward imputation method based on time-series features is used. This involves using the previous valid value to fill in subsequent missing values ​​in the time series, while simultaneously recording a missing value flag to preserve data integrity information. For outlier handling, the system uses a dynamic threshold detection mechanism, employing the moving median combined with the median absolute difference (MAD) method to identify outliers. Values ​​exceeding three standard deviations are truncated using Winsorize, replacing extreme values ​​with boundary values ​​of specified quantiles. Finally, the system uses the mean and standard deviation calculated from the training set for Z-score standardization to normalize features, ensuring that the processing method for the test set is consistent with that for the training set.

[0079] It needs to be explained that the preset correlation coefficient matrix is ​​estimated using statistical methods based on the training data. Based on the preprocessed training data, the system estimates the preset correlation coefficient matrix using statistical methods: Spearman rank correlation coefficient is used to calculate the nonlinear correlation between each feature pair. Its calculation is based on the rank relationship of the features rather than the original values, which can better capture monotonic relationships. For high-dimensional feature spaces, Ledoit-Wolf shrinking covariance estimation is applied to improve the stability of matrix estimation under small sample conditions. The final correlation coefficient matrix is ​​sparsified to retain only the statistically significant correlation coefficients, forming a fixed preset coefficient matrix for the inference stage. This matrix accurately quantifies the coupling correlation between key feature pairs such as thermal shock index and temperature rise rate, and filter time constant and protection duration, providing a reliable mathematical basis for subsequent correlation feature calculation.

[0080] In a specific embodiment, specifically in a test embodiment, when a certain type of MOSFET target device completes the current round of overcurrent protection testing and is determined to be in a "test passed state," the system uses its number of successful protections as a key input, while integrating multi-dimensional feature data: including the thermal shock index (0.32) calculated in this test, the cumulative number of successful protections (127 times), the longest protection trigger delay time in this round of testing (2.3ms), the currently used filter time constant (8.5ms), and the maximum temperature (68.2℃) and average temperature rise rate (3.5℃ / s) recorded in the temperature rise test; after these feature data undergoes standardized preprocessing, they are fed as input vectors into a trained deep learning lifetime prediction model; within the model, the coupling relationship between features is first calculated using a preset correlation coefficient matrix, which is statistically derived from historical training data, such as the thermal shock index... The correlation coefficient between the number of successful protection cycles and the temperature rise rate is 0.76, and the correlation coefficient between the number of successful protection cycles and the filtering time constant is -0.58. These coefficients are used to generate weighted correlation features to quantify the impact of different feature combinations on the lifetime. Subsequently, the time-series feature sequence is input into the LSTM network to extract the time pattern, while the correlation features are processed through a dedicated branch. Finally, the two are fused and calculated by the regression layer to output the predicted remaining lifetime value of the component as 1,850 protection cycles (or equivalent running time). This continuous value represents the expected length of the component's lifespan under the existing test parameters. The system stores this predicted value along with the feature data in the database to update the component's lifetime curve and provide a basis for subsequent maintenance decisions. The entire prediction process makes full use of the coupling correlation between features and, through the powerful nonlinear mapping capability of the deep learning model, achieves reliable prediction from multi-source test data to accurate lifetime values.

[0081] The predicted remaining lifetime values ​​mentioned above are continuous values, used to represent the expected length of time that the target component can still function normally in its current state.

[0082] The aforementioned correlation coefficients are the pre-defined correlation coefficients between specific input features in the deep learning lifetime prediction model, used to quantify the coupling correlation between specific input features.

[0083] Furthermore, fault diagnosis of the battery protection board is performed based on a deep learning model. The specific diagnosis process is as follows: if the current protection test of the target component is in a test failure state, the complete test data of the current protection test is extracted, including transient stress parameters, thermal shock index change curve, voltage and current waveform data during the protection test, and temperature rise test history. The complete test data of the protection test is then input into the deep learning model.

[0084] The aforementioned deep learning models, including deep learning fault diagnosis models, employ multi-layer convolutional neural networks combined with an attention mechanism. They identify abnormal patterns in the complete test data of protection tests through feature extraction layers, and generate fault type probability distributions through classification output layers.

[0085] It should be explained that the deep learning fault diagnosis model, an intelligent diagnostic system trained with systematic multimodal data, is based on a core architecture combining convolutional neural networks and attention mechanisms. The training data for this model comes from historically accumulated battery protection board fault cases, including high-frequency sampled waveforms from complete test cycles, spectral features after time-frequency transformation, voltage and current waveform data during protection tests, historical temperature rise test records, transient stress parameters, and thermal shock index sequences—multi-source information. Each sample is accompanied by a fault type label confirmed by failure analysis and corresponding fault component location information. In the data preprocessing stage, the system performs time synchronization and resampling on the original waveform data, uses short-time Fourier transform to generate a time-spectrum graph, and simultaneously standardizes transient parameters and fills in missing values. During training, a multi-branch network structure is used: a 1D-CNN branch extracts local temporal features of the original waveform, a 2D-CNN branch processes the frequency domain features of the time-spectrum graph, an MLP branch processes statistical features, and finally, a cross-channel attention mechanism fuses multimodal information, and a Transformer encoder captures key event segments in the time dimension. The model addresses class imbalance through a weighted cross-entropy loss function and combines it with attention-constrained loss to ensure accurate fault location. Training employs the Adam optimizer with a stepped learning rate decay strategy, and the dataset is divided according to equipment batches to avoid information leakage. After deployment, when protection tests fail, the system automatically collects multi-channel waveform data during the fault period, calculates time-spectrum features, and integrates historical parameters, inputting them into the trained diagnostic model. The model outputs the probability distribution of the fault type through forward inference, and simultaneously generates channel attention and temporal saliency heatmaps. Combined with the preset "channel-module" mapping relationship, it accurately identifies faulty components (such as drive circuits or power MOSFETs) and their occurrence time segments. Finally, it outputs a three-element diagnostic result containing fault type identification, confidence score, and location information, providing a reliable basis for maintenance decisions and system optimization.

[0086] In one specific embodiment, when the target component fails the protection function test during the final testing phase, the system immediately and automatically collects multimodal data from the testing process, including high-frequency time-series waveforms of voltage and current, transient stress parameters, thermal shock index variation curves, and historical temperature rise test records. This raw data, after preprocessing such as time synchronization, resampling, and short-time Fourier transform, is input in parallel into a multi-branch deep learning model. The 1D-CNN branch extracts local temporal features of the waveform, the 2D-CNN branch analyzes the frequency domain patterns of the time spectrum, and the MLP branch processes statistical features. Finally, multimodal information fusion and key time segment capture are achieved through a cross-channel attention mechanism and a Transformer encoder. The model output layer generates a three-element diagnostic result, including fault type identification (such as MOSFET device aging or drive circuit abnormality), fault confidence score (quantifying diagnostic reliability), and fault location information (accurate to specific components and time segments), thus providing a complete diagnostic basis for maintenance decisions that integrates fault type determination, confidence assessment, and component location.

[0087] The above output layer generates a probability distribution of fault types, including fault type identifiers, fault confidence scores, and fault location information.

[0088] The above fault type identifiers are used to indicate specific fault categories, including MOSFET device aging, drive circuit abnormality, protection chip failure, or heat dissipation system failure.

[0089] The above fault confidence score is used to indicate the reliability of the diagnostic results.

[0090] The above fault location information is used to indicate the specific module or component where the fault occurred.

[0091] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A battery protection board life prediction and fault diagnosis system based on deep learning, characterized in that, include: The test initialization module is used to mark the MOSFET components of the battery protection board as target components, perform temperature rise tests on the target components step by step, and initialize the test parameters for the protection test of the target components based on the final results of the temperature rise tests. The test optimization module is used to perform protection tests on target components, acquire and analyze the transient stress parameters of the target components when the protection action is triggered during the protection test, so as to match the impact risk level of the target components and optimize the test parameters for the next protection test of the target components. The prediction module is used to obtain the protection test results of the target component and determine whether to perform life prediction on the target component. If it is determined that life prediction should be performed on the target component, the life prediction of the battery protection board is performed based on deep learning. If it is determined that life prediction should not be performed on the target component, the fault diagnosis of the battery protection board is performed based on deep learning.

2. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 1, characterized in that: The target component is subjected to a gradual temperature rise test. The specific test process is as follows: The target component is subjected to a temperature rise test under preset temperature rise test conditions. The temperature and voltage of the target component are recorded in real time at the end of this temperature rise test, and the test status judgment conditions of the target component are analyzed. The test status determination conditions for the target component include further temperature rise test conditions, temperature rise test end conditions, and temperature rise test termination conditions. The further temperature rise test conditions refer to the temperature rise rate of the target component being less than the upper limit of the temperature rise rate and the maximum temperature of the target component being less than the upper limit of the maximum temperature. The temperature rise test termination condition refers to the temperature rise rate of the target component being equal to the upper limit of the temperature rise rate and the maximum temperature of the target component being equal to the upper limit of the maximum temperature. The termination of the temperature rise test condition refers to the temperature rise rate of the target component being greater than the upper limit of the temperature rise rate or the maximum temperature of the target component being greater than the upper limit of the maximum temperature. If there are further temperature rise test conditions at the end of this temperature rise test, the target component will be tested under the preset next temperature rise test conditions until there are temperature rise test end conditions or temperature rise test termination conditions at the end of this temperature rise test. If a temperature rise test termination condition exists at the end of this step, then mark this step as the final test; If there is a condition to terminate the temperature rise test when this temperature rise test ends, then mark the temperature rise test performed under the previous temperature rise test condition as the final test; The upper limit of the temperature rise rate is preset in the test database and refers to the maximum allowable temperature rise rate. The maximum temperature limit is a preset value in the test database, referring to the maximum allowable value of the maximum temperature.

3. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 1, characterized in that: The test parameters for the target component protection test are initialized based on the final results of the temperature rise test. The specific initialization process is as follows: The final result of the temperature rise test is the final test result, which includes the final test voltage data set, the final test temperature data set, the final test average temperature rise rate, and the final test temperature rise end time. S11. First, based on the final results of the temperature rise test and the preset temperature rise test conditions, calculate the on-resistance of the target component, and establish the characteristic curve of the on-resistance changing with temperature based on this data. S12. Based on the maximum temperature and average temperature rise rate obtained from the temperature rise test, initialize the starting ambient temperature and load current range of the protection test; S13. Initialize the preset value of the protection delay duration based on the duration of temperature rise; S14. Based on the maximum temperature in the temperature rise test, set the maximum allowable junction temperature rise threshold during the protection test.

4. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 1, characterized in that: The impact risk level of the target component is matched, and the specific matching process is as follows: The transient stress parameters of the target component include the on-resistance of the target component when the protection is triggered, the power of the target component when the protection is triggered, and the voltage of the target component when the protection is triggered. By introducing weighted influence coefficients, the influence of the ratio between the on-resistance of the target component when protection is triggered and the on-resistance when protection is triggered, the ratio between the power of the target component when protection is triggered and the power when protection is triggered, and the ratio between the maximum drain-source voltage of the target component and the voltage when protection is triggered on the thermal shock index of the target component are quantified. The influence of each degree of influence is coupled to obtain the thermal shock index of the target component. If the thermal shock index of the target component falls within the thermal shock index range corresponding to the preset safety shock risk level in the test database, then the shock risk level of the target component is matched as the safety shock risk level, and the current protection test of the target component is marked as a pass. The target component continues to undergo protection testing with the current test parameters. If the thermal shock index of the target component falls within the thermal shock index range corresponding to the preset warning impact risk level in the test database, then the impact risk level of the target component is matched as the warning impact risk level, and the test parameters for the next protection test of the target component are optimized. If the thermal shock index of the target component falls within the thermal shock index range corresponding to the preset dangerous impact risk level in the test database, then the impact risk level of the target component is matched as the dangerous impact risk level, and the current protection test of the target component is marked as a test failure state. Fault diagnosis of the battery protection board is performed based on deep learning.

5. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 4, characterized in that: The target component is continuously subjected to protection testing with the current test parameters. The specific test process is as follows: The target component is continuously subjected to protection tests with the current test parameters, and the number of successful protection attempts of the target component is recorded. Obtain the longest protection trigger delay time of the target element and input the number of successful protections of the target element into the deep learning model; The number of successful protections refers to the cumulative number of times the protection trigger delay time of the target element is less than or equal to the upper limit of the preset protection trigger delay time in the test database.

6. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 4, characterized in that: The optimization process for the test parameters of the target component for the next protection test is as follows: Based on the thermal shock index of the target component, the initial filter time constant reduction constant is matched from the test database, and the filter time constant of the protection test is initially optimized and adjusted, that is, the test parameters of the next protection test of the target component are optimized. Obtain the optimized thermal shock index of the target component and then match it with the impact risk level of the target component. If the impact risk level of the target component is matched again to the safe impact risk level, the current protection test of the target component is marked as a pass, and the target component continues to undergo protection testing with optimized test parameters. If the impact risk level of the target component is matched again to the warning impact risk level, the thermal shock index of the optimized target component is compared with the thermal shock index of the target component before optimization. If the thermal shock index of the optimized target component is less than the thermal shock index of the target component before optimization, the test parameters of the target component for the next protection test are further adjusted. If the thermal shock index of the optimized target component is greater than or equal to the thermal shock index of the unoptimized target component, then the current protection test of the target component is marked as a test failure, and fault diagnosis of the battery protection board is performed based on deep learning.

7. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 6, characterized in that: The further adjustment of the test parameters for the next protection test of the target component is as follows: Based on the optimized thermal shock index of the target component, the filter time constant reduction coefficient is matched from the test database, and the filter time constant of the protection test is further adjusted, that is, the test parameters of the next protection test of the target component are further adjusted. Obtain the thermal shock index of the target component after further adjustment, and compare it with the thermal shock index range corresponding to the preset safety shock risk level in the test database; If the thermal shock index of the target component after further adjustment falls within the thermal shock index range corresponding to the preset safety shock risk level in the test database, then the current protection test of the target component is marked as a pass, and the target component continues to undergo protection testing with the further adjusted test parameters. If the thermal shock index of the target component does not fall within the thermal shock index range corresponding to the preset safety shock risk level in the test database after further adjustment, the current protection test of the target component is marked as a test failure, and the battery protection board is diagnosed based on deep learning.

8. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 1, characterized in that: The specific determination process for whether to perform lifetime prediction on the target component is as follows: If the current protection test of the target component is in a pass state, then the lifespan prediction of the target component is determined, and the lifespan prediction of the battery protection board is performed based on a deep learning model. If the current protection test of the target component is in a failed state, it is determined that no life prediction will be performed on the target component, and fault diagnosis of the battery protection board will be performed based on a deep learning model.

9. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 8, characterized in that: The lifespan prediction of the battery protection board based on the deep learning model is specifically performed as follows: If the current protection test of the target component is passed, the thermal shock index of the target component, the number of successful protections of the target component, the longest protection trigger delay of the target component, the filtering time constant of the target component, and the maximum temperature and temperature rise rate in the final result of the temperature rise test are used as input features and input into the deep learning model. The deep learning model includes a deep learning lifetime prediction model, which uses a long short-term memory network. It calculates the input features through the preset correlation coefficients in the deep learning lifetime prediction model to obtain the associated features, analyzes the nonlinear mapping relationship between the input features and the associated features and the battery protection board lifetime, and calculates the predicted remaining lifetime value of the target component. The predicted remaining lifetime value is a continuous numerical value used to represent the expected length of time that the target component can still work normally in its current state; The correlation coefficient is a pre-defined correlation coefficient between specific input features in the deep learning lifetime prediction model, used to quantify the coupling correlation between specific input features.

10. The battery protection board life prediction and fault diagnosis system based on deep learning according to claim 8, characterized in that: The fault diagnosis process for the battery protection board based on the deep learning model is as follows: If the current protection test of the target component is in a test failure state, extract the complete test data of the current protection test, including transient stress parameters, thermal shock index change curve, voltage and current waveform data during the protection test, and temperature rise test history. Input the above complete test data of the protection test into the deep learning model. The deep learning model, including the deep learning fault diagnosis model, uses a multi-layer convolutional neural network combined with an attention mechanism. It identifies abnormal patterns in the complete test data of the protection test through the feature extraction layer and generates a fault type probability distribution through the classification output layer. The output layer generates a fault type probability distribution, including fault type identifier, fault confidence score and fault location information; The fault type identifier is used to indicate the specific fault category, including MOSFET device aging, drive circuit abnormality, protection chip failure, or heat dissipation system failure. The fault confidence score is used to indicate the reliability of the diagnostic results; The fault location information is used to indicate the specific module or component where the fault occurred.

Images