A high-temperature reliability test process adaptive control method

Abstract

The application discloses a high-temperature reliability test process self-adaptive control method, and belongs to the technical field of reliability environment test, industrial automatic control and cross technology of component failure analysis. The application is applied to the high-temperature life, high-temperature aging, high-temperature storage reliability test scene of semiconductor components, automobile electronics and aerospace products, and is characterized in that a multi-source data real-time acquisition-working condition intelligent identification-adaptive stress matching-disturbance feedforward compensation-failure prediction optimization-closed-loop feedback control whole-process closed-loop self-adaptive control architecture is constructed, a traditional open-loop fixed parameter test mode is replaced, dynamic self-adaptive adjustment and high-precision management and control of the high-temperature reliability test process are realized, test stress control precision is greatly improved, test effectiveness is significantly enhanced, failure prediction and risk control capacity are comprehensively upgraded, whole-process automation and safety guarantee are realized, and manual cost is greatly reduced.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of reliability environmental testing technology, industrial automatic control, and component failure analysis, and specifically relates to an adaptive control method for high-temperature reliability testing process. Background Technology

[0002] High-temperature reliability testing is a core step in verifying the long-term operational stability and environmental tolerance of electronic products. It is an essential procedure for product development, mass production quality inspection, and reliability certification in fields such as semiconductors, automotive electronics, and aerospace. According to industry statistics, more than 60% of electronic product failures are related to performance degradation and thermal stress failure under high-temperature environments. The accuracy and effectiveness of high-temperature reliability testing directly determine the product's lifespan and operational safety.

[0003] Traditional high-temperature reliability testing generally adopts an open-loop fixed-parameter control mode. Before the test, fixed parameters such as temperature, voltage, and test cycle are manually set according to standard requirements. During the test, the parameters are kept constant, without real-time feedback and dynamic adjustment capabilities. This has core pain points that cannot be solved in practical applications. Improvements to high-temperature test control mostly focus on optimizing a single temperature PID parameter, which can only improve the accuracy of constant temperature control but cannot solve the core problems of adaptive matching of test stress and sample characteristics, failure prediction and test strategy optimization, and multi-source disturbance compensation. A few solutions introduce simple parameter adjustment logic, but they rely on manual experience to set rules, have weak adaptive capabilities, cannot adapt to complex nonlinear scenarios with different samples and different operating conditions, and cannot achieve full-process closed-loop adaptive control.

[0004] Therefore, an adaptive control method for high-temperature reliability testing is proposed. Summary of the Invention

[0005] The purpose of this invention is to overcome the limitations of traditional high-temperature reliability tests, which generally adopt an open-loop fixed parameter control mode. Before the test, fixed parameters such as temperature, voltage, and test cycle are manually set according to standard requirements. During the test, the parameters are kept constant without real-time feedback or dynamic adjustment capabilities. Improvements to high-temperature test control are mostly focused on optimizing a single temperature PID parameter, which can only improve the accuracy of constant temperature control but cannot solve the problems of adaptive matching of test stress and sample characteristics, failure prediction and test strategy optimization, and multi-source disturbance compensation.

[0006] The technical solution adopted by this invention to solve its technical problem is: an adaptive control method for high-temperature reliability testing, applied to high-temperature life, high-temperature aging, and high-temperature storage reliability testing scenarios for semiconductor components, automotive electronics, and aerospace products. It constructs a full-process closed-loop adaptive control architecture, including real-time multi-source data acquisition, intelligent operating condition identification, adaptive stress matching, disturbance feedforward compensation, failure prediction optimization, and closed-loop feedback control, replacing the traditional open-loop fixed-parameter testing mode. This achieves dynamic adaptive adjustment and high-precision control of the high-temperature reliability testing process, specifically including the following steps:

[0007] S1 Pre-test parameter calibration and adaptive model initialization: Complete the basic parameter input and performance calibration of test samples and test equipment; based on test standards and sample characteristics, construct the sample failure mechanism model and test stress-life mapping model; and complete the initialization and pre-training of the adaptive control model.

[0008] S2 Multi-dimensional Status Data Synchronous Real-time Acquisition: Through a distributed sensor network, it synchronously acquires three categories of multi-dimensional data: environmental parameters of the test chamber, sample operating status parameters, and operating parameters of the test equipment. After filtering and noise reduction and timestamp synchronization and alignment, it outputs a standardized real-time status dataset.

[0009] S3 Test Condition Intelligent Identification and Stress Benchmark Matching: Based on real-time state datasets, the test process is divided into four major conditions through a multi-feature fusion classification model: heating and isothermal stage, steady-state test stage, cooling and recovery stage, and abnormal disturbance stage. Combined with test standards and sample failure mechanism models, the benchmark test stress parameters under the corresponding conditions are matched.

[0010] S4 Multi-source Disturbance Real-time Observation and Feedforward Compensation: Construct a linear extended state observer to observe three types of multi-source disturbances during the test process: temperature fluctuation inside the chamber, voltage fluctuation of the power grid, and sample performance drift. Generate disturbance feedforward compensation amount, which is superimposed on the reference test stress parameters to eliminate the influence of disturbance on the test stress accuracy.

[0011] S5 is based on an improved fuzzy neural network for adaptive dynamic adjustment of test parameters: taking real-time state dataset, reference stress parameters, and disturbance compensation as inputs, it dynamically optimizes and adjusts four core test parameters—test temperature, power supply voltage, test cycle, and load parameters—through an improved fuzzy neural network adaptive control model, and outputs the optimal real-time control command.

[0012] S6 Sample Failure Prediction and Intelligent Optimization of Test Strategies: Based on the real-time operating status parameters of the sample, the time-series prediction model predicts the performance degradation trend and failure risk of the sample, and intelligently optimizes the test strategy based on the failure prediction results: For qualified samples with stable performance, the test cycle and stress intensity are optimized to improve test efficiency; for samples with early failure, an early warning is triggered and the test strategy is adjusted to avoid overstress damage to the sample; for samples that have failed, the test is automatically terminated and failure data is recorded.

[0013] S7 Closed-Loop Feedback Control and Safety Interlock Protection: Optimized control commands are sent to the high-temperature test chamber, power supply, and load equipment for execution. The actual status data after execution is collected synchronously and compared with the target value. Closed-loop feedback correction is completed through an adaptive PID controller. At the same time, a multi-level safety interlock protection mechanism is set up to immediately trigger emergency protection actions when over-temperature, over-pressure, or sample hard failure occurs.

[0014] S8 Test Termination and Full-Process Data Archiving: When the test termination conditions are met, the gradient cooling procedure is automatically executed. After the test is completed, a standardized test report is generated, and the full-process test data, control parameters, and failure records are archived simultaneously. The training dataset of the adaptive control model is updated simultaneously to complete the model iterative optimization.

[0015] Preferably, the test samples in step S1 include semiconductor chips, discrete devices, automotive electronic controllers, and aerospace airborne products; the test standards include industry standards such as JEDEC, AEC-Q100, GJB128, and GJB548; the sample failure mechanism models include thermo-oxidative aging failure models, electromigration failure models, and thermal stress fatigue failure models; the adaptive control model pre-training is completed using historical test datasets of similar samples, and the stress control error of the pre-trained model is ≤±0.5%.

[0016] Preferably, the acquisition frequency of the multi-dimensional state data in step S2 is 1Hz~100Hz, and the specific classification and acquisition requirements are as follows:

[0017] Test chamber environmental parameters: including temperature, humidity, heating rate, cooling rate, and temperature uniformity of the test chamber working area, which are collected using a high-precision platinum resistance temperature sensor with a measurement accuracy of ≤±0.2℃;

[0018] Sample operating status parameters: including sample operating current, operating voltage, leakage current, output power, junction temperature, and key pin impedance, are collected using a high-precision digital multimeter, current sensor, and infrared thermometer. Voltage measurement accuracy is ≤ ±0.1%FS, and current measurement accuracy is ≤ ±0.5%FS.

[0019] Test equipment operating parameters: including the output power of the test chamber heating / cooling module, the output voltage / current of the power supply, the operating status of the load equipment, the acquisition frequency is synchronized with the sample parameters, and the data timestamp alignment error is guaranteed to be ≤10ms.

[0020] Preferably, the multi-feature fusion classification model in step S3 adopts the random forest classification algorithm, using temperature change rate, sample current fluctuation rate, and test run time as input features, with a working condition identification accuracy of ≥99.5%; the matching rules for the benchmark stress parameters of the four working conditions are as follows:

[0021] Heating and isothermal stage: A gradient heating strategy is adopted, and the heating rate is adaptively matched to the thermal expansion coefficient of the sample to avoid sample damage caused by thermal shock.

[0022] Steady-state test phase: Based on the rated temperature and rated voltage specified in the test standard, dynamic fine-tuning is performed in combination with the real-time state of the sample;

[0023] Cooling and recovery phase: A gradient cooling strategy is adopted, with the cooling rate adaptively matched to the thermal stress tolerance threshold of the sample;

[0024] Abnormal disturbance stage: Lock the test stress reference, prioritize the completion of disturbance compensation, and resume normal test procedures after the state stabilizes.

[0025] Preferably, the linear expansion state observer described in step S4 treats the unmodeled dynamics and unknown disturbances during the test as a total disturbance, observes them in real time and performs feedforward compensation, sets the observer bandwidth to 10 times the system bandwidth, the disturbance observation response time is ≤0.5s, the test temperature control accuracy after compensation is ≤±0.3℃, the power supply voltage control accuracy is ≤±0.5%FS, and the test stress deviation caused by the disturbance is completely eliminated.

[0026] Preferably, the improved fuzzy neural network adaptive control model described in step S5 has a 5-layer structure, consisting of an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer, and an output layer. The model takes the experimental temperature deviation, voltage deviation, temperature change rate, and sample leakage current change rate as inputs, and the adjustment amounts of experimental temperature, power supply voltage, load parameters, and experimental cycle as outputs. The model uses a genetic algorithm to optimize the membership function and fuzzy rules, solving the problem that traditional fuzzy control rules rely on human experience. The control command response time is ≤1s, and there is no steady-state error in steady-state control.

[0027] Preferably, the time-series prediction model in step S6 is an improved LSTM (Long Short-Term Memory) network, which uses the historical operating status time-series data of the sample as input to predict the performance degradation trend and failure probability of the sample within the next 24 hours, with a failure prediction accuracy of ≥98%; the intelligent optimization of the experimental strategy specifically includes:

[0028] When the predicted failure probability of the sample is <1%, within the allowable range of the test standard, the stress intensity and test cycle can be optimized, which can shorten the test time by up to 30%.

[0029] When the predicted failure probability of a sample is 1% to 10%, a yellow warning is triggered, the test stress intensity is reduced, the data acquisition frequency is increased, and performance changes are continuously monitored.

[0030] When the predicted failure probability of a sample is greater than 10%, a red alert is triggered, the test is automatically paused, and abnormal information is pushed to the test personnel to avoid damage to the sample due to excessive stress.

[0031] When a hard failure is detected in the sample, immediately cut off the power supply to the sample, terminate the single-circuit test, and record the failure time and failure status data.

[0032] Preferably, the adaptive PID controller described in step S7 uses fuzzy rules to adaptively adjust the proportional, integral, and derivative parameters of the PID controller online, replacing the traditional fixed-parameter PID controller. This solves the control problems of large inertia, large lag, and nonlinearity in high-temperature testing processes, with a temperature control overshoot ≤1℃ and a settling time ≤300s. The multi-level safety interlock protection mechanism is divided into three levels:

[0033] Level 1 warning: When the parameter deviation is ≤5%, a pop-up warning will appear on the interface, and the control parameters will be automatically adjusted.

[0034] Level 2 protection: If the parameter exceeds the tolerance by 5%~10%, the test stress is locked, the sample load is cut off, and an SMS alarm is sent.

[0035] Level 3 Emergency Shutdown: If parameters exceed tolerance by more than 10%, or if a sample short circuit or test chamber overheating occurs, immediately disconnect the heating power supply and sample power supply, execute gradient cooling, and trigger an audible and visual alarm.

[0036] Preferably, this method supports multi-channel independent parallel control, which can simultaneously perform independent adaptive control on up to 64 different samples. The test parameters and control strategies of each sample are configured independently and do not interfere with each other. At the same time, it supports the standard communication protocol for docking with test equipment, realizing seamless docking with high temperature test chambers, programmable power supplies and electronic loads without the need for hardware modification of existing equipment.

[0037] The advantages of this invention are:

[0038] 1. Significantly improved test stress control precision and enhanced test effectiveness: Completely eliminates test stress deviations caused by disturbances, ensuring that test stress is precisely matched with standard requirements and sample characteristics, avoiding test invalidity due to insufficient stress and sample damage due to excessive stress, greatly improving test efficiency and significantly shortening the test cycle: Through failure prediction and intelligent optimization of test strategies, the test cycle for stable and qualified samples can be shortened by up to 30%, greatly improving the efficiency of product development and mass production quality inspection, and reducing the consumption of test equipment and energy.

[0039] 2. Comprehensive upgrade in failure prediction and risk management capabilities: Through the LSTM time-series prediction model, sample failure risks can be predicted more than 12 hours in advance, solving the problem of lagging failure identification in traditional tests; providing key data support for product failure analysis and reliability improvement, improving product failure analysis efficiency by more than 60%, strong adaptive control capability, and wide adaptability: solving the problems of traditional control methods relying on manual experience and poor adaptability; supporting up to 64 parallel independent control channels of samples, which can simultaneously adapt to the test requirements of different types of samples, and improving batch test consistency from the traditional 80% to more than 99.5%.

[0040] 3. Full-process automation and safety assurance, significantly reducing labor costs: Completely eliminates safety hazards in the testing process, ensuring the safety of equipment, samples and personnel, with strong standardization and traceability, meeting industry certification requirements: Perfectly adaptable to the stringent testing requirements of semiconductor, automotive electronics, aerospace and other fields, and has extremely strong industrialization and promotion value. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a block diagram of the closed-loop architecture of the adaptive control method for high-temperature reliability testing in this invention.

[0043] Figure 2 This is an architecture diagram of the improved fuzzy neural network adaptive control model of the present invention;

[0044] Figure 3 This is the control block diagram for multi-source disturbance observation and feedforward compensation in this invention;

[0045] Figure 4 This is a logic block diagram for sample failure prediction and test strategy optimization in this invention;

[0046] Figure 5 This is a flowchart illustrating the triggering process of the multi-level safety interlocking protection mechanism of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The adaptive control method for high-temperature reliability testing of this invention is based on a closed-loop architecture encompassing multi-source data acquisition, operating condition identification, stress matching, disturbance compensation, parameter optimization, strategy adjustment, and closed-loop control. The detailed technical solutions for the eight core steps are as follows:

[0049] S1 Pre-experiment parameter calibration and adaptive model initialization

[0050] This step provides a benchmark and model foundation for the experiment, ensuring that the adaptive control system is accurately adapted to the test object and test standards. The specific execution process is as follows:

[0051] 1. Basic Information Input: Input the basic parameters of the test sample, including sample type, model and specifications, rated operating voltage, rated junction temperature, thermal resistance, failure mechanism type, and stress threshold; input the test requirements, including the execution standard, total test duration, rated test temperature, rated power supply voltage, and pass / fail criteria; input the test equipment parameters, including the temperature control range, heating and cooling rate, and temperature control accuracy of the high-temperature test chamber, and the range, accuracy, and communication protocol of the programmable power supply and electronic load.

[0052] 2. Equipment and Sample Performance Calibration: Before the test, complete the metrological calibration of the test equipment, and perform zero-point calibration and range calibration of the temperature sensor and voltage and current sensor to ensure that the measurement accuracy meets the test requirements; perform initial performance testing on the test sample, and record the initial operating current, leakage current, output characteristics, insulation impedance and other benchmark parameters of the sample as the benchmark for subsequent performance degradation judgment.

[0053] 3. Failure Mechanism and Stress-Life Model Construction: Based on sample type and failure mode, corresponding failure mechanism models are constructed, including electromigration failure model, thermo-oxidative aging failure model, and thermal stress fatigue failure model of device packaging. Based on the Arrhenius accelerated life equation, a test stress-sample life mapping model is constructed to quantify the correlation between test temperature, voltage stress and sample life and degradation rate, providing a theoretical basis for test stress matching.

[0054] 4. Adaptive control model initialization: Using historical test datasets of similar samples, the improved fuzzy neural network control model and LSTM failure prediction model are pre-trained to optimize the model's weight parameters and fuzzy rules. Based on test standards and sample characteristics, the model's input and output boundaries, control parameter thresholds, and safety protection thresholds are set to complete the model initialization, ensuring that the stress control error after model pre-training is ≤ ±0.5% and the failure prediction accuracy is ≥ 95%.

[0055] 5. Test channel configuration: For multi-sample parallel testing scenarios, test parameters, control strategies, and pass / fail criteria are independently configured for each sample, and independent data acquisition channels and control loops are set up to ensure that multi-channel tests do not interfere with each other.

[0056] S2 Multi-Dimensional Status Data Synchronous Real-Time Acquisition

[0057] This step provides real-time data input for adaptive control, constructs a distributed sensor network, and achieves synchronized and high-precision acquisition of full-dimensional state data. The specific execution flow is as follows:

[0058] 1. Distributed sensor network setup: Deploy at least three high-precision platinum resistance temperature sensors in the working area of ​​the test chamber to collect temperature distribution data within the chamber; configure a high-precision Hall current sensor and voltage sampling module in each sample power supply circuit to collect the sample's operating voltage, operating current, and leakage current in real time; configure status acquisition modules at the test chamber's heating / cooling module, power supply, and load terminals to obtain equipment operating parameters; use a non-contact infrared thermometer to collect the sample surface temperature and calculate the sample junction temperature.

[0059] 2. Synchronous acquisition and timestamp alignment: Set the data acquisition frequency. The basic acquisition frequency is 1Hz during the steady-state test phase. During the heating / cooling phase and under abnormal conditions, the acquisition frequency is automatically increased to 100Hz. A unified clock synchronization module is used to add high-precision timestamps to all acquired data to ensure that the time alignment error of environmental parameters, sample parameters, and equipment parameters is ≤10ms, avoiding control deviations caused by data timing misalignment.

[0060] 3. Data Preprocessing: The raw collected data is preprocessed by using moving average filtering to remove impulse noise, amplitude limiting filtering to remove outliers, and linear interpolation to fill in lost data. Parameters of different dimensions are normalized and mapped to the [0,1] interval to output a standardized real-time status dataset, providing high-quality data input for subsequent working condition identification and model calculation.

[0061] 4. Real-time data caching and backup: The collected raw data and preprocessed data are stored in the local edge database in real time, and off-site backup is also performed. The data storage cycle is no less than 3 times the test cycle to ensure that the data in the entire test process is traceable and tamper-proof, and meets the industry standard requirements for archiving test data.

[0062] S3 test condition intelligent identification and stress benchmark matching

[0063] This step, based on real-time status data, accurately identifies different operating conditions during the test, matches corresponding benchmark stress parameters, and provides a benchmark target for adaptive control. The specific execution process is as follows:

[0064] 1. Operating condition feature extraction: Extract core feature parameters from the real-time status dataset, including the test chamber temperature change rate, the deviation between the set temperature and the actual temperature, the sample operating current fluctuation rate, the leakage current change rate, the cumulative test running time, and the temperature rise and fall stage identifiers, and construct an operating condition identification feature vector.

[0065] 2. Multi-feature fusion for work condition classification: A pre-trained random forest classification model is used, with feature vectors as input, to accurately divide the experimental process into four major work conditions. The work condition identification accuracy is ≥99.5%. The definitions and characteristics of the four major work conditions are as follows:

[0066] Heating and holding phase: After the test is started, the temperature rises from room temperature to the rated test temperature and then enters the holding phase. The core characteristic is that the temperature change rate is >0 and the deviation between the actual temperature and the set value gradually decreases.

[0067] Steady-state test stage: The core test process in which the temperature stabilizes at the rated test temperature and the sample is in rated working condition. The core characteristics are temperature fluctuation ≤ ±0.5℃, continuous increase in cumulative test duration, and stable degradation of sample performance.

[0068] Cooling recovery phase: When the test ends or is abnormally paused, the process of cooling down from the rated temperature gradient to room temperature. The core characteristic is that the temperature change rate is <0 and the power supply to the sample is gradually cut off.

[0069] Abnormal disturbance stage: The process in which the temperature fluctuates greatly, the voltage changes abruptly, and the sample performance changes abnormally during the test. The core characteristic is that the parameter deviation exceeds the set threshold and the sample is in an unstable operating state.

[0070] 3. Benchmark stress matching for operating conditions: Based on the identified operating condition type, combined with the test execution standard and sample failure mechanism model, the benchmark test stress parameters under the corresponding operating condition are matched. The core matching rules are as follows:

[0071] Heating and isothermal stage: A gradient heating strategy is adopted, and the heating rate is adaptively matched with the thermal expansion coefficient and thermal stress tolerance threshold of the sample to avoid thermal shock damage to the sample caused by rapid heating; the isothermal platform is set in stages to eliminate the internal temperature gradient of the sample and ensure that the sample enters steady-state test after reaching the rated test temperature.

[0072] Steady-state test phase: Based on the rated temperature, rated voltage, and rated load specified in the test standard, and combined with the real-time performance degradation state of the sample, an adaptive adjustment boundary for stress parameters is set to ensure that the adjustment range meets the requirements of the test standard.

[0073] Cooling and recovery phase: A gradient cooling strategy is adopted, with the cooling rate adaptively matched to the thermal stress tolerance threshold of the sample to avoid encapsulation cracking and bonding wire detachment failure caused by rapid cooling; a constant temperature platform is set in stages to eliminate internal thermal stress of the sample, and all power supply is cut off after cooling to room temperature.

[0074] Abnormal disturbance phase: Lock the benchmark test stress parameters, suspend adaptive adjustment, prioritize disturbance compensation and anomaly investigation, and re-enter the normal control process for the corresponding working condition after the system state returns to stability.

[0075] S4 Multi-source Disturbance Real-time Observation and Feedforward Compensation

[0076] This step addresses multi-source disturbances during high-temperature testing by constructing a linear extended state observer to achieve real-time observation and feedforward compensation of the disturbances, thereby eliminating the impact of disturbances on the accuracy of the test stress. The specific execution process is as follows:

[0077] 1. Classification and Characteristic Analysis of Multi-Source Disturbances: Disturbances during high-temperature testing are classified into three main categories, and the characteristics of each type of disturbance and its impact on the test are clarified:

[0078] Environmental and equipment disturbances include power grid voltage fluctuations, ambient temperature changes, temperature shocks caused by opening and closing the test chamber door, and performance degradation of the heating / cooling modules. These mainly affect the stability of the test temperature and power supply voltage and have the characteristics of large inertia and slow time-varying.

[0079] Sample characteristic disturbances: These include operating current drift, leakage current increase, and junction temperature change caused by sample performance degradation. These are endogenous disturbances in the experimental process and have nonlinear and strongly coupled characteristics.

[0080] Unmodeled dynamic disturbances: These include disturbances that cannot be accurately modeled, such as the nonlinearity, hysteresis characteristics, and sensor measurement noise of the experimental system. They are characterized by randomness and rapid time-varying nature.

[0081] 2. Linear Extended State Observer Design: All disturbances and unmodeled dynamics mentioned above are considered as the total disturbance of the system. A third-order linear extended state observer is constructed to observe the total disturbance in real time. The observer takes the test temperature setpoint, control command output, and actual temperature feedback as inputs, and outputs the system state observation value and the total disturbance observation value. The observer bandwidth is set to 10 times the system bandwidth to ensure the speed and accuracy of disturbance observation; the disturbance observation response time is ≤0.5s.

[0082] 3. Disturbance Feedforward Compensation Calculation: Based on the total disturbance observation value output by the observer, a corresponding feedforward compensation amount is generated and superimposed on the control command of the reference test stress parameters to proactively offset the impact of disturbances on the system output. For the temperature control loop, the compensation amount is superimposed on the power output command of the heating / cooling module; for the voltage control loop, the compensation amount is superimposed on the output voltage command of the programmable power supply, achieving active suppression of disturbances rather than passive feedback correction.

[0083] 4. Compensation effect verification and closed-loop optimization: Real-time acquisition of actual stress parameters after compensation, and comparison with the benchmark value to verify the disturbance compensation effect; when the deviation after compensation still exceeds the threshold, the bandwidth and gain parameters of the observer are automatically optimized to improve the disturbance observation accuracy, ensuring that the test temperature control accuracy after compensation is ≤±0.3℃ and the power supply voltage control accuracy is ≤±0.5%FS, and completely eliminating the test stress deviation caused by multi-source disturbances.

[0084] S5 is based on an improved fuzzy neural network for adaptive dynamic adjustment of test parameters.

[0085] This step is the core control component of this invention. Through an improved fuzzy neural network model, adaptive dynamic optimization of experimental parameters is achieved, solving the problem that traditional control methods cannot adapt to nonlinear and large time-delay systems. The specific execution flow is as follows:

[0086] 1. Improved Fuzzy Neural Network Model Architecture Design: The model adopts a 5-layer feedforward network structure to solve the problems of traditional fuzzy control rules relying on human experience and having poor adaptability, while retaining the nonlinear fitting ability of neural networks and the robustness of fuzzy control. The structure and function of each layer are as follows:

[0087] Input layer: There are 4 input nodes in total. The input parameters are: deviation between the actual value and the set value of the test temperature, temperature change rate, sample working voltage deviation, and sample leakage current change rate. The input parameters are normalized and then passed to the next layer.

[0088] Fuzzification layer: Each input node corresponds to 7 fuzzy subsets. The Gaussian membership function is used to fuzzify the input parameters. The output is the membership degree of each input parameter corresponding to each fuzzy subset. The parameters of the membership function can be automatically optimized through learning.

[0089] Fuzzy Inference Layer: Each node corresponds to a fuzzy rule, performs the "AND" operation of fuzzy logic, outputs the applicability of each fuzzy rule, realizes the fuzzy inference process, does not require manual pre-setting of fuzzy rules, and can automatically generate the optimal rule base through data learning;

[0090] Normalization layer: Normalizes the output of the fuzzy inference layer to eliminate the magnitude difference in the applicability of different rules and improve the stability of the model;

[0091] Output layer: There are 4 output nodes, which output the test temperature adjustment, power supply voltage adjustment, load parameter adjustment, and test cycle optimization coefficient respectively. The output values ​​are inversely normalized to generate the final control command.

[0092] 2. Model Learning and Optimization: A genetic algorithm is used to globally optimize the membership function parameters and fuzzy rule weights of the model, with the optimization objectives of minimizing control deviation and maximizing response speed, thus avoiding the model from getting trapped in local optima. Real-time data during the experiment is used for online learning, and the model parameters are fine-tuned every 10 minutes to ensure that the model continuously adapts to changes in system characteristics and sample performance degradation during the experiment, ensuring stable control accuracy throughout the process.

[0093] 3. Adaptive Optimization Logic for Experimental Parameters: The control commands output by the model, based on the real-time state of the sample and the system operating conditions, achieve dynamic optimization of the four core experimental parameters:

[0094] Test temperature optimization: Within the range allowed by the test standard, the test chamber temperature is finely adjusted based on the sample junction temperature and leakage current change rate to ensure that the actual junction temperature of the sample is stable at the rated value and to avoid deviation of the actual stress of the sample caused by ambient temperature fluctuations; when the sample shows signs of early degradation, the temperature stress is slightly reduced within the range allowed by the standard to avoid accelerated sample failure.

[0095] Power supply voltage optimization: Based on the drift of the sample's operating current, the power supply voltage is dynamically fine-tuned to ensure that the sample operates at its rated power, eliminating the deviation in actual operating voltage caused by voltage drop in the power supply line and changes in the sample's internal resistance; for samples with abnormally increased leakage current, the power supply voltage is slightly reduced to avoid short-circuit failure.

[0096] Load parameter optimization: Based on the changes in sample output characteristics, the parameters of the electronic load are dynamically adjusted to ensure that the sample works under rated load conditions, simulate the load changes in real use scenarios, and improve the effectiveness of the test;

[0097] Test cycle optimization: Based on the sample performance degradation rate, the test cycle optimization coefficient is output. For samples with stable performance and degradation rate far below expectations, the test cycle is optimized under the premise of meeting the test standard requirements, which can shorten the test time by up to 30%.

[0098] 4. Control command boundary constraints: To ensure that the test meets the standard requirements, hard boundary constraints are set for the control commands output by the model. The temperature adjustment range shall not exceed ±5℃ of the standard rated value, and the voltage adjustment range shall not exceed ±10% of the rated value. This ensures that all adjustments are within the range allowed by the test standard and avoids invalid test results due to parameters exceeding the range.

[0099] S6 Sample Failure Prediction and Intelligent Optimization of Testing Strategies

[0100] This step uses a time-series prediction model to predict sample performance degradation trends and provide early warnings of failure risks. Based on the prediction results, it intelligently optimizes the testing strategy, solving the problems of lagging and inefficient failure identification in traditional testing. The specific execution process is as follows:

[0101] 1. Construction of sample performance degradation time series dataset: Based on the initial performance parameters of the sample, the relative change rate of parameters such as sample operating current, leakage current, output characteristics, and junction temperature collected in real time is calculated to construct a performance degradation time series dataset with equal time intervals. The data sequence length covers the experimental data of the past 72 hours, and the sampling interval is 1 minute.

[0102] 2. Improved LSTM Failure Prediction Model Design: A two-layer LSTM (Long Short-Term Memory) network time-series prediction model is constructed. Using historical performance degradation time-series datasets as input, it predicts the performance degradation trend of samples within the next 24 hours and outputs the failure probability of the samples. An attention mechanism is introduced to enhance the ability to capture abrupt changes in degradation trends, and a Dropout layer is used to suppress overfitting. After pre-training, the model achieves a failure prediction accuracy of ≥98% and an early warning time of ≥12 hours.

[0103] 3. Failure Risk Classification and Test Strategy Optimization: Based on the failure probability output by the model, the sample risk is divided into three levels, and corresponding intelligent test strategies are matched to achieve differentiated management.

[0104] Low risk (failure probability < 1%): The sample performance is stable, and the degradation rate meets expectations, thus it is judged as a qualified sample. Within the allowable range of the test standard, the test stress and cycle are optimized based on the stress-life model to shorten the test time and improve the test efficiency; at the same time, the normal data acquisition frequency is maintained to continuously monitor performance changes.

[0105] Medium risk (failure probability 1%~10%): The sample shows early signs of degradation and has a potential failure risk. A yellow alert is triggered, an abnormal information is displayed in a pop-up window on the interface, the test stress intensity is automatically reduced, the data acquisition frequency is increased to 10Hz, and the monitoring of sample performance changes is intensified; at the same time, abnormal data is recorded to provide a basis for subsequent failure analysis and to avoid data loss due to unexpected sample failure.

[0106] High risk (failure probability > 10%): The sample shows a clear trend of degradation and the risk of failure is extremely high. A red alert is triggered, automatically pausing the test, cutting off the sample load, and sending an SMS alarm to the test personnel. The test will continue only after manual confirmation. This prevents complete damage to the sample due to overstress, preserving the sample's pre-failure state and providing a complete sample for failure analysis.

[0107] 4. Automatic handling of failed samples: When a hard failure (short circuit, open circuit, leakage current exceeding the standard by more than 10 times) is detected in real time, emergency handling is immediately performed: the power supply to the sample is cut off, the single-channel test is terminated, and the failure time, status data at the time of failure, and performance changes before failure are recorded; at the same time, the normal test of other channels is not affected, so as to avoid the failure of a single sample affecting the progress of the entire batch of tests.

[0108] S7 Closed-Loop Feedback Control and Safety Interlock Protection

[0109] This step utilizes an adaptive PID controller to achieve closed-loop feedback correction, while simultaneously setting up a multi-level safety interlock protection mechanism to ensure control accuracy and operational safety during the experiment. The specific execution flow is as follows:

[0110] 1. Adaptive PID closed-loop feedback control: The optimized control command output by the fuzzy neural network model is used as the set target value and sent to the test chamber, power supply, and load equipment for execution; the actual state parameters of the equipment after execution are collected synchronously, and the deviation from the target value is calculated. The closed-loop feedback correction is completed through the adaptive PID controller to eliminate the deviation between the model output and the actual execution.

[0111] The adaptive PID controller employs fuzzy rules to adjust the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd online: when the deviation is large, Kp is increased to improve the response speed, and Ki is decreased to avoid integral saturation; when the deviation is small, Ki is increased to eliminate steady-state error, and Kd is adjusted to suppress overshoot. This solves the problem that traditional fixed-parameter PID controllers cannot adapt to the large inertia, large hysteresis, and nonlinearity of high-temperature test systems, achieving temperature control overshoot ≤1℃, settling time ≤300s, and steady-state control with zero steady-state error.

[0112] 2. Multi-level safety interlock protection mechanism: Based on the degree of deviation of test parameters and the level of fault risk, a three-level safety interlock protection is set up to achieve full-scenario fault coverage and ensure the safety of the test and the sample.

[0113] Level 1 Warning: Parameter deviation ≤5%, no safety risk, only affects test accuracy. The system automatically pops up a warning window, adjusts parameters through adaptive control to eliminate deviation, and records the warning information without manual intervention.

[0114] Level 2 protection: When parameters exceed tolerance by 5%~10%, a potential safety risk exists. The system immediately locks the test stress parameters, cuts off the sample load, pauses the test process, and sends an SMS alarm to the test personnel. The test can continue after the fault is identified and the status is restored, preventing the fault from escalating.

[0115] Level 3 Emergency Shutdown: Parameter deviations exceeding 10%, or serious malfunctions such as chamber overheating, sample short circuit, power overcurrent, or sensor failure. The system immediately executes the emergency shutdown procedure: disconnecting the heating power supply and all sample power supply, shutting down the cooling module, triggering audible and visual alarms, sending emergency alarm messages, and simultaneously executing a gradient cooling program to prevent sample damage due to sudden heating or cooling, ensuring equipment and personnel safety.

[0116] 3. Control Access and Emergency Manual Takeover: The system has three levels of access permissions. Administrators can configure parameters, modify models, and perform emergency shutdowns; operators can start and stop tests and view parameters; visitors can only view data. When automatic control malfunctions, test personnel can manually take over control at any time and switch to manual mode to ensure test controllability in any scenario.

[0117] S8 Trial Termination and Full Process Data Archiving

[0118] This step completes the standardized conclusion of the experiment and data archiving, while simultaneously achieving iterative optimization of the adaptive model, forming a complete closed-loop management. The specific execution flow is as follows:

[0119] 1. Test Termination Condition Determination: The system determines the test termination conditions in real time and automatically triggers the test termination procedure when any of the following conditions are met: ① The preset total test duration is reached and the performance of all samples meets the qualification standards; ② All samples have failed and no further testing is required; ③ A serious fault occurs and the test cannot continue safely; ④ The test personnel manually issue a termination command.

[0120] 2. Standardized test termination procedure: After the test is terminated, the system automatically executes the gradient cooling program, lowering the test chamber temperature to room temperature according to the preset cooling rate, and continuously monitoring the sample status during the cooling process; after the temperature stabilizes, the power supply to the sample and the equipment are cut off in sequence to complete the test termination; at the same time, all test samples are finally tested, the performance parameters after the test are recorded, compared with the initial parameters, the performance degradation is calculated, and the pass / fail determination is completed.

[0121] 3. Automatic generation of standardized test reports: The system automatically generates standardized test reports that conform to industry standards. The report content includes: basic test information, implementation standards, sample information, equipment information, test process parameter curves, sample performance change data, failure records, pass / fail judgment results, and full-process abnormality warning records. The report supports export in PDF and Excel formats and generates an encrypted QR code. Scanning the code allows for the traceability of complete original test data, meeting the traceability requirements of reliability certification.

[0122] 4. Full-process data archiving and model iteration: All raw data collected throughout the entire test process, control parameters, model outputs, early warning records, failure data, and test reports are archived into the database and stored according to test batches and sample types. The data is tamper-proof and permanently traceable. At the same time, the effective data from this test is added to the model training dataset to incrementally train and iteratively optimize the adaptive control model and failure prediction model, continuously improving the control accuracy and prediction accuracy of the model, and realizing the digital accumulation and reuse of test experience.

[0123] Example 1: Application of HTOL High-Temperature Operating Life Test for Automotive-Grade Semiconductor Chips

[0124] 1. Application Scenarios and Testing Requirements

[0125] This embodiment is applied to the HTOL high-temperature operating life test of automotive-grade MCU chips for mass production by automotive electronic chip manufacturers. The test follows the AEC-Q100 standard, with a rated test temperature of 125°C, a rated operating voltage of 5V, and a total test duration of 1000 hours. Each batch of test samples contains 256 chips, divided into 4 test channels, with 64 samples per channel. Traditional open-loop testing methods suffer from drawbacks such as poor temperature control accuracy, inability to detect sample failures in a timely manner, fixed test cycles with low efficiency, and high manual monitoring costs. Therefore, it is necessary to achieve adaptive closed-loop control of the test process to improve test accuracy and efficiency and meet the traceability requirements of automotive-grade certification.

[0126] 2. System Deployment and Pre-Test Preparation

[0127] This embodiment deploys the high-temperature reliability test adaptive control system of the present invention, which connects to 4 high and low temperature test chambers, 16-channel programmable power supply, and 32-channel electronic load. Each sample is equipped with an independent voltage and current acquisition module. Five high-precision platinum resistance temperature sensors are arranged in the test chamber, with an acquisition frequency of 1Hz, which is automatically increased to 100Hz in abnormal conditions. The system is equipped with an i.MX8MPlus edge computing unit to run the adaptive control model and failure prediction model to achieve local real-time control.

[0128] Before the test, complete the following preparations: input the rated parameters of the MCU chip, the failure mechanism model (electromigration, thermo-oxidative aging), and the AEC-Q100 test standard requirements; complete the metrological calibration of the temperature sensor and power module, test the initial performance parameters of the sample, and record the initial leakage current and operating current reference values; use the historical HTOL test dataset of the same model chip to complete the pre-training of the adaptive control model and the LSTM failure prediction model; configure the test parameters independently for the four test channels and complete the model initialization.

[0129] 3. Experiment Execution and Adaptive Control Flow

[0130] Heating and isothermal stage: The system identifies the heating and isothermal operation as a working condition and adopts a gradient heating strategy to heat the sample to 80℃ at a rate of 5℃ / min, hold the temperature for 30 minutes to eliminate thermal stress, and then heat the sample to 125℃ at a rate of 3℃ / min, hold the temperature for 60 minutes to ensure that the sample reaches the rated temperature. During the process, the temperature fluctuation during the heating process is compensated in real time by the disturbance observer. The temperature overshoot is only 0.8℃, which is far lower than the overshoot of more than 3℃ in traditional control.

[0131] Steady-state test phase: The system identifies the test condition as steady-state and uses 125℃ and 5V as reference parameters. The test parameters are adjusted in real time through an improved fuzzy neural network model. Based on the changes in sample leakage current and operating current, the test chamber temperature and power supply voltage are finely adjusted to ensure that the sample junction temperature is stable at the rated value. During the process, the temperature control accuracy is stable at ±0.2℃ and the voltage control accuracy is ±0.3%FS, which is far superior to the requirements of the AEC-Q100 standard.

[0132] Disturbance compensation and closed-loop control: When disturbances such as grid voltage fluctuations and ambient temperature changes occur during the test, the linear extended state observer completes disturbance observation and feedforward compensation within 0.3s, without any obvious temperature and voltage deviations, completely eliminating the impact of disturbances on test stress; the adaptive PID controller corrects the execution deviation in real time, with no steady-state error throughout the process, and the control stability is far superior to that of traditional fixed parameter PID.

[0133] Failure prediction and test strategy optimization: During the test, the LSTM model continuously monitored the performance degradation trend of each sample. After 450 hours of testing, the failure probability of 3 samples was predicted to reach 12%, triggering a red alert. The system automatically paused the test of these 3 samples and pushed alarm information. After testing, it was confirmed that the samples showed early leakage current degradation, which prevented the samples from completely failing and preserved complete pre-failure data. For the remaining qualified samples with stable performance, the test cycle was optimized based on the stress-life model. Under the premise of meeting the requirements of AEC-Q100 standard, the total test time was shortened from 1000 hours to 750 hours, and the test efficiency was improved by 25%.

[0134] Safety interlock protection: If the test chamber door is accidentally opened once during the test, the system will immediately identify it as an abnormal disturbance, lock the test parameters, trigger the secondary protection, push an alarm message, and start the heating module to compensate for the temperature loss. After the chamber door is closed, the temperature will quickly return to the rated temperature without significantly affecting the test process and no sample damage will occur throughout the process.

[0135] 4. Implementation Results

[0136] This system operated continuously and stably for 750 hours in this batch of HTOL tests, achieving fully unattended adaptive control. The core effects are as follows: test temperature control accuracy ±0.2℃, voltage control accuracy ±0.3%FS, far exceeding the standard requirements; successfully predicted the failure of 3 early failure samples in advance, with a failure prediction accuracy of 100%; the test cycle was shortened by 25%, and equipment energy consumption was reduced by 22%; the workload of manual labor was reduced by 90%, and no full-time supervision was required; the data of the whole process is complete and traceable, and the generated test report fully meets the AEC-Q100 automotive-grade certification requirements, with batch test consistency reaching 99.8%, which is a qualitative improvement over the traditional open-loop test mode.

[0137] Example 2: Application of high-temperature aging test for aerospace airborne products

[0138] 1. Application Scenarios and Testing Requirements

[0139] This embodiment is applied to the high-temperature aging test of airborne power controllers for aerospace applications. The test follows the GJB150 military standard, with a rated test temperature of 85°C and a test duration of 500 hours. The products are 12 airborne power controllers of various models, each with independent channel control. The test scenario involves power grid fluctuations and large changes in ambient temperature. The test process requires precise and controllable stress, real-time monitoring of product performance, prediction of failure risks, and meets the high reliability and high safety requirements of military products.

[0140] 2. Implementation process and results

[0141] The adaptive control method of this invention allows for independent configuration of test parameters and control strategies for each product, constructing a corresponding failure mechanism model. During the test, multi-dimensional data acquisition monitors the product's output voltage, current, power, temperature, and other parameters in real time. An improved fuzzy neural network model adaptively adjusts the test parameters to compensate for disturbances caused by power grid fluctuations. The test temperature control accuracy is stable at ±0.3℃, and the power supply voltage control accuracy is ±0.4%FS. An LSTM model predicts the degradation trend of a product's output characteristics in advance, triggering an early warning and adjusting the test stress to prevent product failure. The entire test process is automated, requiring no manual intervention, ultimately shortening the test cycle by 20%. All product test data is complete and traceable, fully meeting the requirements of GJB military standards, and has received high recognition from users.

[0142] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. An adaptive control method for high-temperature reliability testing processes, applied to high-temperature life, high-temperature aging, and high-temperature storage reliability testing scenarios for semiconductor components, automotive electronics, and aerospace products, characterized in that... A closed-loop adaptive control architecture is constructed to replace the traditional open-loop fixed-parameter test mode. This architecture replaces the traditional open-loop fixed-parameter test mode, enabling dynamic adaptive adjustment and high-precision control during high-temperature reliability testing. The specific steps include: S1 Pre-test parameter calibration and adaptive model initialization: Complete the basic parameter input and performance calibration of test samples and test equipment; based on test standards and sample characteristics, construct the sample failure mechanism model and test stress-life mapping model; and complete the initialization and pre-training of the adaptive control model. S2 Multi-dimensional Status Data Synchronous Real-time Acquisition: Through a distributed sensor network, it synchronously acquires three categories of multi-dimensional data: environmental parameters of the test chamber, sample operating status parameters, and operating parameters of the test equipment. After filtering and noise reduction and timestamp synchronization and alignment, it outputs a standardized real-time status dataset. S3 Test Condition Intelligent Identification and Stress Benchmark Matching: Based on real-time state datasets, the test process is divided into four major conditions through a multi-feature fusion classification model: heating and isothermal stage, steady-state test stage, cooling and recovery stage, and abnormal disturbance stage. Combined with test standards and sample failure mechanism models, the benchmark test stress parameters under the corresponding conditions are matched. S4 Multi-source Disturbance Real-time Observation and Feedforward Compensation: Construct a linear extended state observer to observe three types of multi-source disturbances during the test process: temperature fluctuation inside the chamber, voltage fluctuation of the power grid, and sample performance drift. Generate disturbance feedforward compensation amount, which is superimposed on the reference test stress parameters to eliminate the influence of disturbance on the test stress accuracy. S5 is based on an improved fuzzy neural network for adaptive dynamic adjustment of test parameters: taking real-time state dataset, reference stress parameters, and disturbance compensation as inputs, it dynamically optimizes and adjusts four core test parameters—test temperature, power supply voltage, test cycle, and load parameters—through an improved fuzzy neural network adaptive control model, and outputs the optimal real-time control command. S6 Sample Failure Prediction and Intelligent Optimization of Test Strategies: Based on the real-time operating status parameters of the sample, the time-series prediction model predicts the performance degradation trend and failure risk of the sample, and intelligently optimizes the test strategy based on the failure prediction results: For qualified samples with stable performance, the test cycle and stress intensity are optimized to improve test efficiency; for samples with early failure, an early warning is triggered and the test strategy is adjusted to avoid overstress damage to the sample; for samples that have failed, the test is automatically terminated and failure data is recorded. S7 Closed-Loop Feedback Control and Safety Interlock Protection: Optimized control commands are sent to the high-temperature test chamber, power supply, and load equipment for execution. The actual status data after execution is collected synchronously and compared with the target value. Closed-loop feedback correction is completed through an adaptive PID controller. At the same time, a multi-level safety interlock protection mechanism is set up to immediately trigger emergency protection actions when over-temperature, over-pressure, or sample hard failure occurs. S8 Test Termination and Full-Process Data Archiving: When the test termination conditions are met, the gradient cooling procedure is automatically executed. After the test is completed, a standardized test report is generated, and the full-process test data, control parameters, and failure records are archived simultaneously. The training dataset of the adaptive control model is updated simultaneously to complete the model iterative optimization.

2. The method according to claim 1, characterized in that, The test samples mentioned in step S1 include semiconductor chips, discrete devices, automotive electronic controllers, and aerospace airborne products; the test standards include industry standards such as JEDEC, AEC-Q100, GJB128, and GJB548; the sample failure mechanism models include thermo-oxidative aging failure models, electromigration failure models, and thermal stress fatigue failure models; the adaptive control model pre-training is completed using historical test datasets of similar samples, and the stress control error of the pre-trained model is ≤±0.5%.

3. The method according to claim 1, characterized in that, The acquisition frequency of the multi-dimensional state data in step S2 is 1Hz~100Hz, and the specific classification and acquisition requirements are as follows: Test chamber environmental parameters: including temperature, humidity, heating rate, cooling rate, and temperature uniformity of the test chamber working area, which are collected using a high-precision platinum resistance temperature sensor with a measurement accuracy of ≤±0.2℃; Sample operating status parameters: including sample operating current, operating voltage, leakage current, output power, junction temperature, and key pin impedance, are collected using a high-precision digital multimeter, current sensor, and infrared thermometer. Voltage measurement accuracy is ≤ ±0.1%FS, and current measurement accuracy is ≤ ±0.5%FS. Test equipment operating parameters: including the output power of the test chamber heating / cooling module, the output voltage / current of the power supply, the operating status of the load equipment, the acquisition frequency is synchronized with the sample parameters, and the data timestamp alignment error is guaranteed to be ≤10ms.

4. The method according to claim 1, characterized in that, The multi-feature fusion classification model described in step S3 uses a random forest classification algorithm with temperature change rate, sample current fluctuation rate, and test run time as input features, achieving an accuracy rate of ≥99.5% for working condition identification. The matching rules for the baseline stress parameters of the four working conditions are as follows: Heating and isothermal stage: A gradient heating strategy is adopted, and the heating rate is adaptively matched to the thermal expansion coefficient of the sample to avoid sample damage caused by thermal shock. Steady-state test phase: Based on the rated temperature and rated voltage specified in the test standard, dynamic fine-tuning is performed in combination with the real-time state of the sample; Cooling and recovery phase: A gradient cooling strategy is adopted, with the cooling rate adaptively matched to the thermal stress tolerance threshold of the sample; Abnormal disturbance stage: Lock the test stress reference, prioritize the completion of disturbance compensation, and resume normal test procedures after the state stabilizes.

5. The method according to claim 1, characterized in that, The linear expansion state observer described in step S4 treats the unmodeled dynamics and unknown disturbances during the test as a total disturbance, observes them in real time and performs feedforward compensation. The observer bandwidth is set to 10 times the system bandwidth, the disturbance observation response time is ≤0.5s, the test temperature control accuracy after compensation is ≤±0.3℃, the power supply voltage control accuracy is ≤±0.5%FS, and the test stress deviation caused by the disturbance is completely eliminated.

6. The method according to claim 1, characterized in that, The improved fuzzy neural network adaptive control model described in step S5 has a 5-layer structure, consisting of an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer, and an output layer. The model takes the experimental temperature deviation, voltage deviation, temperature change rate, and sample leakage current change rate as inputs, and the adjustment amounts of experimental temperature, power supply voltage, load parameters, and experimental cycle as outputs. The model uses a genetic algorithm to optimize the membership function and fuzzy rules, solving the problem that traditional fuzzy control rules rely on human experience. The control command response time is ≤1s, and there is no steady-state error in steady-state control.

7. The method according to claim 1, characterized in that, The time-series prediction model described in step S6 is an improved LSTM (Long Short-Term Memory) network. Using historical operational time-series data of the sample as input, it predicts the performance degradation trend and failure probability of the sample within the next 24 hours, with a failure prediction accuracy ≥98%. The intelligent optimization of the experimental strategy specifically involves: When the predicted failure probability of the sample is <1%, within the allowable range of the test standard, the stress intensity and test cycle can be optimized, which can shorten the test time by up to 30%. When the predicted failure probability of a sample is 1% to 10%, a yellow warning is triggered, the test stress intensity is reduced, the data acquisition frequency is increased, and performance changes are continuously monitored. When the predicted failure probability of a sample is greater than 10%, a red alert is triggered, the test is automatically paused, and abnormal information is pushed to the test personnel to avoid damage to the sample due to excessive stress. When a hard failure is detected in the sample, immediately cut off the power supply to the sample, terminate the single-circuit test, and record the failure time and failure status data.

8. The method according to claim 1, characterized in that, The adaptive PID controller described in step S7 uses fuzzy rules to adaptively adjust the proportional, integral, and derivative parameters of the PID controller online, replacing the traditional fixed-parameter PID controller. This solves the control problems of large inertia, large lag, and nonlinearity in high-temperature testing processes, achieving a temperature control overshoot ≤1℃ and a settling time ≤300s. The multi-level safety interlock protection mechanism is divided into three levels: Level 1 warning: When the parameter deviation is ≤5%, a pop-up warning will appear on the interface, and the control parameters will be automatically adjusted. Level 2 protection: If the parameter exceeds the tolerance by 5%~10%, the test stress is locked, the sample load is cut off, and an SMS alarm is sent. Level 3 Emergency Shutdown: If parameters exceed tolerance by more than 10%, or if a sample short circuit or test chamber overheating occurs, immediately disconnect the heating power supply and sample power supply, execute gradient cooling, and trigger an audible and visual alarm.

9. The method according to claim 1, characterized in that, This method supports multi-channel independent parallel control, and can simultaneously perform independent adaptive control on up to 64 different samples. The test parameters and control strategies of each sample are configured independently without interference. It also supports the standard communication protocol for docking with test equipment, achieving seamless docking with high-temperature test chambers, programmable power supplies, and electronic loads without the need for hardware modifications to existing equipment.

Images