Intelligent optimization method for secondary screening of electronic components

Abstract

The invention relates to the technical field of reliability test and intelligent screening of electronic components, and discloses an intelligent optimization method for secondary screening of electronic components, which comprises the following steps of: 1, firstly, constructing a multi-dimensional screening parameter system of the electronic components; according to the method, a screening parameter system covers four categories of electrical performance, environmental adaptability, reliability and process consistency, and judgment deviation caused by a single parameter is avoided; meanwhile, parameter anomaly detection and relevance verification are introduced, even if a single parameter is qualified, relevance anomaly is still marked to be unqualified, and the missed judgment rate is reduced to 2% or below from traditional 8%. On the other hand, the scene adaptation module divides the application into four grades, each grade corresponds to different parameter threshold coefficients, and it is ensured that the screening standard is matched with the scene requirement; the fault feedback module counts the fault correlation degree through FMEA, when the fault contribution degree of a certain parameter is larger than or equal to 5%, the weight of the parameter is automatically improved, the threshold value is tightened, the screening model is optimized along with actual fault data, and the fault rate of downstream elements is reduced by 60% or above.

Description

Technical Field

[0001] This invention relates to the field of electronic component reliability testing and intelligent screening technology, specifically to an intelligent optimization method for secondary screening of electronic components. Background Technology

[0002] In the production and application of electronic components, secondary screening is a crucial process for eliminating early-failure components and ensuring equipment reliability. However, traditional secondary screening methods for electronic components have five major problems, making it difficult to meet the high-precision and high-efficiency screening requirements in different scenarios: The parameters are too narrow and the coverage is incomplete: Traditional screening focuses on electrical performance parameters (such as rated voltage and on-resistance) and ignores environmental adaptability (such as temperature resistance and vibration resistance), reliability (such as aging degradation) and process consistency (such as pin coplanarity). This leads to the failure of "qualified components" in actual applications due to insufficient environmental tolerance or process defects. For example, in automotive-grade scenarios, vibration stability is not screened. After the components are installed, poor contact occurs due to bumps, and the failure rate exceeds 8%.

[0003] Fixed and rigid thresholds lead to poor scenario adaptability: Traditional screening uses a uniform threshold (e.g., all scenarios have a temperature resistance of -40℃ to 85℃), without distinguishing between different levels of reliability requirements such as civilian, industrial, and automotive. Automotive-grade components are not thoroughly screened due to overly wide thresholds, while civilian-grade components are over-screened due to overly strict thresholds (with a pass rate of only 60%), resulting in serious waste of resources.

[0004] The model has low accuracy and weak anti-interference ability: Traditional screening often uses simple threshold comparison or a single algorithm (such as basic decision tree), which cannot handle small samples and imbalanced data (the proportion of unqualified samples is often <10%), and the classification error rate exceeds 5%; in addition, there is no anomaly detection and correlation verification, which can easily lead to missed judgments due to single parameter being qualified but correlation being abnormal (such as the mismatch between conduction resistance and temperature), and potential faulty components flowing downstream.

[0005] The screening process is inefficient, time-consuming, and costly: Traditional screening uses a fixed testing sequence (such as first testing the time-consuming temperature cycle, and then testing the fast on-resistance). Unqualified components need to complete the entire process of testing, and the average testing time for a single component exceeds 60 seconds. Moreover, there is no efficiency optimization strategy. In continuous batch screening, parameters with high pass rates are still sampled at 100%, resulting in a waste of more than 30% of testing resources.

[0006] Without data closure, the model does not evolve: Traditional screening only outputs a qualified list, does not track fault data in actual application of components, and cannot correct screening parameters and models; when component processes are upgraded or application scenarios change, the model still uses old parameters, screening accuracy decreases over time, and the false judgment rate rises to more than 10% after 1 year. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent optimization method for secondary screening of electronic components, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent optimization method for secondary screening of electronic components, comprising the following steps: Step 1: First, construct a multi-dimensional screening parameter system for electronic components. This system covers four major categories: electrical performance parameters, environmental adaptability parameters, reliability parameters, and process consistency parameters. Among them, electrical performance parameters include rated voltage, rated current, on-resistance, breakdown voltage, and response time; environmental adaptability parameters include operating temperature range, humidity tolerance, vibration stability, and shock resistance; reliability parameters include mean time between failures (MTBF), high-temperature aging degradation rate, and low-temperature start-up success rate; and process consistency parameters include appearance defect level, pin coplanarity, and package sealing.

[0009] Step 2: Next, establish an intelligent screening model, using a combination of an improved random forest algorithm and adaptive threshold optimization. The specific steps are as follows: The first step is to construct the dataset by collecting screening parameter data for N batches of electronic components of the same model, with each batch containing ≥1000 samples, including qualified and unqualified samples, and dividing them into training and test sets in a 7:3 ratio. The second step is feature importance assessment, which calculates the importance weight of each parameter using the Gini index of the random forest. The formula is ,in For the number of decision trees, For the number of categories, For the first Category in trees The sample proportions are sorted by importance, and the top K parameters are retained. The third step is model training, which uses 5-fold cross-validation to optimize the hyperparameters of the random forest. The objective function is to minimize the classification error rate. in, It is a true positive. It is a true negative. It was a false positive. It is a false negative; The fourth step is adaptive threshold generation. For each parameter j, based on the parameter distribution of qualified samples in the training set, the probability density function is calculated using kernel density estimation. Set threshold range [ ],satisfy ≥99%, and introduce parameter correlation coefficient. For strongly correlated parameters, a joint threshold is used.

[0010] Step 3: The intelligent screening process is then executed, including three stages: initial screening, fine screening, and optimization iteration. In the initial screening stage, multi-dimensional parameters of the components to be screened are collected using automated testing equipment and compared with adaptive thresholds. Components with single parameters exceeding the threshold are eliminated, and a comprehensive pass index is calculated for the remaining components. ,in For parameters Standardized scores, retain Elements ≥0.6 are screened for precision. In the fine screening stage, the components that passed the initial screening were subjected to enhanced testing, and the parameter drift after enhancement was collected. Calculate the drift influence coefficient Set drift threshold , ≤ The components are marked as "qualified after screening"; In the optimization iteration phase, after screening every M batches of components, components that have failed in actual use are added to the training set. The parameter importance weights and adaptive thresholds are recalculated, and the intelligent screening model is updated. The iterative optimization formula is as follows: ,in For the number of iterations, For learning rate, For the parameter Changes in error rate due to misjudgment.

[0011] Step 4: Finally, output the screening results and optimization report. The results include a list of qualified components and statistics on the classification of unqualified components. The report includes the pass rate, false positive rate, and screening efficiency of this screening, and compares the performance improvement of traditional screening methods, while also proposing parameter optimization suggestions.

[0012] Preferably, when constructing a multi-dimensional screening parameter system, a dynamic parameter adjustment mechanism also needs to be established. This mechanism realizes real-time parameter updates and weight adaptation based on component application scenarios and historical fault data, specifically including a scenario adaptation module and a fault feedback module: The scenario adaptation module divides application scenarios into four levels: civilian, industrial, automotive, and military. Each level corresponds to a different parameter threshold adjustment coefficient. For example, the threshold coefficient for the operating temperature range in automotive-grade applications. =1.2, the threshold coefficient for vibration stability =0.8, parameter In the scene The threshold below is , ; The fault feedback module uses the FMEA method to statistically analyze the contribution of each parameter in historical faults and calculate the fault correlation. = (due to parameters) (Number of failures caused by exceeding the standard) / Total number of failures, when When the percentage is ≥5%, increase the importance weight of this parameter. and tighten its threshold ( , ).

[0013] Preferably, during the training process of the intelligent screening model, a weighted loss function and an ensemble learning strategy are also introduced to improve the classification accuracy of small samples and imbalanced data. The weighted loss function addresses the issue of a low proportion of non-compliant samples by assigning different weights to samples of different categories. The formula for the loss function is as follows: ,in For the first The weights of each sample, Cross-entropy loss ; The ensemble learning strategy employs a fusion model of random forest and gradient boosting tree. First, the random forest model is trained to obtain the predicted probabilities. Train the GBDT model to obtain the predicted probability Then, logistic regression is used to fuse the models and output the final probability. ,in For the sigmoid function, , , These are the fusion parameters.

[0014] Preferably, the initial screening stage further includes sub-steps for parameter anomaly detection and correlation verification to improve screening accuracy. An anomaly detection algorithm using parameters employs an improved Isolation Forest algorithm. An anomaly scoring model is constructed for the initially screened parameters, and an anomaly score is calculated for each element. in For the first The path length of a sample in an isolated tree. This represents the average path length. A higher value indicates a higher probability of an anomaly; an anomaly threshold should be set. , A≥ The component is marked as "suspected abnormal" and requires manual review; Correlation verification targets strongly correlated parameters and establishes parameter correlation models. For example, the correlation model between on-resistance R and temperature T is as follows: If the deviation between the measured value and the model prediction value ΔR = |R_measured - R_predicted| / R_predicted > 5%, it is judged as an abnormal correlation, even if all individual parameters are within the threshold. It is still marked as "abnormal correlation". For components that are "suspected abnormal" and "abnormal correlation", supplementary testing is initiated, and the average of the tests is used for re-evaluation. If it is still abnormal, it is classified as unqualified to ensure that no potential faulty components are missed in the initial screening stage. At the same time, abnormal cases are added to the model training set, and the abnormal threshold of the isolated forest is updated once for each batch.

[0015] Preferably, the enhanced testing in the fine screening stage adopts an adaptive stress loading strategy, dynamically adjusting the test stress and duration based on the performance of the initial screening parameters, specifically as follows: First, calculate the robustness index of the initial screening parameters. ( The mean of the qualified samples. (Standard deviation) A value closer to 1 indicates a more robust parameter; calculate the average robustness index for all parameters. ,according to Stress level classification: ≥0.8 indicates low stress, 0.5≤ <0.8 indicates moderate stress. <0.5 indicates high stress; The stress loading sequence of the reinforcement test adopts an orthogonal experimental design to avoid misjudgment caused by stress superposition effect. For example, the combination sequence of temperature cycling and vibration test is "high temperature → vibration → low temperature → vibration". After each stress stage, the test is left to stand for 30 minutes before collecting parameters. The screening stage also introduces an equivalent conversion for accelerated life testing. The relationship between high-temperature testing duration and actual lifespan is calculated using the Arrhenius model, with the formula as follows: ,in High temperature The test duration is as follows. room temperature The equivalent lifetime below, To activate energy, The Boltzmann constant is used to require that the equivalent life of the finely screened components be ≥ 1.5 times the target life.

[0016] Preferably, the optimization iteration stage further includes a screening efficiency optimization submodule, which shortens the testing time while ensuring screening quality by dynamically adjusting the test items and test order. The specific method is as follows: Establish a database of test item time consumption to record the average test time for each parameter. Calculate the "cost-effectiveness" index of the parameters ; according to Sort the test items from highest to lowest, and test the highest-ranking ones first. Parameters: When a component fails to meet the high-weight parameters, the subsequent test is immediately terminated and marked as unqualified. For parameters with a pass rate ≥ 95% in three consecutive screening batches, reduce the testing frequency; for parameters with a pass rate < 80%, increase the testing frequency. A parallel testing strategy is introduced, in which non-interference parameter test items are arranged to be executed in parallel at the same workstation, and data is collected synchronously through the multi-channel test interface of the automated equipment. The synchronization error of the parallel test is ≤10ms. The filtering efficiency optimization submodule calculates the time saving rate after each batch of filtering. ,Require The error rate is ≥30%, and an efficiency-quality balance model is established. When the error rate increases to ≥0.5%, the frequency or order of some test items is automatically restored.

[0017] Preferably, it also includes a closed-loop management step for screening data, achieving continuous evolution of the screening model through full lifecycle data tracking. The specific process is as follows: Each selected component is assigned a unique identifier, and its working status data is recorded during the application of the component and uploaded to the cloud database in real time through the industrial IoT module; The cloud-based data processing unit periodically compares the actual failure rate of components with the screening and prediction results to calculate the prediction accuracy. ; Root cause analysis was performed on cases of prediction errors, and the fishbone diagram method was used to locate the error type. For cases with missing parameters, corresponding test items were added. For cases where the threshold is unreasonable, the threshold range of the parameter is recalculated; for cases where the model is biased, the transfer learning method is used to fuse new scene data with historical data, freeze the underlying parameters of the model, and only fine-tune the top classifier to reduce the amount of computation required for retraining. Closed-loop management also generates component health curves, predicting the remaining lifespan of components based on screening parameters and actual operating data. ,in This is the initial lifetime prediction value. The attenuation coefficient is... For the time already worked, when ≤30% In time, a replacement alert is pushed to the user, realizing the optimization of the entire chain from screening to operation and maintenance.

[0018] This invention provides an intelligent optimization method for secondary screening of electronic components. It has the following beneficial effects: 1. The screening parameter system of this invention covers four major categories: electrical performance, environmental adaptability, reliability, and process consistency, avoiding judgment bias caused by a single parameter. It also introduces parameter anomaly detection and correlation verification; even if a single parameter is qualified, anomalies in the correlation are still marked as unqualified, reducing the false negative rate from the traditional 8% to below 2%. Furthermore, the scenario adaptation module divides applications into four levels, each corresponding to different parameter threshold coefficients, ensuring that the screening criteria match the scenario requirements. The fault feedback module uses FMEA to statistically analyze fault correlation; when the fault contribution of a certain parameter is ≥5%, its weight is automatically increased and the threshold is tightened, allowing the screening model to optimize with actual fault data, reducing the downstream component failure rate by more than 60%.

[0019] 2. This invention employs an improved random forest + ensemble learning strategy—random forest calculates parameter importance using the Gini index, and optimizes hyperparameters using 5-fold cross-validation; it integrates the GBDT model and logistic regression to output the final prediction probability; a weighted loss function is introduced to handle imbalanced data, reducing the classification error rate from 5% to below 1.5%, and achieving an F1 score ≥0.95 for automotive / military grade screening. Adaptive stress loading classifies stress levels according to the robustness index of the initial screening parameters, avoiding overtesting and shortening high-temperature testing time by 40%; the testing order is sorted according to the "cost-effectiveness index," and testing is immediately terminated if high-weight parameters fail, reducing invalid testing; parallel testing executes non-interference items simultaneously, reducing the average unit component testing time from 60s to 25s, a time saving rate ≥30%; at the same time, the sampling frequency for high-pass-rate parameters is reduced, reducing testing resource waste by 25% and overall screening cost by 35%. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the intelligent optimization method for secondary screening of electronic components according to the present invention; Figure 2 This is a flowchart of the multi-dimensional screening parameter preprocessing process of the present invention; Figure 3 This is a flowchart of the intelligent screening model training process of the present invention; Figure 4 This is a flowchart of the intelligent screening execution and optimization iteration process of the present invention. Detailed Implementation

[0021] 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.

[0022] Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention.

[0023] Example 1 A preferred embodiment of the intelligent optimization method for secondary screening of electronic components provided by the present invention is as follows: Figure 1-4 The intelligent optimization method for secondary screening of electronic components includes the following steps: Step 1: First, construct a multi-dimensional screening parameter system for electronic components. This system covers four major categories: electrical performance parameters, environmental adaptability parameters, reliability parameters, and process consistency parameters. Among them, electrical performance parameters include rated voltage, rated current, on-resistance, breakdown voltage, and response time; environmental adaptability parameters include operating temperature range, humidity tolerance, vibration stability, and shock resistance; reliability parameters include mean time between failures (MTBF), high-temperature aging degradation rate, and low-temperature start-up success rate; and process consistency parameters include appearance defect level, pin coplanarity, and package sealing.

[0024] Step 2: Next, establish an intelligent screening model, using a combination of an improved random forest algorithm and adaptive threshold optimization. The specific steps are as follows: The first step is to construct the dataset by collecting screening parameter data for N batches of electronic components of the same model, with each batch containing ≥1000 samples, including qualified and unqualified samples, and dividing them into training and test sets in a 7:3 ratio. The second step is feature importance assessment, which calculates the importance weight of each parameter using the Gini index of the random forest. The formula is ,in For the number of decision trees, For the number of categories, For the first Category in trees The sample proportions are sorted by importance, and the top K parameters are retained. The third step is model training, which uses 5-fold cross-validation to optimize the hyperparameters of the random forest (number of decision trees: 500-1000, maximum depth: 10-20, minimum number of leaf node samples: 5-10). The objective function is to minimize the classification error rate. in, It is a true positive. It is a true negative. It was a false positive. It is a false negative; The fourth step is adaptive threshold generation. For each parameter j, based on the parameter distribution of qualified samples in the training set, the probability density function is calculated using kernel density estimation (KDE). Set threshold range [ ],satisfy ≥99% (confidence level), and introduce parameter correlation coefficient. (Pearson correlation coefficient, | |≥0.7 is considered a strong correlation), and a joint threshold is used for strongly correlated parameters (such as the joint threshold of rated voltage and breakdown voltage satisfying Vbr≥1.2×U).

[0025] Step 3: The intelligent screening process is then executed, including three stages: initial screening, fine screening, and optimization iteration. In the initial screening stage, multi-dimensional parameters of the components to be screened are collected through automated testing equipment and compared with adaptive thresholds. Components with single parameters exceeding the threshold are eliminated (marked as "initial screening unqualified"). A comprehensive pass index is calculated for the remaining components. ,in For parameters Standardized score ( , The mean of the qualified samples. Standard deviation, When ≤0, the contribution of this parameter is 0; retain. Components with a strength ≥0.6 enter the fine screening stage; during the fine screening stage, components that pass the initial screening undergo enhanced testing (such as thermal shock cycling at high temperature of 125℃ / 2h and low temperature of -55℃ / 2h), and the parameter drift after enhancement is collected. Calculate the drift influence coefficient Set drift threshold (Based on reliability requirements, typically ≤5%) ≤ The components are marked as "qualified after screening"; during the optimization iteration phase, after screening every M batches of components, the components that malfunctioned in actual use (feedback data) are added to the training set, the parameter importance weights and adaptive thresholds are recalculated, and the intelligent screening model is updated. The iterative optimization formula is as follows: ,in For the number of iterations, The learning rate is 0.01 to 0.1. For the parameter Changes in error rate due to misjudgment ensure that the model is dynamically optimized according to the application scenario.

[0026] Step 4: Finally, output the screening results and optimization report. The results include a list of qualified components (with measured values ​​of each parameter, comprehensive pass index S, and drift influence coefficient D), and statistics on the classification of unqualified components (by parameter exceeding type and failure mode). The report includes the pass rate (number of qualified components / total number of components screened), the false positive rate ((FP+FN) / total number of components screened, target ≤2%), and the screening efficiency (average test time per component, target ≤30s). It also compares the performance improvement of traditional screening methods (e.g., pass rate increased by 10%~20%, false positive rate reduced by more than 50%), and provides parameter optimization suggestions (e.g., "the on-resistance drift of a certain batch of components is large, it is recommended to increase the weight of this parameter in the fine screening").

[0027] When constructing a multi-dimensional screening parameter system, a dynamic parameter adjustment mechanism also needs to be established. This mechanism realizes real-time parameter updates and weight adaptation based on component application scenarios and historical fault data, specifically including a scenario adaptation module and a fault feedback module: The scenario adaptation module categorizes application scenarios into four levels: civilian (temperature range -20℃~70℃, vibration requirement ≤10G), industrial (-40℃~85℃, vibration requirement ≤30G), automotive (-40℃~125℃, vibration requirement ≤50G), and military (-55℃~150℃, vibration requirement ≤100G). Each level corresponds to a different parameter threshold adjustment coefficient. For example, the threshold coefficient for the operating temperature range in automotive-grade applications. =1.2 (i.e., the acceptable range is expanded by 20%), the threshold coefficient for vibration stability. =0.8 (meaning the allowable drift range is reduced by 20%), parameter In the scene The threshold below is , ; The fault feedback module uses FMEA (Failure Mode and Effects Analysis) to statistically analyze the contribution of each parameter in historical faults and calculate the fault correlation. = (due to parameters) (Number of failures caused by exceeding the standard) / Total number of failures, when When the percentage is ≥5%, increase the importance weight of this parameter. and tighten its threshold ( , The dynamic adjustment mechanism also supports user-defined parameters, allowing users to add new parameter items according to special needs (such as radiation tolerance in the aerospace field). The initial weight of the new parameter is assigned according to its correlation with the existing parameters (the higher the correlation coefficient, the closer the initial weight is to the relevant parameter), and the weight is recalculated based on actual data after three batches of screening to ensure that the parameter system always matches the application requirements.

[0028] During the training of the intelligent screening model, a weighted loss function and an ensemble learning strategy are also introduced to improve the classification accuracy for small samples and imbalanced data: the weighted loss function addresses the problem of a low proportion of unqualified samples (usually <10%) by assigning different weights to samples of different categories. The loss function formula is as follows: ,in For the first The weights of each sample (positive samples) =1, negative sample , The number of positive samples (Negative sample size) Cross-entropy loss ( For real labels, (To predict probabilities for the model); the ensemble learning strategy employs a fusion model of random forest and gradient boosting tree (GBDT). First, the random forest model is trained to obtain the predicted probabilities. Train the GBDT model (learning rate 0.1, number of trees 200) to obtain the predicted probability. Then, logistic regression is used to fuse the models and output the final probability. ,in For the sigmoid function, , , To fuse parameters (optimized via grid search, aiming to minimize the AUC on the test set); to avoid overfitting, a regularization term was added during model training, and each tree in the random forest was sampled using bootstrap sampling (70% sampling ratio), with a limit on the maximum number of leaf nodes; GBDT incorporated L2 regularization (regularization coefficient 0.01), and an early stopping strategy was adopted (training stopped if there was no improvement on the validation set for 5 consecutive rounds); in addition to error rate, the model evaluation metrics also included accuracy. Recall rate F1 score The requirements are that the F1 score should be ≥0.95 for automotive-grade and military-grade screening, and ≥0.90 for civilian-grade and industrial-grade screening, to ensure the classification reliability of the model in different scenarios.

[0029] The initial screening stage also includes sub-steps for parameter anomaly detection and correlation verification to improve screening accuracy: Parameter anomaly detection employs an improved isolated forest algorithm to construct an anomaly scoring model for the initial screening parameters and calculate anomaly scores for each component. in For the first The path length of a sample in an isolated tree. Average path length ( (where H is the harmonic function). A higher value indicates a higher probability of an anomaly; an anomaly threshold should be set. (Based on the score distribution of outlier samples in the training set, taking the 95th percentile), A≥ Components marked as "suspected anomalies" require manual review; correlation checks target strongly correlated parameters (| (≥0.7), establish a parameter correlation model, for example, the correlation model between on-resistance R and temperature T is: ( (The temperature coefficient is calculated through linear regression). If the deviation between the measured value and the model prediction value ΔR = |R_measured - R_predicted| / R_predicted > 5%, it is judged as an abnormal correlation, even if all individual parameters are within the threshold. It is still marked as an "abnormal correlation". For components that are "suspected abnormal" and "abnormal correlation", supplementary testing (such as 3 additional repeated tests) is initiated. The average of the tests is taken for re-evaluation. If it is still abnormal, it is classified as unqualified. This ensures that no potential faulty components are missed in the initial screening stage. At the same time, abnormal cases are added to the model training set. The abnormal threshold of the isolated forest is updated once for each batch to improve the adaptability of abnormal detection.

[0030] The enhanced testing in the fine screening stage adopts an adaptive stress loading strategy, dynamically adjusting the test stress and duration based on the performance of the initial screening parameters. Specifically, the robustness index of the initial screening parameters is first calculated. ( The mean of the qualified samples. (Standard deviation) A value closer to 1 indicates a more robust parameter; calculate the average robustness index for all parameters. ,according to Stress level classification: ≥0.8 indicates low stress (high temperature 85℃ / 1h, low temperature -40℃ / 1h, vibration 10G), 0.5≤ <0.8 indicates medium stress (high temperature 105℃ / 2h, low temperature -55℃ / 2h, vibration 20G). <0.5 indicates high stress (high temperature 125℃ / 4h, low temperature -65℃ / 4h, vibration 30G); the stress loading sequence of the intensification test adopts an orthogonal experimental design to avoid misjudgment caused by stress superposition effect. For example, the combination sequence of temperature cycling and vibration testing is "high temperature → vibration → low temperature → vibration", and parameters are collected after each stress stage by resting for 30 minutes; the fine screening stage also introduces the equivalent transformation of accelerated life testing. The correspondence between high temperature test duration and actual life is calculated according to the Arrhenius model, and the formula is: ,in High temperature The test duration is as follows. room temperature The equivalent lifetime below, The activation energy is typically 0.7 eV for electronic components. Boltzmann constant (8.617 × 10⁻⁶) -5 (eV / K), requiring that the equivalent life of the components after fine screening be ≥ 1.5 times the target life (e.g., if the target life is 10 years, the equivalent test must be ≥ 15 years) to ensure that the components that pass the fine screening have long-term reliability.

[0031] The optimization iteration phase also includes a screening efficiency optimization submodule. By dynamically adjusting test items and test order, the testing time is shortened while ensuring screening quality. Specifically, a database of test item time consumption is established to record the average testing time for each parameter. (For example, conduction resistance test for 0.5s, temperature cycling test for 300s), calculate the "cost-effectiveness" index of the parameters. (Ratio of importance to time spent); by Sort the test items from highest to lowest, and test the highest-ranking ones first. Parameters, when a component has high weighting parameters (cumulative) If any parameter fails (≥60%), subsequent testing is immediately terminated and the parameter is marked as unqualified. For parameters with a pass rate ≥95% in three consecutive batches, the testing frequency is reduced (e.g., from 100% sampling to 50%). For parameters with a pass rate <80%, the testing frequency is increased (increased to 100% sampling and the number of tests is increased). A parallel testing strategy is introduced, where non-interference parameter tests (e.g., appearance inspection and continuity resistance testing) are performed in parallel at the same workstation. Data is collected synchronously through the multi-channel test interface of the automated equipment, and the synchronization error of parallel testing is ≤10ms. The screening efficiency optimization submodule calculates the time saving rate after each batch of screening. ,Require The error rate is ≥30%, and an efficiency-quality balance model is established. When the error rate increases to ≥0.5%, the frequency or order of some test items is automatically restored to ensure that efficiency improvement does not come at the expense of quality.

[0032] Example 2 Please see Figures 1-4 Furthermore, based on Example 1, the following steps are added: a closed-loop management step for screening data is also included. This step achieves continuous evolution of the screening model through full lifecycle data tracking. Specifically, each screened component is assigned a unique identifier (a QR code containing batch, test time, and device number). During component application (e.g., assembly, debugging, and maintenance phases), its operating status data (temperature, voltage, failure time, etc.) is recorded and uploaded to the cloud database in real time via the Industrial Internet of Things (IIoT) module. The cloud data processing unit periodically (monthly) compares the actual failure rate of the components with the screening prediction results to calculate the prediction accuracy. For cases of prediction errors, root cause analysis is performed, and the fishbone diagram method is used to locate the error type (missing parameters, unreasonable thresholds, model bias, etc.). For missing parameters, corresponding test items are added (e.g., if the failure is caused by not measuring humidity parameters, humidity tolerance is added to the parameter system). For unreasonable thresholds, the threshold range of the parameter is recalculated (expanded or reduced by 5%~10%). For model bias, transfer learning is used to fuse new scenario data with historical data, freeze the underlying parameters of the model, and only fine-tune the top-level classifier to reduce the computational cost of retraining. Closed-loop management also generates component health curves, predicting the remaining lifespan of components based on the selected parameters and actual working data. ,in This is the initial lifetime prediction value. This is the attenuation coefficient (fitted based on actual attenuation data). For the time already worked, when ≤30% In time, a replacement alert is pushed to the user, realizing the optimization of the entire chain from screening to operation and maintenance.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0034] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent optimization method for secondary screening of electronic components, characterized in that, Includes the following steps: Step 1: First, construct a multi-dimensional screening parameter system for electronic components. This system covers four major categories: electrical performance parameters, environmental adaptability parameters, reliability parameters, and process consistency parameters. Among them, electrical performance parameters include rated voltage, rated current, on-resistance, breakdown voltage, and response time; environmental adaptability parameters include operating temperature range, humidity tolerance, vibration stability, and shock resistance; reliability parameters include mean time between failures (MTBF), high-temperature aging degradation rate, and low-temperature start-up success rate; and process consistency parameters include appearance defect level, pin coplanarity, and package sealing. Step 2: Next, establish an intelligent screening model, using a combination of an improved random forest algorithm and adaptive threshold optimization. The specific steps are as follows: The first step is to construct the dataset by collecting screening parameter data for N batches of electronic components of the same model, with each batch containing ≥1000 samples, including qualified and unqualified samples, and dividing them into training and test sets in a 7:3 ratio. The second step is feature importance assessment, which calculates the importance weight of each parameter using the Gini index of the random forest. The formula is ,in For the number of decision trees, For the number of categories, For the first Category in trees The sample proportions are sorted by importance, and the top K parameters are retained. The third step is model training, which uses 5-fold cross-validation to optimize the hyperparameters of the random forest. The objective function is to minimize the classification error rate. in, It is a true positive. It is a true negative. It was a false positive. It is a false negative; The fourth step is adaptive threshold generation. For each parameter j, based on the parameter distribution of qualified samples in the training set, the probability density function is calculated using kernel density estimation. Set threshold range [ ],satisfy ≥99%, and introduce parameter correlation coefficient. For strongly correlated parameters, a joint threshold is used; Step 3: The intelligent screening process is then executed, including three stages: initial screening, fine screening, and optimization iteration. In the initial screening stage, multi-dimensional parameters of the components to be screened are collected using automated testing equipment and compared with adaptive thresholds. Components with single parameters exceeding the threshold are eliminated, and a comprehensive pass index is calculated for the remaining components. ,in For parameters Standardized scores, retain Elements ≥0.6 are screened for precision. In the fine screening stage, the components that passed the initial screening were subjected to enhanced testing, and the parameter drift after enhancement was collected. Calculate the drift influence coefficient Set drift threshold , ≤ The components are marked "qualified after screening"; In the optimization iteration phase, after screening every M batches of components, components that have failed in actual use are added to the training set. The parameter importance weights and adaptive thresholds are recalculated, and the intelligent screening model is updated. The iterative optimization formula is as follows: ,in For the number of iterations, For learning rate, For the parameter Changes in error rate due to misjudgment; Step 4: Finally, output the screening results and optimization report. The results include a list of qualified components and statistics on the classification of unqualified components. The report includes the pass rate, false positive rate, and screening efficiency of this screening, and compares the performance improvement of traditional screening methods, while also proposing parameter optimization suggestions.

2. The intelligent optimization method for secondary screening of electronic components according to claim 1, characterized in that, When constructing a multi-dimensional screening parameter system, a dynamic parameter adjustment mechanism also needs to be established. This mechanism realizes real-time parameter updates and weight adaptation based on component application scenarios and historical fault data, specifically including a scenario adaptation module and a fault feedback module: The scenario adaptation module divides application scenarios into four levels: civilian, industrial, automotive, and military. Each level corresponds to a different parameter threshold adjustment coefficient. For example, the threshold coefficient for the operating temperature range in automotive-grade applications. =1.2, the threshold coefficient for vibration stability =0.8, parameter In the scene The threshold below is , ; The fault feedback module uses the FMEA method to statistically analyze the contribution of each parameter in historical faults and calculate the fault correlation. = (due to parameters) (Number of failures caused by exceeding the standard) / Total number of failures, when When the percentage is ≥5%, increase the importance weight of this parameter. and tighten its threshold ( , ).

3. The intelligent optimization method for secondary screening of electronic components according to claim 1, characterized in that, During the training process of the intelligent screening model, a weighted loss function and an ensemble learning strategy are also introduced to improve the classification accuracy of small samples and imbalanced data. The weighted loss function addresses the issue of a low proportion of non-compliant samples by assigning different weights to samples of different categories. The formula for the loss function is as follows: ,in For the first The weights of each sample, Cross-entropy loss ; The ensemble learning strategy employs a fusion model of random forest and gradient boosting tree. First, the random forest model is trained to obtain the predicted probabilities. Train the GBDT model to obtain the predicted probability Then, logistic regression is used to fuse the models and output the final probability. ,in For the sigmoid function, , , These are the fusion parameters.

4. The intelligent optimization method for secondary screening of electronic components according to claim 1, characterized in that, The initial screening stage also includes sub-steps for parameter anomaly detection and correlation verification to improve screening accuracy. An anomaly detection algorithm using parameters employs an improved Isolation Forest algorithm. An anomaly scoring model is constructed for the initially screened parameters, and an anomaly score is calculated for each element. in For the first The path length of a sample in an isolated tree. This represents the average path length. A higher value indicates a higher probability of an anomaly; an anomaly threshold should be set. , A≥ The component is marked as "suspected abnormal" and requires manual review; Correlation verification targets strongly correlated parameters and establishes parameter correlation models. For example, the correlation model between on-resistance R and temperature T is as follows: If the deviation between the measured value and the model prediction value ΔR = |R_measured - R_predicted| / R_predicted > 5%, it is judged as an abnormal correlation, even if all individual parameters are within the threshold. It is still marked as "abnormal correlation". For components that are "suspected abnormal" and "abnormal correlation", supplementary testing is initiated, and the average of the tests is used for re-evaluation. If it is still abnormal, it is classified as unqualified to ensure that no potential faulty components are missed in the initial screening stage. At the same time, abnormal cases are added to the model training set, and the abnormal threshold of the isolated forest is updated once for each batch.

5. The intelligent optimization method for secondary screening of electronic components according to claim 1, characterized in that, The enhanced testing in the fine screening stage employs an adaptive stress loading strategy, dynamically adjusting the test stress and duration based on the performance of the initial screening parameters. Specifically: First, calculate the robustness index of the initial screening parameters. ( The mean of the qualified samples. (Standard deviation) A value closer to 1 indicates a more robust parameter; calculate the average robustness index for all parameters. ,according to Stress level classification: ≥0.8 indicates low stress, 0.5≤ <0.8 indicates moderate stress. <0.5 indicates high stress; The stress loading sequence of the reinforcement test adopts an orthogonal experimental design to avoid misjudgment caused by stress superposition effect. For example, the combination sequence of temperature cycling and vibration test is "high temperature → vibration → low temperature → vibration". After each stress stage, the test is allowed to stand for 30 minutes before the parameters are collected. The screening stage also introduces an equivalent conversion for accelerated life testing. The relationship between high-temperature testing duration and actual lifespan is calculated using the Arrhenius model, with the formula as follows: ,in High temperature The test duration is as follows. room temperature The equivalent lifetime below, To activate energy, The Boltzmann constant is used to require that the equivalent life of the finely screened components be ≥ 1.5 times the target life.

6. The intelligent optimization method for secondary screening of electronic components according to claim 1, characterized in that, The optimization iteration phase also includes a screening efficiency optimization submodule, which shortens the testing time while ensuring screening quality by dynamically adjusting test items and test order. The specific method is as follows: Establish a database of test item time consumption to record the average test time for each parameter. Calculate the "cost-effectiveness" index of the parameters. ; according to Sort the test items from highest to lowest, and test the highest-ranking ones first. Parameters: When a component fails to meet the high-weight parameters, the subsequent test is immediately terminated and marked as unqualified. For parameters with a pass rate ≥ 95% in three consecutive screening batches, reduce the testing frequency; for parameters with a pass rate < 80%, increase the testing frequency. A parallel testing strategy is introduced, in which non-interference parameter test items are arranged to be executed in parallel at the same workstation, and data is collected synchronously through the multi-channel test interface of the automated equipment. The synchronization error of the parallel test is ≤10ms. The filtering efficiency optimization submodule calculates the time saving rate after each batch of filtering. ,Require The error rate is ≥30%, and an efficiency-quality balance model is established. When the error rate increases to ≥0.5%, the frequency or order of some test items is automatically restored.

7. The intelligent optimization method for secondary screening of electronic components according to claim 1, characterized in that, It also includes a closed-loop management step for data screening, enabling continuous evolution of the screening model through full lifecycle data tracking. The specific process is as follows: Each selected component is assigned a unique identifier, and its working status data is recorded during the application of the component and uploaded to the cloud database in real time through the industrial IoT module; The cloud-based data processing unit periodically compares the actual failure rate of components with the screening and prediction results to calculate the prediction accuracy. ; Root cause analysis was performed on cases of prediction errors, and the fishbone diagram method was used to locate the error type. For cases with missing parameters, corresponding test items were added. For cases where the threshold is unreasonable, the threshold range of the parameter is recalculated; for cases where the model is biased, the transfer learning method is used to fuse new scene data with historical data, freeze the underlying parameters of the model, and only fine-tune the top classifier to reduce the amount of computation required for retraining. Closed-loop management also generates component health curves, predicting the remaining lifespan of components based on screening parameters and actual operating data. ,in This is the initial lifetime prediction value. The attenuation coefficient is... For the time already worked, when ≤30% In time, a replacement alert is pushed to the user, realizing the optimization of the entire chain from screening to operation and maintenance.

Images