Solid state disk testing method and device for big data storage and storage medium

Abstract

The invention discloses a solid state disk testing method and device for big data storage and a storage medium, and relates to the technical field of big data storage.The method includes the steps that multi-dimensional production testing of covering components, processes, functions and reliability is executed, the result is subjected to digital scoring, and the comprehensive quality score Qtotal is calculated based on the preset weight; further extracting a production sample to be verified on a benchmark test platform for simulating actual application of a customer, establishing a sample library for marking customer standard conformity states and corresponding Qtotal and sub-item data, accurately aligning internal test standards with actual requirements of the customer in a data driving mode, dynamically adjusting test parameters through statistical modeling and algorithm iteration, and finally obtaining the standard conformity state of the customer. Therefore, the over-test problem is scientifically quantified and optimized, on the premise of ensuring that the solid state disk meets the real reliability requirement of a client, excessive elimination is remarkably reduced, the test cost and time are reduced, and self-adaptive balance of the production test standard between economical efficiency and reliability is achieved.

Description

Technical Field

[0001] This invention relates to the field of big data storage technology, specifically to a solid-state drive testing method, apparatus, and storage medium for big data storage. Background Technology

[0002] With the rapid development of technologies such as cloud computing, the Internet of Things, and artificial intelligence, the global data volume is exploding, placing higher demands on the capacity, performance, and reliability of big data storage systems. Solid-state drives (SSDs), due to their advantages such as fast read and write speeds, low power consumption, and strong shock resistance, have become the core storage medium in big data storage centers, with deployments reaching the scale of millions or even tens of millions of units.

[0003] In this application context, the overall reliability of the storage system is highly dependent on the failure rate of individual SSDs. To ensure reputation and service quality, SSD manufacturers generally adopt a conservative strategy: in factory testing, they implement internal testing standards that are far higher than those actually used by customers. While this "over-designed" testing method can reduce the risk of field failures to some extent, it also has problems such as long testing times, increased energy consumption, significant capacity constraints, and "over-rejection" (i.e., judging many SSDs that actually meet customer requirements as unqualified), significantly increasing production costs and delivery pressure.

[0004] However, related technologies lack a scientific and precise method to dynamically adjust and optimize internal production testing standards while ensuring that SSDs meet customers' real reliability requirements. Current practices rely on statically setting test thresholds based on experience, failing to quantitatively assess the degree of "passing the test" or to deduce the most economical and reasonable test parameter boundaries based on feedback data from actual customer usage scenarios. Therefore, establishing a data-driven production testing optimization system that dynamically balances "reliability" and "cost" has become a core technical problem urgently needing to be solved in this field. Summary of the Invention

[0005] To overcome the shortcomings mentioned above, this invention aims to provide a technical solution that addresses the aforementioned problems by providing a solid-state drive testing method, apparatus, and medium for big data storage.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The solid-state drive testing method for big data storage includes the following steps: S100 performs multi-dimensional production tests on solid-state drives, covering components, processes, functions, and reliability, and provides digital scoring for the results of each test item. S200. Based on the scores and preset weights of each test sub-item, calculate the overall quality score Q_total for each solid-state drive. S300. Extract production samples and verify them on a benchmark testing platform that simulates actual customer applications. Based on the verification results, establish a sample library labeled with the customer's standard compliance status and its corresponding Q_total and sub-item data. S400. Based on the Q_total distribution of “compliant” and “non-compliant” samples in the sample library, an iterative approximation algorithm is used to determine a customer standard critical comprehensive score Q_critical with a prediction misjudgment rate within an acceptable risk tolerance. S500: Analyze the statistical characteristics of critical sample groups located in the neighborhood of Q_critical on each test item, reverse deduce the new and optimized test pass threshold for each production test item, and integrate them to form a new production test standard specification. S600. Apply the new production testing standard to subsequent production testing, and continuously perform dynamic closed-loop updates to the testing standard through steps S300 to S500. The dynamic closed-loop update process includes a core optimization cycle and a supporting version management process.

[0007] As a further aspect of the present invention: step S100 specifically includes: S110, Component-level testing and scoring: Perform raw bad block count and raw bit error rate testing on the NAND flash memory chips of the SSD, perform basic function and power consumption testing on the main control chip, perform impedance and connectivity testing on the PCB substrate, and perform capacitance and frequency accuracy testing on the auxiliary capacitors and crystal oscillators. Record all test data and score them according to the deviation of each test result from the standard value using a preset first scoring model; The first scoring model converts the deviation of each test parameter into a sub-item score using a piecewise linear function based on the degree to which its measured value deviates from the standard value. The weighted sum of all relevant sub-item scores constitutes the final score of this component. S120. Process and structural testing and scoring: Perform AOI optical inspection on the SSD board after mounting and soldering to identify component misalignment, cold solder joints, missing components and record the coordinates and types of defects. Perform X-ray inspection to check the quality of internal solder joints, chip stack alignment, and the presence of voids; Perform online ICT circuit testing to verify the connectivity of all electrical nodes and the static parameters of key components; Perform a preliminary power-on test to confirm that the board can be recognized and enter the initialization state; Based on the number of defects detected, their severity level, and deviations in electrical parameters, a pre-set second scoring model is used for scoring; The second scoring model is a deduction system model, which sets a basic deduction value for each type of defect and multiplies it by the corresponding deduction coefficient according to the severity level of the defect. Test items with no defects or fully qualified parameters are given a benchmark full score. S130 Function and Firmware Testing and Scoring: Firmware flashing and verification are performed on SSD boards that have passed the S120 test to ensure firmware integrity. Perform functional logic testing to verify the correct response of all user command interfaces, security management functions, and basic read / write commands; Compatibility testing and initial performance screening were conducted on a standard testing platform, and the recognition success rate and initial sequential read / write bandwidth values ​​were recorded under different host interfaces. Based on the achievement of the standards, a pre-set third scoring model is used for scoring; The third scoring model uses Boolean scoring for binary pass / fail tests and converts the score for continuous variable tests by the percentage of the measured value relative to the target value. S140. Reliability Enhancement Test and Rating: Perform full-capacity sequential and random mode read and write tests on the assembled SSD, and record any errors or latency anomalies that occur during the process; Perform bad block retesting and management to verify the reliability of the bad block replacement mechanism; High-temperature aging tests were conducted in a high-temperature chamber to monitor steady-state operation. High and low temperature cyclic stress tests were conducted to examine the stability under sudden temperature changes. Simulate abnormal power outage scenarios and conduct multiple power outage protection tests to verify data consistency; Accelerated P / E cycle tests were performed under specific loads, and the performance degradation trend was recorded. Perform EMC electromagnetic compatibility tests and record the radiated and conducted interference values; Based on the error count, performance retention rate, number of passes, and interference values ​​exceeding the standard for each test item, a pre-set fourth scoring model is used for quantitative scoring. The fourth scoring model is based on the principle of statistical process control. It sets an acceptable "lower limit" for each reliability index. The score is based on the difference or ratio between the measured value and the lower limit, and is mapped through an exponential decay function to reflect the comprehensive assessment of early failures and long-term stability.

[0008] As a further aspect of the present invention: step S200 specifically includes: S210. Constructing the initial weight matrix: Based on product design specifications, historical fault analysis data, and expert experience, initial weights are assigned to the scores of each test sub-item in S110 to S140, forming a multi-dimensional initial weight matrix; among them, the test sub-items that have a decisive impact on the long-term reliability and data integrity of the SSD have higher weights than the test sub-items that have an impact on instantaneous performance or appearance and process. S220. Data normalization preprocessing: Before weighted calculation, the scores output in S110 to S140 are normalized to map all scores into a uniform numerical range, so as to eliminate the deviation caused by different scales or ranges of the scoring model to the comprehensive calculation results. S230. Calculate the weighted overall quality score: For each SSD that has completed all tests, the normalized sub-item scores are weighted and summed with the corresponding weight values ​​in the initial weight matrix to obtain the initial overall quality score of the SSD. S240. Introduce a batch correction factor: Calculate the average or median score of all SSDs in the current production batch on specific key test items, and compare it with the data of the historical baseline batch; based on the comparison deviation, generate a batch correction factor to fine-tune the initial comprehensive quality score to eliminate systematic deviations caused by material batch fluctuations or environmental drift, and finally obtain the comprehensive quality score for subsequent analysis.

[0009] As a further aspect of the present invention: step S300 specifically includes: S310. Formulate sampling strategy and sample preparation: From each production batch for which the comprehensive quality score Q_total is calculated by S200, extract physical samples SSD according to a preset sampling ratio and rules; the sampling rules ensure that the samples cover multiple score ranges from high Q_total to low Q_total within the batch, and give preference to SSDs whose Q_total is close to the customer standard critical comprehensive score Q_critical used in current production. S320, Building and Deploying a Benchmark Test Platform: Build a benchmark test platform independent of the production line. This platform consists of multiple test servers, racks, a temperature control system, and load generation and management software. Its hardware configuration, interface type, heat dissipation environment, and firmware driver version all simulate the actual application environment of the target big data storage customer. S330. Define a customer standard compliance test procedure: Perform long-term reliability operation tests on the benchmark test platform. The test procedure includes: S331, Load Model Test: Run a load script that simulates typical I / O modes of big data storage, including high-proportion sequential writes, garbage collection pressure, mixed read and write, and quiet periods, and record any errors, delays, or data consistency check failures that occur during the process. S332. Environmental stress test: Under the maximum operating ambient temperature specified in the customer's equipment specifications, continuously run the load model and monitor the steady-state performance and failure rate of the SSD. S333, Lifetime and Data Retention Test: The cumulative write volume reaches the TBW value specified in the customer's warranty terms by converting the write amplification factor, and data integrity is verified after completion; S340. Perform tests and determine compliance: Perform the complete test procedure defined in S330 on each sampled SSD; if no preset critical failure conditions are triggered during the test period and all data verifications pass, the sample is determined to "comply" with the customer's usage standards; otherwise, record the failure mode and time point, and determine it to "not comply". S350, Building and Updating the Sample Database: Create a record for each completed SSD sample, which includes at least: Sample ID, identifier of the source production batch, original comprehensive quality score Q_total calculated in S200, detailed scores or raw data of each key test sub-item in S100, failure mode and time data on the benchmark test platform, and the final "compliant" or "non-compliant" judgment label; all records constitute a continuously updated customer standard compliance sample library; S360. Dynamic Optimization of the Weight Matrix: Using the customer standard compliance sample library and its corresponding detailed sub-item scores established in S350 as training data, correlation analysis is used to analyze the correlation strength between each test sub-item score and the final customer standard compliance. Based on the analysis results, the weight values ​​in the initial weight matrix in S210 are periodically iteratively optimized to generate an optimized weight matrix. This optimized weight matrix will replace the initial weight matrix in S210 and be applied to the calculation of the overall quality score Q_total in the S200 step of subsequent production batches of SSDs, so as to continuously enhance the predictive ability of the overall quality score Q_total for customer standard compliance.

[0010] As a further aspect of the present invention: step S400 specifically includes: S410, Data Extraction and Interval Initialization: Extract the comprehensive quality score Q_total and its corresponding "compliant" or "non-compliant" label for all samples from the customer standard compliance sample library constructed in S350; calculate the minimum Q_total value Q_min_pass for "compliant" samples and the maximum Q_total value Q_max_fail for "non-compliant" samples, and use the interval [Q_max_fail, Q_min_pass] as the initial search interval; S420. Set approximation parameters and target: Preset an acceptable maximum prediction misclassification rate R_max as the convergence target. The misclassification rate includes the first type error rate of misclassifying actual "non-compliant" samples as qualified, and the second type error rate of misclassifying actual "compliant" samples as unqualified. S430. Execute the iterative approximation algorithm: Use an iterative algorithm to find the critical score threshold within the search interval. Each iteration includes: S431. Candidate threshold generation: Calculate a new candidate threshold Q_candidate based on the current search interval, preferably using the binary search method to find the midpoint of the interval; S432. False positive rate simulation calculation: Using the candidate threshold Q_candidate as the virtual factory pass line, that is, in the simulation, if Q_total ≥ Q_candidate, it is predicted as "pass", otherwise it is predicted as "fail"; traverse the sample library, count the number of samples whose prediction results do not match the real labels, calculate the first type error rate and the second type error rate respectively, and use the weighted sum and / or the maximum value of the two as the overall verification false positive rate of the current candidate threshold; S433. Search Interval Update and Convergence Judgment: Update the search interval based on simulation results: If the current overall verification misclassification rate is higher than R_max and is mainly caused by Type I errors, then increase the lower limit of the search interval; if it is mainly caused by Type II errors, then decrease the upper limit of the search interval; if the overall verification misclassification rate is lower than R_max, then record the current candidate threshold and its misclassification rate; the iteration process continues until the search interval width is less than the preset accuracy threshold ε. S440. Determine the final critical score: After the iteration converges, select the threshold with the lowest Q_total value from the candidate thresholds that meet the condition that the misclassification rate is lower than R_max, and formally determine it as the customer standard critical comprehensive score Q_critical. S450, Confidence Assessment and Data Augmentation Instruction: Based on the final determined Q_critical, assess its confidence level in the current sample database. If the confidence level is lower than the preset standard, generate an instruction to be fed back to step S310, requiring supplementary sampling of SSDs with Q_total values ​​in the vicinity of Q_critical in subsequent batches to enrich the data density of this critical area for use in the next round of optimization cycle.

[0011] As a further aspect of the present invention: step S500 specifically includes: S510. Define and screen the critical sample group: Taking the customer standard critical comprehensive score Q_critical determined in S440 as the center, set a neighborhood range ΔQ that fluctuates up and down. From the customer standard compliance sample library constructed in S350, screen all SSD samples whose comprehensive quality score Q_total falls within the interval [Q_critical - ΔQ, Q_critical + ΔQ] to form the critical sample group. S520. Analyze the sub-item data distribution of the critical sample group and determine the new threshold: For each independent production test sub-item defined in S100, an optimized test pass threshold is calculated based on the actual performance of the critical sample group. The specific operation is as follows: S521. Extracting and separating data: From the critical sample group defined in S510, extract the original measurement data of all samples on the specific test item; according to the “compliant” / “non-compliant” labels recorded in the sample library, the dataset is clearly separated into two subsets: the “compliant” sample data subset and the “non-compliant” sample data subset, and all data values ​​in the “compliant” sample data subset are sorted in ascending order according to their numerical values; S522. Define indicator types and select statistical strategies: Based on the physical nature represented by this test item, determine which of the following types it belongs to: Category A: Performance / Benefit Metrics: Metrics that measure a product’s “capability” or “health”, with the expectation that “the higher the better”. Their failure usually leads to performance degradation rather than immediate loss of functionality. Category B: Defect / Error Indicators: Indicators that measure manufacturing “flaws” or “potential failure points”. The desired direction is always “the lower the better”, and their deterioration is directly related to sudden functional failure. Choose a statistical strategy: For Category A indicators, analyze the subset of sample data that "meets" the criteria and calculate its P1 percentile, where P1 is the number less than 50. For Category B indicators, analyze the subset of sample data that "meets" the criteria and calculate its P2 percentile, where P2 is a number greater than 50. S523. Calculate and determine the recommended threshold: Perform the statistical calculations selected in S522; For Category A indicators, the calculated P1 percentile value will be directly used as the new suggested test pass threshold for that test sub-item. For Category B indicators, the calculated P2 percentile value will be used as the new recommended test pass threshold for that sub-item. S530. Convert the suggested test pass threshold into executable test parameters: For each test sub-item, based on its physical meaning and the requirements of the test equipment, convert the suggested test pass threshold obtained in S523 into specific parameters that can be directly called by the production line test program; the conversion includes: S531. For electrical parameter testing, convert the threshold to upper or lower limits for voltage, current, or timing; S532. For functional and performance testing, convert the threshold to minimum throughput, maximum permissible delay, or minimum bandwidth. S533. For reliability testing, the threshold is converted to the maximum permissible error count, the minimum performance retention rate, or the maximum permissible interference level. S540: Generate and output optimized test standard document: Integrate the specific parameters of all test sub-items after conversion by S530, and compile them into a structured optimized test standard specification document according to the order and format requirements of the production line test process; this document is used to directly update the configuration of the production line test system and serve as the basis for the inspection of subsequent batches of SSDs.

[0012] As a further aspect of the present invention: step S600 specifically includes: S610 Deploy and verify the new test standard: Import the optimized test standard specification file generated by S540 into the production line test system and replace the original test parameter thresholds; before the new standard is officially implemented, extract a small batch of trial production SSDs and use the old and new standards to conduct parallel testing and compare the results to confirm the effectiveness and safety of the new standard in engineering applications. S620, Monitoring and collecting production data under the new standard: After the optimized testing standard is officially applied to subsequent production batches, continue to execute the complete testing and scoring process from S100 to S200, and collect detailed sub-item scores, overall quality score Q_total and production pass rate data for all SSDs in this stage. S630. Evaluation of Optimization Effect and System Convergence Judgment: Based on the production data collected in S620, and combined with the continuously running S300 to S500 processes, the optimization effect is evaluated, and the judgment includes: S631. Compare whether the overall production pass rate before and after the implementation of the new standard has improved as expected; S632. Analyze whether the compliance rate of the new batch of samples on the S300 benchmark test platform is stably maintained at the target level; S633. Observe whether the customer standard critical comprehensive score Q_critical determined in the latest round of S400 tends to be stable and whether its fluctuation is less than the preset stability threshold. S640, Trigger standard iteration update or maintain the current standard: Make a decision based on the evaluation results of S630: If S633 indicates that the system has converged and the results of S631 and S632 meet the requirements, then maintain the current optimized test standard; otherwise, incorporate the new batch production data collected in S620 and its verification results under step S300 into the customer standard compliance sample library, and then automatically trigger a new round of dynamic threshold approximation starting from S400 and threshold reverse derivation process from S500 to generate and deploy the next-generation optimized test standard.

[0013] As a further aspect of the present invention: the dynamic closed-loop update process consists of a core optimization loop and a supporting management process, wherein the supporting management process specifically includes: S710. Each time S540 is executed to generate a new optimized test standard specification file, a version snapshot is created and stored synchronously. This snapshot is uniquely associated with and records: the identifier of this version standard, the data timestamp of the customer standard compliance sample library on which it is based, the version of the weight matrix used when generating the standard, the critical score Q_critical and its confidence interval, and the percentile parameter (P1, P2) combination used in S520. S720 Before performing parallel testing of the old and new S610 standards, the system automatically compares the changes in the thresholds of all test sub-items in the old and new versions, identifies the top K items with the "largest relaxation" and the "only tightened" item, forms a "List of Key Change Items", and prioritizes the verification resources of S610 to conduct in-depth testing and risk confirmation of the items on this list. During the S730 and S620 monitoring process, the system monitors two major indicators in real time: 1) The first pass rate of any test sub-item has decreased significantly relative to the historical average; 2) A cluster of failure modes that were not previously recorded have appeared in the S300 benchmark test. Once a monitoring alarm is triggered, the system can automatically suspend the widespread application of the new standard and trigger root cause analysis, while providing an operation interface for rolling back to the previous stable version of the standard. The evaluation reports for S740 and S630 need to integrate core indicators, the preliminary review summary of the impact of S720, and the runtime monitoring records of S730 to form a comprehensive decision report. The system automatically provides decision suggestions based on preset strategies, and the entire "report-approval-execution" process generates tamper-proof audit logs, which are stored in association with the version snapshot of S710.

[0014] An apparatus includes: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described solid-state drive testing method for big data storage.

[0015] A storage medium comprising a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the above-described solid-state drive testing method for big data storage.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention performs multi-dimensional production testing covering components, processes, functions, and reliability, and digitally scores the results (S100). Based on preset weights, it calculates a comprehensive quality score Q_total (S200). Then, it extracts production samples for verification on a benchmark testing platform simulating actual customer applications, establishing a sample library that labels customer standard compliance status and its corresponding Q_total and sub-item data (S300). Based on this, it analyzes the Q_total distribution of "compliant" and "non-compliant" samples in the sample library using an iterative approximation algorithm, determining the customer standard critical comprehensive score Q_critical (S400) within an acceptable risk tolerance range for predicted misjudgment rate. Finally, it performs reverse analysis to determine the position of Q_total within the Q_critical range. The statistical characteristics of critical sample groups within the ical neighborhood on each test sub-item are used to derive the optimized test pass threshold for each production test sub-item. These are then integrated to form a new production test standard specification (S500), which is applied to subsequent production tests. Simultaneously, dynamic closed-loop updates are continuously achieved through sample verification and threshold optimization steps (S600). The basic principle is to accurately align internal test standards with actual customer needs in a data-driven manner. By dynamically adjusting test parameters through statistical modeling and algorithm iteration, the "pass test" problem is scientifically quantified and optimized. Under the premise of ensuring that solid-state drives meet the real reliability requirements of customers, excessive obsolescence is significantly reduced, test costs and time are lowered, and an adaptive balance between economy and reliability is achieved in the production test standards. Attached Figure Description

[0017] Figure 1 This is a flowchart of steps S100-S600 in this invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 A testing method for solid-state drives used for big data storage includes the following steps: S100 performs multi-dimensional production tests on solid-state drives, covering components, processes, functions, and reliability, and provides digital scoring for the results of each test item. S200. Based on the scores and preset weights of each test sub-item, calculate the overall quality score Q_total for each solid-state drive. S300. Extract production samples and verify them on a benchmark testing platform that simulates actual customer applications. Based on the verification results, establish a sample library labeled with the customer's standard compliance status and its corresponding Q_total and sub-item data. S400. Based on the Q_total distribution of “compliant” and “non-compliant” samples in the sample library, an iterative approximation algorithm is used to determine a customer standard critical comprehensive score Q_critical with a prediction misjudgment rate within an acceptable risk tolerance. S500: Analyze the statistical characteristics of critical sample groups located in the neighborhood of Q_critical on each test item, reverse deduce the new and optimized test pass threshold for each production test item, and integrate them to form a new production test standard specification. S600. Apply the new production testing standard to subsequent production testing, and continuously update the testing standard dynamically in a closed loop through steps S300 to S500. The dynamic closed loop update process includes a core optimization loop (S610-S640) and a supporting version management process (S800 series steps). The problem of static solidification of traditional SSD production testing standards is solved by building a data-driven dynamic closed-loop optimization system. First, multi-dimensional production tests covering components, processes, functions, and reliability are performed, and the results of each sub-item are digitized and scored (S100) to lay the foundation for quantitative analysis; Then, based on the preset weights, the comprehensive quality score Q_total (S200) is calculated to achieve a normalized representation of multidimensional performance; Then, production samples are extracted and verified on a benchmark testing platform that simulates real application scenarios, generating a sample library (S300) labeled with customer standard compliance status and its corresponding Q_total and details of each sub-item, thereby building an objective bridge between internal test data and external customer needs. Based on this, the Q_total statistical distribution of the “compliant” and “non-compliant” groups in the sample library is analyzed by iterative approximation algorithm, and the critical comprehensive score Q_critical (S400) of the customer standard with the prediction misjudgment rate within the acceptable risk is accurately calculated. This step scientifically locates the theoretical boundary that the test standard can be relaxed. Next, we deeply analyze the data characteristics of the critical sample group in the neighborhood of Q_critical on each test item, reversely deduce the new and optimized pass threshold for each item, and integrate them into a new production test specification (S500). This process realizes accurate traceability and optimization from "overall results" to "individual parameters". Finally, the new specifications are applied to subsequent tests, and dynamic closed-loop updates (S600) are performed by continuously executing steps S300 to S500, enabling the system to have adaptive evolution capabilities. First, it changed the paradigm of test standard setting, shifting from static setting based on experience to dynamic optimization based on actual customer data. Through S300 benchmark verification and S400 threshold calculation, it achieved precise quantification and calibration of the "pass test" level. Second, it brought significant economic benefits. The optimized thresholds derived through reverse derivation in the S500 steps can directly reduce "over-elimination," allowing a large number of SSDs that actually meet customer requirements but were misjudged by the original stringent standards to pass the test and be shipped out. This directly reduced material scrap costs. At the same time, the shortened testing time due to optimized test thresholds (especially in high-intensity reliability tests) released testing capacity and reduced energy consumption. First, it can save huge amounts of costs in large-scale production. Second, it improves the accuracy and foresight of quality control. By continuously tracking critical samples (S600), the system can adaptively adjust the test boundaries according to changes in customer usage patterns or supply chain processes, which not only continuously ensures the bottom line of reliability but also avoids unnecessary test redundancy. Third, it provides a complete decision support system that integrates the originally separate production test data, customer verification data, and statistical models, enabling enterprises to continuously fine-tune their test strategies based on empirical data, and ultimately achieve a balance between quality, cost, and delivery efficiency in SSD production within a strictly controllable field failure risk tolerance. This invention performs multi-dimensional production testing covering components, processes, functions, and reliability, and digitally scores the results (S100). Based on preset weights, it calculates a comprehensive quality score Q_total (S200). Then, it extracts production samples for verification on a benchmark testing platform simulating actual customer applications, establishing a sample library that labels customer standard compliance status and its corresponding Q_total and sub-item data (S300). Based on this, it analyzes the Q_total distribution of "compliant" and "non-compliant" samples in the sample library using an iterative approximation algorithm, determining the customer standard critical comprehensive score Q_critical (S400) within an acceptable risk tolerance range for predicted misjudgment rate. Finally, it performs reverse analysis to determine the position of Q_total within the Q_critical range. The statistical characteristics of critical sample groups within the ical neighborhood on each test sub-item are used to derive the optimized test pass threshold for each production test sub-item. These are then integrated to form a new production test standard specification (S500), which is applied to subsequent production tests. Simultaneously, dynamic closed-loop updates are continuously achieved through sample verification and threshold optimization steps (S600). The basic principle is to accurately align internal test standards with actual customer needs in a data-driven manner. By dynamically adjusting test parameters through statistical modeling and algorithm iteration, the "pass test" problem is scientifically quantified and optimized. Under the premise of ensuring that solid-state drives meet the real reliability requirements of customers, excessive obsolescence is significantly reduced, test costs and time are lowered, and an adaptive balance between economy and reliability is achieved in the production test standards.

[0020] In this embodiment of the invention, step S100 specifically includes: S110, Component-level testing and scoring: Perform raw bad block count and raw bit error rate testing on the NAND flash memory chips of the SSD, perform basic function and power consumption testing on the main control chip, perform impedance and connectivity testing on the PCB substrate, and perform capacitance and frequency accuracy testing on the auxiliary capacitors and crystal oscillators. Record all test data and score them according to the deviation of each test result from the standard value using a preset first scoring model; The first scoring model converts the deviation of each test parameter into a sub-item score using a piecewise linear function based on the degree to which its measured value deviates from the standard value. The weighted sum of all relevant sub-item scores constitutes the final score of this component. S120. Process and structural testing and scoring: Perform AOI optical inspection on the SSD board after mounting and soldering to identify component misalignment, cold solder joints, missing components and record the coordinates and types of defects. Perform X-ray inspection to check the quality of internal solder joints, chip stack alignment, and the presence of voids; Perform online ICT circuit testing to verify the connectivity of all electrical nodes and the static parameters of key components; Perform a preliminary power-on test to confirm that the board can be recognized and enter the initialization state; Based on the number of defects detected, their severity level, and deviations in electrical parameters, a pre-set second scoring model is used for scoring; The second scoring model is a deduction system model, which sets a basic deduction value for each type of defect and multiplies it by the corresponding deduction coefficient according to the severity level of the defect. Test items with no defects or fully qualified parameters are given a benchmark full score. S130 Function and Firmware Testing and Scoring: Firmware flashing and verification are performed on SSD boards that have passed the S120 test to ensure firmware integrity. Perform functional logic testing to verify the correct response of all user command interfaces, security management functions, and basic read / write commands; Compatibility testing and initial performance screening were conducted on a standard testing platform, and the recognition success rate and initial sequential read / write bandwidth values ​​were recorded under different host interfaces. Based on the achievement of the standards, a pre-set third scoring model is used for scoring; The third scoring model uses Boolean scoring for binary pass / fail tests (such as firmware verification and command response), and converts the score for continuous variable tests (such as calibration accuracy and performance bandwidth) by the percentage of the measured value relative to the target value. S140. Reliability Enhancement Test and Rating: Perform full-capacity sequential and random mode read and write tests on the assembled SSD, and record any errors or latency anomalies that occur during the process; Perform bad block retesting and management to verify the reliability of the bad block replacement mechanism; High-temperature aging tests were conducted in a high-temperature chamber to monitor steady-state operation. High and low temperature cyclic stress tests were conducted to examine the stability under sudden temperature changes. Simulate abnormal power outage scenarios and conduct multiple power outage protection tests to verify data consistency; Accelerated P / E cycle tests were performed under specific loads, and the performance degradation trend was recorded. Perform EMC electromagnetic compatibility tests and record the radiated and conducted interference values; Based on the error count, performance retention rate, number of passes, and interference values ​​exceeding the standard for each test item, a pre-set fourth scoring model is used for quantitative scoring. The fourth scoring model is based on the principle of statistical process control. It sets an acceptable "lower limit" for each reliability index. The score is based on the difference or ratio between the measured value and the lower limit, and is mapped through an exponential decay function to reflect the comprehensive assessment of early failure and long-term stability. This step establishes a comprehensive and quantifiable quality data collection system, providing a refined data foundation for subsequent intelligent decision-making. Production testing in related technologies often only records a binary result of "pass / fail". However, this solution uses steps S110 to S140 to perform parametric measurement and digital scoring on the reliability of each link from component materials to finished products. For example, at the component level (S110), not only is the quality of NAND chips detected, but the quality gradient of materials is accurately characterized by the measurement of continuous variables such as the original bit error rate and the scoring of piecewise linear functions. This helps to trace the impact of supply chain fluctuations on the final quality. This deep data processing makes each SSD have a complete "quality check-up report", rather than just a final qualified label. Secondly, its differentiated scoring models designed for different testing dimensions enhance the scientific rigor and guidance of quality assessment. The solution does not employ a "one-size-fits-all" scoring method, but rather intelligently matches the model to the testing objectives: for process defects (S120), a deduction system is used, directly linked to production yield, clearly pinpointing weaknesses in the manufacturing process; for functional testing (S130), a combination of Boolean and percentage scoring is used, balancing the rigidity of basic functions with the flexibility of performance parameters; for reliability testing (S140), an exponential decay function scoring based on statistical process control is used, sensitively reflecting the early failure risk of a product on the "borderline" of acceptable quality, rather than simply judging whether it passes a fixed threshold. This scientific scoring transforms quality from a "black box" to a "white box," enabling managers to accurately understand which aspect—"materials," "process," "function," or "long-term reliability"—is lowering the overall quality score, thus allowing for targeted improvements. Furthermore, this solution directly enables refined management of production costs and optimization of testing efficiency; Because the scoring is continuous and itemized, it allows for more flexible decision-making on the production side. For example, an SSD with a high overall score (Q_total) but a slightly lower score for a particular component may not need to be scrapped directly as in the traditional "one-vote veto" system. Instead, it can be downgraded for use in scenarios with lower requirements, thereby reducing material waste (for example, in big data storage, the performance of SSDs used in cold data storage functional areas is significantly lower than that in hot data storage functional areas). At the same time, detailed sub-item score data provides key input for the reverse derivation of the "optimized test pass threshold" in subsequent steps (S500): the system can analyze whether an SSD near the critical quality (Q_critical) is struggling in a specific sub-score, such as "high temperature aging score" or "random write error count," thereby accurately and specifically relaxing unnecessarily stringent parameter thresholds in that test, rather than blindly lowering the overall standard. This makes the allocation of testing resources more precise, shortening the testing time of non-critical items while ensuring that critical reliability is not compromised, directly improving production capacity and energy efficiency. Ultimately, this solution achieves a closed loop of "prevention" and "continuous optimization" in quality control. By digitally recording and scoring the defect types and coordinates in process testing (S120), it can be correlated with production equipment parameters to achieve rapid location of production line problems and preventive adjustment of process parameters, reducing batch risks. After all scoring data is incorporated into the sample library, it forms fuel for continuous optimization: when customer application scenarios change (such as changes in compatibility requirements of new operating systems), the system can retrospectively analyze the correlation between the scoring data of S130 (functional and firmware testing) and the latest customer verification results, and adjust the scoring weights of relevant test items or pass thresholds in a timely manner, so that production testing standards can dynamically adapt to real market demands. This invention conducts a comprehensive digital assessment of solid-state drive (SSD) production quality through four hierarchical and model-based testing stages: First, at the component level (S110), parametric testing is performed on core materials such as NAND and controller, and a piecewise linear function scoring model is used to quantify the deviation between measured values ​​and standard values ​​into scores, thereby accurately assessing the quality dispersion of individual materials; Second, at the process and structure level (S120), manufacturing defects are detected through multiple methods such as AOI and X-ray, and a scoring system based on defect type and severity level is used for scoring, achieving strict control over the consistency of the production process; Subsequently, at the functional and firmware level (S130), firmware logic and basic performance are verified, and a model combining Boolean scoring and percentage conversion is used for scoring, ensuring the completeness of the product's basic functions and interfaces; Finally, at the reliability enhancement level (S140), various stresses and long-term usage scenarios are simulated, and an exponential decay function model based on statistical process control principles is used for scoring, thereby quantitatively assessing the long-term stability and robustness of the product; First, it establishes a well-structured, multi-dimensional testing framework covering the entire product lifecycle, systematically integrating previously isolated, qualitative test items into a quantifiable and traceable data stream, providing a solid and detailed underlying data foundation for subsequent comprehensive quality score (S200) calculations. Second, its differentiated scoring models (such as piecewise linear, deduction system, exponential decay, etc.) specifically designed for different test dimensions (materials, processes, functions, reliability) can more scientifically and fairly reflect the impact of various test indicators (from continuous parameters to discrete defects) on the final quality, avoiding the bias caused by a single scoring method, and making the scoring results more interpretable and guiding. Third, this fine-grained digital scoring method not only realizes a "panoramic portrait" of the production quality of each SSD, but also provides a key data correlation foundation for subsequent steps (such as the reverse derivation of sub-item optimization thresholds in S500), enabling optimization to accurately locate specific test parameters or process links, thereby significantly improving the effectiveness and accuracy of the entire dynamic optimization system.

[0021] In this embodiment of the invention, step S200 specifically includes: S210. Constructing the initial weight matrix: Based on product design specifications, historical fault analysis data, and expert experience, initial weights are assigned to the scores of each test sub-item in S110 to S140, forming a multi-dimensional initial weight matrix; among them, the test sub-items that have a decisive impact on the long-term reliability and data integrity of the SSD have higher weights than the test sub-items that have an impact on instantaneous performance or appearance and process. S220. Data normalization preprocessing: Before weighted calculation, the scores output in S110 to S140 are normalized to map all scores into a uniform numerical range, so as to eliminate the deviation caused by different scales or ranges of the scoring model to the comprehensive calculation results. S230. Calculate the weighted overall quality score: For each SSD that has completed all tests, the normalized sub-item scores are weighted and summed with the corresponding weight values ​​in the initial weight matrix to obtain the initial overall quality score of the SSD. S240. Introduce a batch correction factor: Calculate the average or median score of all SSDs in the current production batch on specific key test items, and compare it with the data of historical baseline batches; based on the comparison deviation, generate a batch correction factor to fine-tune the initial comprehensive quality score to eliminate systematic bias caused by material batch fluctuations or environmental drift, and finally obtain the comprehensive quality score (Q_total) for subsequent analysis. Through a multi-step, self-correcting comprehensive scoring algorithm, the multi-dimensional and heterogeneous test sub-item scores generated in claim 2 are scientifically combined into a highly consistent and comparable comprehensive quality index (Q_total): First, an initial weight matrix is ​​constructed based on design specifications, historical failures, and expert knowledge (S210) to ensure that the scoring system focuses on key quality attributes that have a decisive impact on reliability; then, data normalization processing (S220) eliminates differences in the dimensions and scope of different sub-item scores, laying the foundation for fair weighted calculation; furthermore, a weighted summation calculation is performed to obtain the initial comprehensive score (S230), realizing the transformation from scattered indicators to a single quantitative evaluation; finally, a batch correction factor (S240) is introduced, which is generated based on the ratio or difference between the median scores of the current production batch and the historical baseline batch on key test sub-items; for example, if the ratio method is used, the factor is (historical baseline median / The initial overall quality score is fine-tuned by multiplying the median of the current batch by this factor to eliminate systematic bias. By comparing the overall performance of the current batch with the historical baseline on key indicators, the initial score is fine-tuned to remove interference from systematic fluctuations such as materials or environment, thereby outputting a pure final Q_total that reflects the inherent quality of the individual product. First, it establishes a structured and interpretable decision-making model. The differentiated weighting in S210 ensures that the overall score accurately reflects the quality strategy of "long-term reliability taking precedence over instantaneous performance," directly aligning the score with the ultimate business objective (low failure rate). Second, the normalization process in S220 solves the core technical challenge, enabling unbiased mathematical calculations and horizontal comparisons of scores from different scoring models (such as deduction systems and exponential decay functions) with vastly different numerical ranges, ensuring the internal fairness of the scoring system and the consistency of the calculation results. Third, the most innovative step, S240, introduces a "batch-dimensional" quality perspective, which not only focuses on the absolute score of a single SSD but also perceives the overall quality of the entire batch. The system addresses the collective quality drift of individual production batches. By using a correction factor for fine-tuning, it can automatically offset the general impact of slightly inferior inherent performance of a particular batch of NAND flash memory or fluctuations in ambient temperature and humidity during the week. This makes the final Q_total more accurately represent the relative quality level of individual components during the manufacturing process, improving the comparability of quality scores between SSDs produced in different batches and at different times. This provides a stable and reliable input data foundation for subsequent steps (such as determining the global quality critical value Q_critical in S400), enabling elimination decisions and standard optimizations based on Q_total to eliminate batch-specific noise interference and truly focus on the optimization space of product design and manufacturing itself.

[0022] In this embodiment of the invention, step S300 specifically includes: S310. Formulate sampling strategy and sample preparation: From each production batch for which the comprehensive quality score (Q_total) is calculated via S200, physical samples SSD are drawn according to a preset sampling ratio and rules; the sampling rules ensure that the samples cover multiple score ranges from high Q_total to low Q_total within the batch, and that SSDs with Q_total close to the customer standard critical comprehensive score (Q_critical) used in current production are sampled with emphasis. S320, Building and Deploying a Benchmark Test Platform: Build a benchmark test platform independent of the production line. This platform consists of multiple test servers, racks, a temperature control system, and load generation and management software. Its hardware configuration, interface type, heat dissipation environment, and firmware driver version all simulate the actual application environment of the target big data storage customer. S330. Define a customer standard compliance test procedure: Perform long-term reliability operation tests on the benchmark test platform. The test procedure includes: S331, Load Model Test: Run a load script that simulates typical I / O modes of big data storage, including high-proportion sequential writes, garbage collection pressure, mixed read and write, and quiet periods, and record any errors, delays, or data consistency check failures that occur during the process. S332. Environmental stress test: Under the maximum operating ambient temperature specified in the customer's equipment specifications, continuously run the load model and monitor the steady-state performance and failure rate of the SSD. S333, Lifetime and Data Retention Test: The cumulative write volume reaches the TBW value specified in the customer's warranty terms by converting the write amplification factor. After completion, data integrity is verified. The cumulative write volume is calculated on the benchmark test platform in accelerated write mode by converting the write amplification factor until the TBW value specified in the customer's warranty terms is reached. This test should be completed within a reasonable test duration calculated by the acceleration factor within the customer's expected product lifespan. S340. Perform tests and determine compliance: Perform the complete test procedure defined in S330 on each sampled SSD; if no preset critical failure conditions are triggered during the test period and all data verifications pass, the sample is determined to "comply" with the customer's usage standards; otherwise, record the failure mode and time point, and determine it to "not comply". S350. Building and Updating the Sample Database: Create a record for each completed SSD sample. This record should include at least: Sample ID, identifier of the source production batch, original comprehensive quality score Q_total calculated in S200, detailed scores or raw data of each key test sub-item in S100, failure mode and time data on the benchmark test platform (empty if there are no failures), and the final "compliant" or "non-compliant" judgment label. All records constitute a continuously updated customer standard compliance sample database. S360. Dynamic optimization of the weight matrix: Using the customer standard compliance sample library and its corresponding detailed sub-item scores established in S350 as training data, correlation analysis is used to analyze the correlation strength between each test sub-item score and the final customer standard compliance. Based on the analysis results, the weight values ​​in the initial weight matrix in S210 are periodically iteratively optimized to generate an optimized weight matrix. This optimized weight matrix will replace the initial weight matrix in S210 and be applied to the calculation of the overall quality score (Q_total) of subsequent production batches of SSDs in step S200, so as to continuously enhance the predictive ability of the overall quality score (Q_total) for customer standard compliance. This invention constructs a core verification and learning closed loop connecting internal production and external customer scenarios: First, it establishes an authoritative "customer truth bank," providing an irreplaceable decision-making anchor and risk isolation barrier for the entire optimization system. Through the highly simulated benchmark testing platform of S320 and the test procedures of S330 defined based on the customer's actual load, environment, and lifespan requirements, the system essentially reproduces the key operating conditions of the customer's data center within the factory. Enterprises no longer need to transform vague "customer requirements" into internal specifications that may be over-interpreted, but instead have an objective and repeatable "fact standard." Any adjustment to internal production test standards (such as threshold optimization in S500) must be based on empirical results of "compliance" or "non-compliance" on this platform, thereby fundamentally eliminating the major business risk of subjectively lowering quality standards and causing batch field failures, and providing a safe "testing ground" for subsequent optimization. Second, the intelligent sampling strategy (S310) achieves precision and efficiency in verifying resource investment, improving the input-output ratio of the optimization process. Traditional sampling may be random or uniform, while this solution explicitly requires coverage of the entire score range and focuses on extracting "critical" samples where Q_total is close to the current qualified threshold. The "current qualified production threshold" refers to the comprehensive quality score (Q_total) boundary corresponding to the fixed internal factory test standards currently in use on the company's original production line before the implementation of this patented optimization method. It concentrates valuable long-term reliability testing resources (testing cycles may last for several weeks) on the product group most likely to expose problems and with the most decision ambiguity. On the one hand, it avoids wasting testing resources on samples in high score ranges (almost certainly compliant) and low score ranges (almost certainly non-compliant). On the other hand, it ensures that the final sample library (S350) has a sufficiently high data density in the "critical zone" that determines the quality boundary, making the customer standard critical score Q_critical determined by the algorithm (S400) statistically more robust and accurate, thus making the optimization decision more reliable. Third, the constructed structured sample database (S350) is the core data asset driving the intelligence of the entire system. Its value is not just a simple test record; each record in this database is a complete "production-verification" control experiment, which closely links the "cause" (score) of dozens of test sub-items in the production end with the final "effect" (compliance label and failure mode) in the customer scenario. The profound benefits of this design include: 1) Supporting root cause analysis: When a sample is judged to be "non-compliant", all its production test data can be traced back immediately to quickly locate which component score (S110), which type of process defect (S120), or which reliability test (S140) weakness caused the field failure, driving targeted process improvement; 2) Providing precise "targets" for reverse engineering: In the S500 step, the system needs to analyze the characteristics of samples near Q_critical to optimize the test threshold. The detailed, one-to-one corresponding sub-item data provided in this step is the "map" for performing this reverse engineering, enabling optimization to be precise to specific parameters. Fourth, the dynamic optimization mechanism of the weight matrix introduced in step S360 enables the self-evolution of the system's predictive capabilities, forming the first optimized sub-loop of the internal quality evaluation system. In traditional production testing, the weights of each item remain unchanged for a long time once set. However, this solution utilizes sample library data and conducts correlation analysis to continuously evaluate the actual correlation strength between each internal test item (such as "high temperature aging score" and "original bit error rate") and the final customer's compliance, and adjusts its weight in calculating Q_total accordingly. This enables the internal quality scoring system to "learn" the customer's real concerns. For example, if data analysis finds that the "abnormal power failure test score" is highly correlated with data corruption in the customer's scenario, the system will automatically increase the weight of this sub-item, so that Q_total gives lower scores to SSDs that perform poorly in this item, thus enabling stricter screening at the production end. This dynamic adjustment ensures that the evaluation standard of the comprehensive quality score Q_total is always consistent with the customer's real experience, continuously improving its predictive accuracy and guiding role in the long-term reliability of the product.

[0023] In this embodiment of the invention, step S400 specifically includes: S410, Data Extraction and Interval Initialization: Extract the comprehensive quality score Q_total and its corresponding "compliant" or "non-compliant" label for all samples from the customer standard compliance sample library constructed in S350; calculate the minimum Q_total value Q_min_pass for "compliant" samples and the maximum Q_total value Q_max_fail for "non-compliant" samples, and use the interval [Q_max_fail, Q_min_pass] as the initial search interval; S420. Set approximation parameters and target: Preset an acceptable maximum prediction misclassification rate R_max as the convergence target. The misclassification rate includes the first type error rate of misclassifying actual "non-compliant" samples as qualified, and the second type error rate of misclassifying actual "compliant" samples as unqualified. S430. Execute the iterative approximation algorithm: Use an iterative algorithm to find the critical score threshold within the search interval. Each iteration includes: S431. Candidate threshold generation: Calculate a new candidate threshold Q_candidate based on the current search interval, preferably using the binary search method to find the midpoint of the interval; S432. False positive rate simulation calculation: Using the candidate threshold Q_candidate as the virtual factory pass line, that is, in the simulation, if Q_total ≥ Q_candidate, it is predicted as "pass", otherwise it is predicted as "fail"; traverse the sample library, count the number of samples whose prediction results do not match the real labels, calculate the first type error rate and the second type error rate respectively, and use the weighted sum and / or the maximum value of the two as the overall verification false positive rate of the current candidate threshold; S433. Search Interval Update and Convergence Judgment: Update the search interval based on simulation results: If the current overall verification misclassification rate is higher than R_max and is mainly caused by Type I errors, then increase the lower limit of the search interval; if it is mainly caused by Type II errors, then decrease the upper limit of the search interval; if the overall verification misclassification rate is lower than R_max, then record the current candidate threshold and its misclassification rate; the iteration process continues until the search interval width is less than the preset accuracy threshold ε. S440. Determine the final critical score: After the iteration converges, select the threshold with the lowest Q_total value from the candidate thresholds that meet the condition that the misclassification rate is lower than R_max, and formally determine it as the customer standard critical comprehensive score Q_critical. This selection principle aims to ensure that production cost pressure is reduced to the greatest extent. S450, Confidence Assessment and Data Augmentation Instruction: Based on the final determined Q_critical, assess its confidence level in the current sample database; if the confidence level is lower than the preset standard, generate an instruction to feed back to step S310, requiring supplementary sampling of SSDs with Q_total values ​​near Q_critical in subsequent batches to enrich the data density of this critical area for use in the next round of optimization cycle; First, it achieves a fundamental shift in quality risk from "qualitative concern" to "quantitative control," providing management with clear decision-making basis. In traditional methods, "how much to relax testing standards to ensure safety" is a vague judgment based on experience, lacking data support and carrying high liability risks. This solution uses S420 to preset the acceptable maximum predicted false positive rate (R_max) and decomposes it into Type I error (releasing defective products) and Type II error (falsely rejecting good products). This makes "risk" a quantitative indicator that can be defined in advance and calculated precisely (e.g., allowing a risk of 0.05% defective products leaving the site). The S430 iterative algorithm essentially performs thousands of hypothesis tests automatically: for each candidate threshold (Q_candidate), it immediately simulates the consequences of applying the standard using all historical sample data and calculates the precise misjudgment rate. This ultimately ensures that the determined Q_critical value is not an "estimate," but a "scientific boundary" with known, controllable risk probabilities. Business managers can then make business decisions based on a clear "risk-cost" curve, rather than taking risks in the unknown. The prediction result is determined by the algorithm in the current iteration based on the candidate threshold Q_candidate. Temporary simulation generation is a "hypothetical" factory judgment. The algorithm specifies in the simulation that if we set the factory pass score of the production line as the current Q_candidate, then for each sample in the sample library, we only rely on its existing comprehensive quality score. The real label comes entirely from the execution result of step S340. It is the final and objective judgment given for each sampled sample SSD after long-term and rigorous verification on the "benchmark test platform" according to the "customer standard compliance test procedure". The precision threshold ε is a preset, extremely small positive number that defines the termination condition of the iterative search algorithm when searching for the customer standard critical comprehensive score Q_critical. Second, while adhering to the bottom line of risk, it achieves extreme cost optimization, directly creating economic benefits. The algorithm's goal is not simply to find any threshold that meets the false positive rate requirement, but rather to select the lowest Q_total from all feasible solutions as Q_critical through the S440 rule. This strategy has a strong economic intention: it means that while ensuring that the failure rate at the customer's site does not exceed R_max, the internal factory acceptance line is relaxed as much as possible, allowing more SSDs that would otherwise be eliminated by the stricter old standards to pass the factory. For a company with an annual output of tens of millions of units, even if the overall acceptance rate is increased by less than one percentage point, tens of thousands of unnecessary SSDs can be avoided from being scrapped. The savings in material costs, manufacturing costs, and related testing costs are enormous. Therefore, this step is the core link in directly extracting "profit" from "quality data." Third, its built-in confidence assessment and feedback mechanism ( S450 constructs a lean, self-optimizing decision-making system. The algorithm clearly recognizes the uncertainty inherent in decisions made based on finite samples. The S450 step evaluates the confidence level of Q_critical and automatically generates supplementary sampling instructions for the interval near Q_critical when the confidence level is insufficient. This is a key innovation, avoiding the risk of accumulated bias that may arise from the "one-time setting, long-term use" approach in traditional methods. This feedback mechanism ensures that the system can intelligently identify its own knowledge weaknesses (i.e., insufficient data in the critical region) and guide the "precise deployment" of subsequent validation resources (expensive benchmark tests) to enrich the sample library in the most efficient way. This forms a reinforcing loop of "decision-evaluation-learning-re-decision," which allows the Q_critical value to continuously converge to the most realistic and robust value as production and validation continue, and the overall decision quality of the system continuously improves over time.

[0024] In this embodiment of the invention, step S500 specifically includes: S510. Define and screen the critical sample group: Using the customer standard critical comprehensive score Q_critical determined in S440 as the center, set a neighborhood range (ΔQ) that fluctuates up and down. From the customer standard compliance sample library constructed in S350, screen all SSD samples whose comprehensive quality score Q_total falls within the interval [Q_critical - ΔQ, Q_critical + ΔQ] to form the critical sample group. S520. Analyze the sub-item data distribution of the critical sample group and determine the new threshold: For each independent production test sub-item defined in S100, an optimized test pass threshold is calculated based on the actual performance of the critical sample group. The specific operation is as follows: S521. Extracting and separating data: From the critical sample group defined in S510, extract the original measurement data (or scores) of all samples on the specific test item; based on the "compliant" / "non-compliant" labels recorded in the sample library, clearly separate the dataset into two subsets: the "compliant" sample data subset and the "non-compliant" sample data subset, and sort all data values ​​in the "compliant" sample data subset in ascending order of numerical value; S522. Define indicator types and select statistical strategies: Based on the physical nature represented by this test item, determine which of the following types it belongs to: Category A: Performance / Benefit Metrics: Metrics that measure a product’s “capability” or “health”, with the expectation that “the higher the better” (such as read and write speed). Their failure usually leads to performance degradation rather than immediate loss of functionality. Category B: Defect / Error Indicators: Indicators that measure manufacturing “flaws” or “potential failure points”, with the expectation that the lower the better (e.g., weld void area, number of short circuits), their deterioration is directly related to sudden functional failure; Choose a statistical strategy: For Category A (performance / benign) metrics, we will focus on analyzing the subset of “compliant” sample data and calculate its P1 percentile, where P1 is a number less than 50, such as 20. The significance of this value is that the performance value of the sample ranked in the worst 20% among those SSDs that are ultimately verified as compliant in the critical group will be used as the new standard. In this way, all SSDs that perform better than this “poor student” (accounting for 80% of the “compliant” samples) will easily meet the new standard. For Category B (defects / errors) metrics, we also focus on analyzing the subset of “compliant” sample data, but we calculate its P2 percentile, where P2 is a number greater than 50, such as 90. The significance of this value is: among those SSDs that are ultimately verified as compliant in the critical group, the performance value of the sample ranked in the worst 90% is used as the new standard. In this way, all SSDs that perform better than this “poor student” (accounting for 90% of the “compliant” samples) will easily meet the new standard. S523. Calculate and determine the recommended threshold: Perform the statistical calculations selected in S522; For Category A indicators, the calculated P1 percentile value will be directly used as the new suggested test pass threshold for that test sub-item. This means we will relax our internal standards to a level where the performance of "near-qualified products" is relatively poor (below P1%) but still acceptable; this will significantly reduce the number of products that are phased out due to an excessive pursuit of high performance. For Category B indicators, the calculated P2 percentile value will be used as the new recommended test pass threshold for that sub-item. This means that we have defined a defect tolerance cap that covers the vast majority (P2%) of "near-qualified products". Within the empirically safe range, this level of defects is allowed to improve yield. S530. Convert the suggested test pass threshold into executable test parameters: For each test sub-item, based on its physical meaning and the requirements of the test equipment, convert the suggested test pass threshold obtained in S523 into specific parameters that can be directly called by the production line test program; the conversion includes: S531. For electrical parameter testing, convert the threshold to upper or lower limits for voltage, current, or timing; S532. For functional and performance testing, convert the threshold to minimum throughput, maximum permissible delay, or minimum bandwidth. S533. For reliability testing, the threshold is converted to the maximum permissible error count, the minimum performance retention rate, or the maximum permissible interference level. S540: Generate and output optimized test standard document: Integrate the specific parameters of all test sub-items after conversion by S530, and compile them into a structured optimized test standard specification document according to the order and format requirements of the production line test process; this document is used to directly update the configuration of the production line test system and serve as the basis for the inspection of subsequent batches of SSDs. For Category A metrics (performance / benignity): we want them to be "high" (e.g., speed); we want to relax the standards to allow more products to qualify, but relaxation is not without limits; the bottom line is the performance of the "compliant" sample subset; we need to lower the passing grade (for metrics where higher is always better); taking "sequential read speed" (higher is always better) as an example: the old standard might be 800 MB / s, and we found that the speeds of the "compliant" SSDs in the critical group varied. If we want to save more products, we should lower the standard from 800 MB / s. How much should we lower it to? The system examines the speed data of the "compliant" sample subset, counting from low to high, and takes the value at position P1 (i.e., the P1 percentile). The smaller P1 is (e.g., 10), the more we take the slowest 10% level in the "compliant" sample as the new standard. This new standard is definitely lower than the old standard, and the smaller P1 is, the more the standard is lowered, and the more products can be saved. Therefore, for Category A indicators, in order to relax the standard, we must take the value at the "low tail" of the "compliant" sample data distribution, so P1 must be a small number (<50). For Category B indicators (defects / errors): we want them to be "low" (e.g., void area). We also want to relax the standards, but "relaxing" for defect indicators means allowing larger defect values ​​to exist. We need to increase the upper limit of the allowable defect. Taking "solder void area" (the lower the better) as an example: the old standard might be 1%. We found that the void areas of the SSDs that "complied" in the critical group varied in size. If we want to save more products (i.e. allow slightly defective products to leave the factory), we should increase the upper limit of the allowable defect from 1%. How much should we increase it to? The system examines the void area data of the "compliant" sample subset, counting from low to high, and takes the value at position P2 (i.e., the P2 percentile). The larger P2 is (e.g., 90), the more we are using the level of the largest 10% of void areas in the "compliant" samples (i.e., ranked after 90%) as the new standard. This new standard is definitely higher than the old standard (allowing for larger defects), and the larger P2 is, the higher the standard is raised, and the more products can be saved. Therefore, for Category B indicators, in order to relax the standard (increase the upper limit of defect tolerance), we must take the value at the "high end" of the "compliant" sample data distribution, so P2 must be a large number (>50). First, it establishes a scientific and sophisticated "parameter-level" cost optimization mechanism, fundamentally combating the economic waste of "overtesting"; ΔQ can be determined by multiplying the standard deviation (σ) of the Q_total scores of all samples in the sample library by a scaling factor k (e.g., k=0.5), i.e., ΔQ = k * Traditional methods cannot distinguish which test parameters can be relaxed and which must be strictly adhered to, resulting in a "one-size-fits-all" approach. This solution, however, utilizes the refined operation of the S520 to analyze each test parameter independently. Its core lies in employing drastically different statistical strategies for "performance / benign indicators" and "defect / error indicators": For the former (such as read / write speed and power consumption), a new standard is set by calculating the low percentile (P1) of the "compliant" samples. This essentially sacrifices some "icing on the cake" performance to save a large number of SSDs mistakenly rejected due to minor performance deficiencies. For the latter (such as solder joint voids and bad block count), a tolerance limit is set by calculating the high percentile (P2). This is equivalent to clearly defining a safe "defect tolerance zone" supported by empirical data, allowing products with minor flaws but verified as reliable to leave the factory. This categorized approach ensures that cost optimization is no longer a general "relaxation of standards," but rather a precise measurement and targeted reduction of the "redundancy" of each test, thereby maximizing economic benefits in the numerator. Second, it ensures that the optimization process is risk-controllable and decision-making is safe, and that standard adjustments are based on solid empirical evidence. All new threshold calculations (S523) are strictly based on data from a subset of SSDs in the "critical sample group" that have been verified to "meet" customer requirements. This means that the new standards are not the product of theoretical derivation or subjective guesswork, but rather a summary and generalization of the "actual level achieved by physical products that have successfully met customer needs." For example, setting the defect index threshold at the P2 percentile is equivalent to declaring: "Based on historical data, SSDs with defect levels within this limit have proven to have an acceptable risk of failure in actual use." This provides an irrefutable "safe data endorsement" for relaxing standards, enabling quality decision-makers to confidently implement optimization and fundamentally avoiding the risk of batch field failures that may be caused by blindly relaxing standards. Third, it achieves a seamless closed loop from data analysis to production execution, greatly improving the efficiency and agility of implementing optimization solutions. The S530 and S540 steps specifically address the "last mile" problem of engineering technology transformation. S530 accurately translates abstract statistical values ​​(such as "10th percentile") into engineering instructions that test engineers and equipment can directly understand and execute, such as specific voltage thresholds, minimum bandwidth, or maximum number of errors. S540 integrates these scattered parameters into a structured configuration file that can be directly imported into the production line test system. This design enables the entire complex data analysis results to take effect immediately in the next production batch without tedious manual interpretation and secondary development. As a result, enterprises gain a "dynamic test standard" deployment capability that can respond quickly and iterate continuously, and optimization results can be rapidly transformed into tangible yield improvements and capacity release. Fourth, it creates a transparent, traceable, and auditable standardized process, enhancing the maturity of the quality management system. The entire optimization process (from S510 to S540) is highly structured and data-driven. The source of each new threshold (which critical sample group it comes from), the calculation basis (which type of percentile is used), and the final form (what engineering parameters it is converted to) are all clearly traceable. This completely changes the traditional "black box" standard adjustment model that relies on expert experience, ensuring that any change to the testing standard has a complete "data lineage" and logical chain. This not only greatly enhances the credibility of decision-making but also provides valuable data assets and standardized working paradigms for quality auditing, process traceability, and continuous process optimization.

[0025] In this embodiment of the invention, step S600 specifically includes: S610 Deploy and verify the new test standard: Import the optimized test standard specification file generated by S540 into the production line test system and replace the original test parameter thresholds; before the new standard is officially implemented, extract a small batch of trial production SSDs and use the old and new standards to conduct parallel testing and compare the results to confirm the effectiveness and safety of the new standard in engineering applications. S620, Monitoring and collecting production data under the new standard: After the optimized testing standard is officially applied to subsequent production batches, continue to execute the complete testing and scoring process from S100 to S200, and collect detailed sub-item scores, overall quality score Q_total and production pass rate data for all SSDs in this stage. S630. Evaluation of Optimization Effect and System Convergence Judgment: Based on the production data collected in S620, and combined with the continuously running S300 to S500 processes, the optimization effect is evaluated, and the judgment includes: S631. Compare whether the overall production pass rate before and after the implementation of the new standard has improved as expected; S632. Analyze whether the compliance rate of the new batch of samples on the S300 benchmark test platform is stably maintained at the target level; S633. Observe whether the customer standard critical comprehensive score Q_critical determined in the latest round of S400 tends to be stable, and whether its fluctuation is less than the preset stability threshold. That is, observe whether the relative change between the Q_critical determined in the latest round of S400 and the average value of Q_critical determined in the previous N consecutive rounds (e.g., N=3) is less than the preset stability threshold (e.g., 5%). S640, Trigger standard iteration update or maintain the current standard: Make a decision based on the evaluation results of S630: If S633 indicates that the system has converged and the results of S631 and S632 meet the requirements, then maintain the current optimized test standard; otherwise, incorporate the new batch production data collected in S620 and its verification results (compliance labels) under step S300 into the customer standard compliance sample library, and then automatically trigger a new round of dynamic threshold approximation and threshold reverse derivation process starting from S400 and S500 to generate and deploy the next-generation optimized test standard; First, it achieves a fundamental shift in the optimization process from a "one-off project" to "routine operation," establishing a permanent evolutionary engine for quality control. Traditional methods, even after completing a standard adjustment, cannot cope with subsequent changes in materials, processes, or customer requirements. This solution, through continuous data monitoring of S620 and automatic decision triggering of S640, makes optimization a permanent process running parallel to production. This means that enterprises gain an "always-on" self-optimization system: whenever the characteristics of a new batch of NAND flash memory drift, or when a customer server platform is upgraded, the system can detect the changes through continuous verification of S300 and automatically initiate a new round of optimization (S400-S500) to adjust the testing standards. This gives the enterprise's quality defenses adaptive capabilities, enabling them to continuously maintain the optimal balance between cost and reliability, rather than gradually failing. Secondly, its rigorous "deployment-verification-monitoring" process (S610-S630) provides multiple safeguards for optimization decisions, eliminating batch risks caused by standard changes. The S610 step requires parallel testing of the old and new standards, which is equivalent to a "canary release" in engineering terms. This allows for verification of the safety of the new standard on a small scale, preventing defective parameter settings from directly impacting full-scale production. The S630 evaluation system constructs a multi-dimensional health dashboard: the production pass rate (S631) directly measures cost-benefit; the customer verification compliance rate (S632) is the ultimate quality rule; and the stability of Q_critical (S633) reflects the maturity of system cognition. Only when all three meet the standards will the system determine that the optimization is successful and maintain the status quo. This design ensures that any optimization must pass the dual tests of economic benefits (pass rate improvement) and risk control (compliance rate stability) simultaneously, making the decision extremely robust. Third, by quantifying "system convergence" (S633), it achieves intelligent allocation and efficient utilization of optimized resources (time, computing power, verification cost). Continuous optimization is not an infinite loop. When the Q_critical value fluctuates less than the threshold in multiple iterations, it indicates that the system's understanding of the customer's true quality boundary has become stable and the "optimal solution" under the current conditions has been found. At this time, S640 decides to pause large-scale iterations and switch to monitoring mode. This avoids unnecessary recalculation and verification on the optimized standard and saves valuable test resources and computing power. The system will only be reawakened when external conditions change and cause data drift (such as Q_critical starting to fluctuate continuously again). This allows the entire system to intelligently switch between the two states of "active learning" and "stable operation" and maximize resource efficiency. Fourth, this closed-loop mechanism has solidified the core capability of "data-driven decision-making" at the organizational level, transforming the quality management paradigm from relying on personal experience to inheritable systemic wisdom. The entire S600 process standardizes and automates the core knowledge work of "how to evaluate and update test standards." All decision-making bases (monitoring data, evaluation indicators, convergence judgments) are clearly defined and recorded. This makes the iteration of quality strategies no longer dependent on the experience and intuition of a senior engineer, but transforms into a routine operation based on clear rules and data that is verifiable and inheritable. Even with personnel changes, this system can ensure the continuity and consistency of the company's quality optimization capabilities, becoming a core digital asset of the organization.

[0026] In this embodiment of the invention, the dynamic closed-loop update process (S600) consists of a core optimization loop and a supporting management process, the supporting management process specifically including: S710 (Metadata Anchoring and Version Release): Each time S540 is executed to generate a new optimized test standard specification file, a version snapshot is created and stored synchronously. This snapshot uniquely associates with and records: the identifier of this version standard, the data timestamp of the customer standard compliance sample library on which it is based, the version of the weight matrix used when generating the standard, the critical score Q_critical and its confidence interval, and the percentile parameter (P1, P2) combination used in S520. S720 (Security Deployment and Impact Preliminary Review): Before performing parallel testing of the old and new S610 standards, the system automatically compares the changes in the thresholds of all test sub-items in the old and new versions, identifies the top K items with the "largest relaxation" and the "only tightened" item, forms a "List of Key Change Items", and prioritizes the verification resources of S610 to conduct in-depth testing and risk confirmation of the items on this list. S730 (Runtime Monitoring and Abnormal Circuit Breaker): During the S620 monitoring process, the system monitors two major indicators in real time: 1) The first pass rate of any test sub-item decreases significantly relative to the historical average; 2) A cluster of failure modes that were not previously recorded appear in the S300 benchmark test; Once a monitoring alarm is triggered, the system can automatically suspend the widespread application of the new standard and trigger root cause analysis, while providing an operation interface for rolling back to the previous stable version of the standard. S740 (Iterative Decision and Audit Tracking): The S630 assessment report needs to integrate core indicators (pass rate, compliance rate, Q_critical stability), the S720 impact pre-audit summary, and the S730 runtime monitoring records to form a comprehensive decision report; the system automatically provides decision suggestions based on preset strategies (such as "maintain if the comprehensive report is all green, trigger a review if any red / yellow color is present"), and the entire "report-approval-execution" process generates an immutable audit log, which is stored in association with the S710 version snapshot; In each iteration of the core optimization loop (S600), a complete version snapshot containing all parameters and data sources is first created for the newly generated test standard through S710 (metadata anchoring and version release), realizing digital anchoring and traceability of each change; Next, before deployment, the S720 (Security Deployment and Impact Assessment) system automatically compares the differences between the old and new standards, identifies and prioritizes the verification of the most critical changes with the highest risk, forming a preventative security checkpoint. After the new standard goes live, the S730 (Runtime Monitoring and Circuit Breaker) system monitors production pass rate and failure modes in real time. Once a statistically significant anomaly or an unknown cluster of failures is detected, an automatic warning is issued and a circuit breaker rollback mechanism can be triggered to prevent the risk from spreading. Ultimately, by integrating all technical metrics, risk assessment results, and monitoring alerts through S740 (Iterative Decision and Audit Trail), a structured report supporting "maintain, observe, or review" decisions is generated. The entire process of decisions and operations is linked to version snapshots via immutable logs, completing a closed loop from execution to auditing. Its beneficial effects are concrete and crucial: First, it endows high-frequency data-driven optimization with industrial-grade reliability and security. Through the impact pre-screening of S720 and the real-time circuit breaking of S730, the system can proactively prevent batch misjudgments or quality accidents caused by algorithm "blind action" or sudden changes in the production environment, and control the potential risks in the optimization process to the minimum. Secondly, it achieves full digitalization and transparency of the quality management system. The version snapshot of S710 and the audit log of S740 ensure that the “ins and outs” of every standard change—including why it was changed, what data it was based on, and who approved it—is fully recorded and readily available, greatly satisfying the stringent requirements for quality system auditing and process traceability. Third, it will upgrade the optimization system from a technical tool to an autonomous, intelligent decision-making management entity. Through the S740's automatic decision-making suggestions based on multi-source information (performance data, risk summary, monitoring status), the system can more robustly determine when optimization has converged and when intervention and adjustment are needed. Thus, while continuously pursuing cost-effectiveness, it firmly safeguards the bottom line of quality risk and operational compliance.

[0027] An apparatus includes: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described solid-state drive testing methods for big data storage.

[0028] A storage medium comprising a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described solid-state drive testing methods for big data storage.

[0029] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A testing method for solid-state drives used for big data storage, characterized in that, Includes the following steps: S100 performs multi-dimensional production tests on solid-state drives, covering components, processes, functions, and reliability, and provides digital scoring for the results of each test item. S200. Based on the scores and preset weights of each test sub-item, calculate the overall quality score Q_total for each solid-state drive. S300. Extract production samples and verify them on a benchmark testing platform that simulates actual customer applications. Based on the verification results, establish a sample library labeled with the customer's standard compliance status and its corresponding Q_total and sub-item data. S400. Based on the Q_total distribution of "compliant" and "non-compliant" samples in the sample library, an iterative approximation algorithm is used to determine a customer standard critical comprehensive score Q_critical with a prediction misjudgment rate within an acceptable risk tolerance. S500: Analyze the statistical characteristics of the critical sample group located in the neighborhood of Q_critical on each test item, reverse deduce the new and optimized test pass threshold for each production test item, and integrate them to form a new production test standard specification. S600. Apply the new production testing standard to subsequent production testing, and continuously perform dynamic closed-loop updates to the testing standard through steps S300 to S500. The dynamic closed-loop update process includes a core optimization cycle and a supporting version management process.

2. The solid-state drive testing method for big data storage according to claim 1, characterized in that, Step S100 specifically includes: S110, Component-level testing and scoring: Perform raw bad block count and raw bit error rate testing on the NAND flash memory chips of the SSD, perform basic function and power consumption testing on the main control chip, perform impedance and connectivity testing on the PCB substrate, and perform capacitance and frequency accuracy testing on the auxiliary capacitors and crystal oscillators. Record all test data and score them according to the deviation of each test result from the standard value using a preset first scoring model; The first scoring model converts the deviation of each test parameter into a sub-item score using a piecewise linear function based on the degree to which its measured value deviates from the standard value. The weighted sum of all relevant sub-item scores constitutes the final score of this component. S120. Process and structural testing and scoring: Perform AOI optical inspection on the SSD board after mounting and soldering to identify component misalignment, cold solder joints, missing components and record the coordinates and types of defects. Perform X-ray inspection to check the quality of internal solder joints, chip stack alignment, and the presence of voids; Perform online ICT circuit testing to verify the connectivity of all electrical nodes and the static parameters of key components; Perform a preliminary power-on test to confirm that the board can be recognized and enter the initialization state; Based on the number of defects detected, their severity level, and deviations in electrical parameters, a pre-set second scoring model is used for scoring; The second scoring model is a deduction system model, which sets a basic deduction value for each type of defect and multiplies it by the corresponding deduction coefficient according to the severity level of the defect. Test items with no defects or fully qualified parameters are given a benchmark full score. S130 Function and Firmware Testing and Scoring: Firmware flashing and verification are performed on SSD boards that have passed the S120 test to ensure firmware integrity. Perform functional logic testing to verify the correct response of all user command interfaces, security management functions, and basic read / write commands; Compatibility testing and initial performance screening were conducted on a standard testing platform, and the recognition success rate and initial sequential read / write bandwidth values ​​were recorded under different host interfaces. Based on the achievement of the standards, a pre-set third scoring model is used for scoring; The third scoring model uses Boolean scoring for binary pass / fail tests and converts the score for continuous variable tests by the percentage of the measured value relative to the target value. S140. Reliability Enhancement Test and Rating: Perform full-capacity sequential and random mode read and write tests on the assembled SSD, and record any errors or latency anomalies that occur during the process; Perform bad block retesting and management to verify the reliability of the bad block replacement mechanism; High-temperature aging tests were conducted in a high-temperature chamber to monitor steady-state operation. High and low temperature cyclic stress tests were conducted to examine the stability under sudden temperature changes. Simulate abnormal power outage scenarios and conduct multiple power outage protection tests to verify data consistency; Accelerated P / E cycle tests were performed under specific loads, and the performance degradation trend was recorded. Perform EMC electromagnetic compatibility tests and record the radiated and conducted interference values; Based on the error count, performance retention rate, number of passes, and interference values ​​exceeding the standard for each test item, a pre-set fourth scoring model is used for quantitative scoring. The fourth scoring model is based on the principle of statistical process control. It sets an acceptable "lower limit" for each reliability index. The score is based on the difference or ratio between the measured value and the lower limit, and is mapped through an exponential decay function to reflect the comprehensive assessment of early failures and long-term stability.

3. The solid-state drive testing method for big data storage according to claim 2, characterized in that, Step S200 specifically includes: S210. Constructing the initial weight matrix: Based on product design specifications, historical fault analysis data, and expert experience, initial weights are assigned to the scores of each test sub-item in S110 to S140, forming a multi-dimensional initial weight matrix; among them, the test sub-items that have a decisive impact on the long-term reliability and data integrity of the SSD have higher weights than the test sub-items that have an impact on instantaneous performance or appearance and process. S220. Data normalization preprocessing: Before weighted calculation, the scores output in S110 to S140 are normalized to map all scores into a uniform numerical range, so as to eliminate the deviation caused by different scales or ranges of the scoring model to the comprehensive calculation results. S230. Calculate the weighted overall quality score: For each SSD that has completed all tests, the normalized sub-item scores are weighted and summed with the corresponding weight values ​​in the initial weight matrix to obtain the initial overall quality score of the SSD. S240. Introduce a batch correction factor: Calculate the average or median score of all SSDs in the current production batch on specific key test items, and compare it with the data of the historical baseline batch; based on the comparison deviation, generate a batch correction factor to fine-tune the initial comprehensive quality score to eliminate systematic deviations caused by material batch fluctuations or environmental drift, and finally obtain the comprehensive quality score for subsequent analysis.

4. The solid-state drive testing method for big data storage according to claim 3, characterized in that, Step S300 specifically includes: S310. Formulate sampling strategy and sample preparation: From each production batch for which the comprehensive quality score Q_total is calculated by S200, extract physical samples SSD according to a preset sampling ratio and rules; the sampling rules ensure that the samples cover multiple score ranges from high Q_total to low Q_total within the batch, and give preference to SSDs whose Q_total is close to the customer standard critical comprehensive score Q_critical used in current production. S320, Building and Deploying a Benchmark Test Platform: Build a benchmark test platform independent of the production line. This platform consists of multiple test servers, racks, a temperature control system, and load generation and management software. Its hardware configuration, interface type, heat dissipation environment, and firmware driver version all simulate the actual application environment of the target big data storage customer. S330. Define a customer standard compliance test procedure: Perform long-term reliability operation tests on the benchmark test platform. The test procedure includes: S331, Load Model Test: Run a load script that simulates typical I / O modes of big data storage, including high-proportion sequential writes, garbage collection pressure, mixed read and write, and quiet periods, and record any errors, delays, or data consistency check failures that occur during the process. S332. Environmental stress test: Under the highest operating ambient temperature specified in the customer's equipment specifications, continuously run the load model and monitor the steady-state performance and failure rate of the SSD. S333, Lifetime and Data Retention Test: The cumulative write volume reaches the TBW value specified in the customer's warranty terms by converting the write amplification factor, and data integrity is verified after completion; S340. Perform tests and determine compliance: Perform the complete test procedure defined in S330 on each sampled SSD; if no preset critical failure conditions are triggered during the test period and all data verifications pass, the sample is determined to "comply" with the customer's usage standards; otherwise, record the failure mode and time point, and determine it to "not comply"; S350. Building and Updating the Sample Database: Create a record for each tested sample SSD, which includes at least: Sample ID, identifier of the source production batch, original comprehensive quality score Q_total calculated in S200, detailed scores or raw data of each key test sub-item in S100, failure mode and time data on the benchmark test platform, and the final "compliant" or "non-compliant" judgment label; all records constitute a continuously updated customer standard compliance sample library; S360. Dynamic Optimization of the Weight Matrix: Using the customer standard compliance sample library and its corresponding detailed sub-item scores established in S350 as training data, correlation analysis is used to analyze the correlation strength between each test sub-item score and the final customer standard compliance. Based on the analysis results, the weight values ​​in the initial weight matrix in S210 are periodically iteratively optimized to generate an optimized weight matrix. This optimized weight matrix will replace the initial weight matrix in S210 and be applied to the calculation of the overall quality score Q_total in the S200 step of subsequent production batches of SSDs, so as to continuously enhance the predictive ability of the overall quality score Q_total for customer standard compliance.

5. The solid-state drive testing method for big data storage according to claim 4, characterized in that, Step S400 specifically includes: S410, Data Extraction and Interval Initialization: Extract the comprehensive quality score Q_total and its corresponding "compliant" or "non-compliant" label for all samples from the customer standard compliance sample library constructed in S350; calculate the minimum Q_total value Q_min_pass for "compliant" samples and the maximum Q_total value Q_max_fail for "non-compliant" samples, and use the interval [Q_max_fail, Q_min_pass] as the initial search interval; S420. Set approximation parameters and target: Preset an acceptable maximum prediction misclassification rate R_max as the convergence target. The misclassification rate includes the first type error rate of misclassifying actual "non-compliant" samples as qualified, and the second type error rate of misclassifying actual "compliant" samples as unqualified. S430. Execute the iterative approximation algorithm: Use an iterative algorithm to find the critical score threshold within the search interval. Each iteration includes: S431. Candidate threshold generation: Calculate a new candidate threshold Q_candidate based on the current search interval, preferably using the binary search method to find the midpoint of the interval; S432. False positive rate simulation calculation: The candidate threshold Q_candidate is used as the virtual factory pass line. That is, in the simulation, if Q_total ≥ Q_candidate, it is predicted as "pass", and otherwise it is predicted as "fail". Traverse the sample library, count the number of samples whose prediction results do not match the real labels, calculate the first type error rate and the second type error rate respectively, and use the weighted sum and / or the maximum value of the two as the overall verification false positive rate of the current candidate threshold. S433. Search Interval Update and Convergence Judgment: Update the search interval based on simulation results: If the current overall verification misclassification rate is higher than R_max and is mainly caused by Type I errors, then increase the lower limit of the search interval; if it is mainly caused by Type II errors, then decrease the upper limit of the search interval; if the overall verification misclassification rate is lower than R_max, then record the current candidate threshold and its misclassification rate; the iteration process continues until the search interval width is less than the preset accuracy threshold ε. S440. Determine the final critical score: After the iteration converges, select the threshold with the lowest Q_total value from the candidate thresholds that meet the condition that the misclassification rate is lower than R_max, and formally determine it as the customer standard critical comprehensive score Q_critical. S450, Confidence Assessment and Data Augmentation Instruction: Based on the final determined Q_critical, assess its confidence level in the current sample database. If the confidence level is lower than the preset standard, generate an instruction to be fed back to step S310, requiring supplementary sampling of SSDs with Q_total values ​​in the vicinity of Q_critical in subsequent batches to enrich the data density of this critical area for use in the next round of optimization cycle.

6. The solid-state drive testing method for big data storage according to claim 5, characterized in that, Step S500 specifically includes: S510. Define and screen the critical sample group: Taking the customer standard critical comprehensive score Q_critical determined in S440 as the center, set a neighborhood range ΔQ that fluctuates up and down. From the customer standard compliance sample library constructed in S350, screen all SSD samples whose comprehensive quality score Q_total falls within the interval [Q_critical - ΔQ, Q_critical + ΔQ] to form the critical sample group. S520. Analyze the sub-item data distribution of the critical sample group and determine the new threshold: For each independent production test sub-item defined in S100, an optimized test pass threshold is calculated based on the actual performance of the critical sample group. The specific operation is as follows: S521. Extracting and separating data: From the critical sample group defined in S510, extract the original measurement data of all samples on the specific test item; according to the "compliant" / "non-compliant" labels recorded in the sample library, the dataset is clearly separated into two subsets: the "compliant" sample data subset and the "non-compliant" sample data subset, and all data values ​​in the "compliant" sample data subset are sorted in ascending order according to their numerical values; S522. Define indicator types and select statistical strategies: Based on the physical nature represented by this test item, determine which of the following types it belongs to: Category A: Performance / Benefit Metrics: Metrics that measure a product's "capability" or "health," with the expectation that "the higher the better." Their failure usually leads to performance degradation rather than immediate loss of functionality. Category B: Defect / Error Indicators: Indicators that measure manufacturing "flaws" or "potential failure points". The desired direction is always "the lower the better". Their deterioration is directly related to sudden functional failure. Choose a statistical strategy: For Category A indicators, analyze the subset of sample data that "meets" the criteria and calculate its P1 percentile, where P1 is the number less than 50. For Category B indicators, analyze the subset of sample data that "meets" the criteria and calculate its P2 percentile, where P2 is the number greater than 50. S523. Calculate and determine the recommended threshold: Perform the statistical calculations selected in S522; For Category A indicators, the calculated P1 percentile value will be directly used as the new suggested test pass threshold for that test sub-item. For Category B indicators, the calculated P2 percentile value will be used as the new recommended test pass threshold for that sub-item. S530. Convert the suggested test pass threshold into executable test parameters: For each test sub-item, based on its physical meaning and the requirements of the test equipment, convert the suggested test pass threshold obtained in S523 into specific parameters that can be directly called by the production line test program; the conversion includes: S531. For electrical parameter testing, convert the threshold to upper or lower limits for voltage, current, or timing; S532. For functional and performance testing, convert the threshold to minimum throughput, maximum permissible delay, or minimum bandwidth. S533. For reliability testing, the threshold is converted to the maximum permissible error count, the minimum performance retention rate, or the maximum permissible interference level. S540: Generate and output optimized test standard document: Integrate the specific parameters of all test sub-items after conversion by S530, and compile them into a structured optimized test standard specification document according to the order and format requirements of the production line test process; this document is used to directly update the configuration of the production line test system and serve as the basis for the inspection of subsequent batches of SSDs.

7. The solid-state drive testing method for big data storage according to claim 6, characterized in that, Step S600 specifically includes: S610 Deploy and verify the new test standard: Import the optimized test standard specification file generated by S540 into the production line test system and replace the original test parameter thresholds; before the new standard is officially implemented, extract a small batch of trial production SSDs and use the old and new standards to conduct parallel testing and compare the results to confirm the effectiveness and safety of the new standard in engineering applications. S620, Monitoring and collecting production data under the new standard: After the optimized testing standard is officially applied to subsequent production batches, continue to execute the complete testing and scoring process from S100 to S200, and collect detailed sub-item scores, overall quality score Q_total and production pass rate data for all SSDs in this stage. S630. Evaluation of Optimization Effect and System Convergence Judgment: Based on the production data collected in S620, and combined with the continuously running S300 to S500 processes, the optimization effect is evaluated, and the judgment includes: S631. Compare whether the overall production pass rate before and after the implementation of the new standard has improved as expected; S632. Analyze whether the compliance rate of the new batch of samples on the S300 benchmark test platform is stably maintained at the target level; S633. Observe whether the customer standard critical comprehensive score Q_critical determined in the latest round of S400 tends to be stable and whether its fluctuation is less than the preset stability threshold. S640, Trigger standard iteration update or maintain the current standard: Make a decision based on the evaluation results of S630: If S633 indicates that the system has converged and the results of S631 and S632 meet the requirements, then maintain the current optimized test standard; otherwise, incorporate the new batch production data collected in S620 and its verification results under step S300 into the customer standard compliance sample library, and then automatically trigger a new round of dynamic threshold approximation starting from S400 and threshold reverse derivation process from S500 to generate and deploy the next-generation optimized test standard.

8. The solid-state drive testing method for big data storage according to claim 7, characterized in that, The dynamic closed-loop update process consists of a core optimization loop and a supporting management process, which specifically includes: S710. Each time S540 is executed to generate a new optimized test standard specification file, a version snapshot is created and stored synchronously. This snapshot is uniquely associated with and records: the identifier of this version standard, the data timestamp of the customer standard compliance sample library on which it is based, the version of the weight matrix used when generating the standard, the critical score Q_critical and its confidence interval, and the percentile parameter (P1, P2) combination used in S520. S720 Before performing parallel testing of the old and new S610 standards, the system automatically compares the changes in the thresholds of all test sub-items in the old and new versions, identifies the top K items with the "largest relaxation" and the "only tightened" items, forms a "List of Key Change Items", and prioritizes the verification resources of S610 to conduct in-depth testing and risk confirmation of the items on this list. During the S730 and S620 monitoring process, the system monitors two major indicators in real time: 1) The first pass rate of any test sub-item has decreased significantly relative to the historical average; 2) A cluster of failure modes that were not previously recorded have appeared in the S300 benchmark test. Once a monitoring alarm is triggered, the system can automatically suspend the widespread application of the new standard and trigger root cause analysis, while providing an operation interface for rolling back to the previous stable version of the standard. The evaluation reports for S740 and S630 need to integrate core indicators, the preliminary review summary of the impact of S720, and the runtime monitoring records of S730 to form a comprehensive decision report. The system automatically provides decision suggestions based on preset strategies, and the entire "report-approval-execution" process generates tamper-proof audit logs, which are stored in association with the version snapshot of S710.

9. An apparatus, characterized in that, include: The memory is used to store computer programs; A processor, which executes a computer program to implement the steps of the solid-state drive testing method for big data storage as claimed in any one of claims 1-8.

10. A storage medium, characterized in that, The storage medium includes a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the solid-state drive testing method for big data storage as claimed in any one of claims 1-8.

Images