Method for comprehensively verifying data integrity of solid state disk

Abstract

The invention relates to the technical field of data storage, and discloses a method for comprehensively verifying data integrity of a solid state disk, which comprises the following steps of: preprocessing the solid state disk in a way of at least sending a full-disk erasing instruction to the solid state disk, recording initial SMART information and writing the initial SMART information into a verification file; dividing the solid state disks into multiple groups according to a voltage strategy; performing comprehensive stress test circulation on the grouped multiple groups of solid state disks, wherein the comprehensive stress test circulation at least comprises temperature circulation, load mode switching, transient voltage testing and data integrity verification; data acquisition of SMART data of the solid state disk is carried out, and intelligent analysis of the SMART data is carried out; and judging whether the test is finished or not, and if the test times do not reach the preset times, returning to the previous step to continue the next round of test. According to the method, the data integrity level of the solid state disk is effectively improved, and reliable support is provided for cooperative work of multiple devices in an actual application scene.

Description

Technical Field

[0001] This invention relates to the field of data storage technology, and in particular to a comprehensive method for verifying the data integrity of solid-state drives. Background Technology

[0002] With the widespread adoption of solid-state drives (SSDs) in various enterprise-level applications such as data centers, high-performance computing, high-end storage, and edge computing, SSD usage is surging. However, as the number of SSDs actually running online gradually increases, the probability of unexpected power outages related to SSD maintenance activities such as unforeseen power failures, human operation, machine malfunctions, and data center power outages also increases. Furthermore, during long-term operation, the performance of SSDs may gradually decline, or even malfunction, affecting the stability and security of data storage.

[0003] To improve the reliability of solid-state drives (SSDs), existing technologies typically employ power-loss protection to prevent data loss in the SSD cache. This feature usually includes a voltage detection module and a power storage module to detect interruptions and provide backup power when the external voltage is too low. However, this approach still has some limitations, primarily in its lack of ability to effectively predict and proactively address SSD failures. Summary of the Invention

[0004] This invention provides a comprehensive verification method for data integrity of solid-state drives (SSDs), which can solve the current technical problem of lacking the ability to effectively predict and proactively address SSD failures.

[0005] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is: to provide a comprehensive verification method for data integrity of a solid-state drive, the method comprising: The preprocessing of the solid-state drive includes at least: issuing a full erase command to the solid-state drive, recording initial SMART information and writing it to a verification file; Solid-state drives are divided into multiple groups based on voltage strategy; A comprehensive stress test cycle is performed on the grouped solid-state drives. The comprehensive stress test cycle includes at least temperature cycling, load mode switching, transient voltage testing, and data integrity verification. It collects SMART data from solid-state drives and performs intelligent analysis on the SMART data; Determine if the test is complete. If the preset number of tests has not been reached, return to the previous step and continue to the next round of testing; if the preset number of tests has been reached, proceed to the next step. Test reports are generated based on SMART data and intelligent analysis results.

[0006] The beneficial effects of this invention are as follows: By introducing innovative modules such as dynamic load simulation, rapid transient voltage testing, data persistence verification, and multi-dimensional fault prediction, a multi-stress coupled testing environment involving temperature, voltage, load, timing, and transient interference is constructed. This more realistically simulates the actual working state of solid-state drives (SSDs) in harsh environments such as data centers, industrial control systems, and automotive electronics, overcoming the limitations of existing technologies that only focus on single-indicator testing. Through an active fault injection and prediction mechanism, combined with AI analysis of SMART data, effective prediction and early response to SSD faults are achieved, breaking through the passive detection mode of existing technologies that can only diagnose after a fault occurs. Through a deep data integrity measurement mechanism, introducing data hash verification and data verification tools, "silent data errors" that Diskinfo cannot detect can be discovered, establishing a black-box evaluation system for data reliability and effectively improving the data integrity level of SSDs. Through multi-disk parallel testing and mutual interference mechanisms, the system-level compatibility issues of multiple SSDs running under high load simultaneously are examined, providing reliability support for multi-device collaborative work in practical application scenarios. By adopting the concept of full lifecycle management, and through the full monitoring and data collection and analysis of solid-state drives (SSDs), a comprehensive assessment of SSD performance and reliability is achieved, providing users with a more comprehensive and accurate SSD reliability assessment solution. Attached Figure Description

[0007] Figure 1 This is a flowchart illustrating the comprehensive verification method for data integrity of a solid-state drive according to the first embodiment of the present invention.

[0008] Figure 2 yes Figure 1 A flowchart illustrating step 2.

[0009] Figure 3 yes Figure 1 A flowchart illustrating step 3.

[0010] Figure 4 yes Figure 1 A flowchart illustrating step 4.

[0011] Figure 5 yes Figure 1 A flowchart illustrating step 6. Detailed Implementation

[0012] 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 a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0013] The terms "comprising" and "having," and any variations thereof, used in this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] Figure 1 This is a flowchart illustrating the comprehensive data integrity verification method for solid-state drives according to the first embodiment of the present invention. Figure 1 As shown, the system includes hardware and software components: Step 1: Preprocess the solid-state drive, wherein the preprocessing method includes at least: issuing a full erase command to the solid-state drive, recording the initial SMART information and writing it to a verification file; Step 2: Divide the solid-state drives into multiple groups according to voltage strategy; Step 3: Perform a comprehensive stress test cycle on the grouped solid-state drives. The comprehensive stress test cycle includes at least temperature cycling, load mode switching, transient voltage testing, and data integrity verification. Step 4: Collect SMART data from the solid-state drive and perform intelligent analysis on the SMART data; Step 5: Determine if the test is complete. If the preset number of tests has not been reached, return to the previous step to continue the next round of testing; if the preset number of tests has been reached, proceed to the next step. Step 6: Generate a test report based on SMART data and intelligent analysis results.

[0016] Step 1: SSD Preprocessing (Ensuring Initial State Consistency) Preprocessing requires three core operations: First, execute a full disk erase command using the ATA protocol's "SECURITY ERASE UNIT" or the NVMe protocol's "Format NVM" (selecting "Full Disk Erase" mode) to remove any remaining data and eliminate interference from past usage on test results. Second, record initial SMART information, focusing on key parameters such as remaining lifetime percentage, bad block count, average write cycles, temperature, and error correction count. Use professional tools like CrystalDiskInfo or NVMe CLI to export the raw data and create an "Initial State Archive" for each drive. Third, write a verification file, generating a random binary file (recommended size is 80% of the drive capacity to avoid the impact of an empty drive on testing). Simultaneously, record the file's MD5 or SHA256 checksum to provide a benchmark for subsequent data integrity verification and detect whether data loss or corruption occurred during the test.

[0017] Step 2: In the SSD voltage grouping and special testing settings (simulating differentiated power supply and interference scenarios), the grouping is based on (referencing industry standard voltage ranges): SSDs are divided into three groups according to 12V and 5V power supply types: Group A corresponds to normal power supply scenarios, with a 12V power supply range of 11.4V-12.6V and a 5V power supply range of 4.75V-5.25V; Group B corresponds to unstable power supply scenarios, with a 12V power supply range of 10.8V-11.3V and a 5V power supply range of 4.5V-4.74V; Group C corresponds to power fluctuation overload scenarios, with a 12V power supply range of 12.7V-13.2V and a 5V power supply range of 5.26V-5.5V.

[0018] Operational requirements: Each sample group must contain at least 3 disks to ensure statistical significance; hard drives from the same batch must be evenly distributed across groups to avoid batch variations affecting test results. Power supply control must use a programmable DC power supply (such as Keysight N6705B) to precisely control the power supply voltage for each group, with an error within ±0.05V.

[0019] Intra-group "mutual interference" detection mechanism (corresponding to step 201): simulate electromagnetic interference and power load fluctuation interference when multiple hard drives are working in parallel to verify the stability of hard drives in cluster environments such as server hard drive arrays.

[0020] Specific implementation plan: Hardware layout: Solid-state drives in the same group are installed in a "closely parallel" manner, with the spacing controlled within 2cm (simulating the actual server hard drive bay density), and all hard drives share the same power supply branch (drawn from the same circuit from the programmable DC power supply) to avoid independent power supply and eliminate interference factors; each group needs to be equipped with one electromagnetic interference (EMI) monitor (such as R&S ESR26), with the monitor probe 1cm away from the hard drive casing, to record the interference signal strength in real time in the frequency range of 10kHz-1GHz.

[0021] Interference Detection Scenarios and Indicators: Parallel Read / Write Interference: The trigger condition is that all hard drives in the group simultaneously perform 100% load sequential read / write using the FIO tool. The monitoring indicators are the single-disk throughput fluctuation range and the maximum IO latency. The judgment standard is that the throughput fluctuation exceeds 10%, or the IO latency exceeds 50ms and lasts for 10 seconds, which is judged as "interference exceeding the standard". Power Supply Load Fluctuation Interference: The trigger condition is that 1-2 hard drives in the group suddenly start and stop reading / writing (triggered once every 5 minutes). The monitoring indicators are the power supply branch voltage fluctuation value and the number of hard drive power failures / restarts. The judgment standard is that the voltage fluctuation exceeds ±0.1V, or there is one or more power failures, which is judged as "interference exceeding the standard". Long-Term Parallel Operation Interference: The trigger condition is that all hard drives in the group continuously run under cyclic load for 72 hours. The monitoring indicators are the SMART error count increment and the number of data verification failures. The judgment standard is that the error count increment exceeds 3, or the number of verification failures is not less than 1, which is judged as "interference exceeding the standard".

[0022] Data recording requirements: Generate an "Interference Detection Report" once per hour. The report must include the EMI signal strength curve, throughput fluctuation data, and error records. If "interference exceeds the standard" occurs, use an oscilloscope (such as Tektronix MDO3024) to capture the power supply voltage waveform at the moment of exceeding the standard and retain the key data for subsequent analysis.

[0023] Voltage test point setting (corresponding to step 202): Test point classification and parameter configuration. Rated voltage test point: The standard voltage of the 12V power supply group is 12V (error ±0.05V), and the standard voltage of the 5V power supply group is 5V (error ±0.05V). The test time of this test point accounts for 40%, mainly simulating the normal and stable power supply environment such as desktop computer standby.

[0024] Voltage bias test points: The 12V power supply group includes positive bias (12.6V, corresponding to +5%; 13.2V, corresponding to +10%) and negative bias (11.4V, corresponding to -5%; 10.8V, corresponding to -10%), and the 5V power supply group includes positive bias (5.25V, corresponding to +5%; 5.5V, corresponding to +10%) and negative bias (4.75V, corresponding to -5%; 4.5V, corresponding to -10%). The test time for this test point accounts for 30% of the test duration, simulating scenarios such as power adapter aging and mains voltage fluctuations.

[0025] Dynamic voltage fluctuation test point: The fluctuation frequency of both 12V and 5V power supply groups is 1kHz-10kHz. The fluctuation amplitude of the 12V power supply group is ±0.5V (relative to the rated voltage), and the fluctuation amplitude of the 5V power supply group is ±0.2V (relative to the rated voltage). The waveform is tested by alternating between sine wave and square wave. The test time of this test point accounts for 30% of the test time, simulating scenarios such as switching power supply ripple interference and simultaneous start-up and shutdown of multiple devices.

[0026] Test point execution logic: A single voltage test cycle is 2 hours, cycling in the following order: "Rated voltage (48 minutes) → Positive bias (36 minutes) → Negative bias (36 minutes) → Dynamic fluctuation (36 minutes) → Rated voltage (6 minutes, recovery transition)" to ensure uniform coverage of each test point. The slope must be controlled during voltage switching, with the voltage change rate not exceeding 0.5V / second (e.g., switching from 12V to 13.2V should take at least 2.4 seconds) to avoid sudden voltage changes that could damage the hard drive hardware. The hard drive's operating status must be recorded synchronously at each voltage test point. The "Power OnHours" and "Current Power Consumption" parameters are read in real-time via the NVMe CLI to verify the correlation between hard drive power consumption and operational stability under different voltages.

[0027] Step 3: In the comprehensive stress test cycle (multi-dimensional environment and load simulation), the recommended duration of a single comprehensive stress test cycle is 12 hours, which includes four core test modules: the "transient voltage test" in Step 3 is "sudden pulse interference", and the "dynamic voltage fluctuation test point" is "continuous periodic interference". The triggering times of the two should be staggered to avoid the superposition of test conditions.

[0028] Temperature cycling test (environmental stress): The temperature range is cycled from -40℃ (low temperature storage) to 25℃ (normal temperature transition, 10 minutes) to 70℃ (high temperature operation, 2 hours) to 25℃ (transition, 10 minutes), with each temperature cycle lasting 4 hours. A high and low temperature chamber (such as ESPEC SH-241) is used to control the temperature, and the temperature change rate does not exceed 5℃ / minute to avoid sudden temperature changes that could damage the hard drive.

[0029] Load mode switching (read / write stress): The load mode is switched in the following order: "continuous sequential read / write (1 hour, 50%) → random read / write (1 hour, 40%, block size 4K) → idle state (0.5 hours, 10%)". The load parameters are configured using FIO or IOMeter, and the read / write ratio is set to 1:1 to simulate a balanced usage scenario.

[0030] Transient voltage test (power supply stress): During load mode operation, voltage pulses are inserted (low voltage drop: from standard voltage to 80% of standard voltage, lasting 100 milliseconds; high voltage spike: from standard voltage to 120% of standard voltage, lasting 50 milliseconds), triggered twice per hour; synchronously record whether the hard drive experiences power loss, restart or data interruption.

[0031] Data integrity verification (result verification): After each load mode ends, read the verification file written in step 1, recalculate the MD5 or SHA256 value and compare it with the initial value; if the verification value is inconsistent, mark it as "data error" and record the test environment (temperature, voltage, load type) when the error occurred.

[0032] Step 4: SMART data acquisition and intelligent analysis. SMART data is acquired once after each comprehensive stress test cycle (12 hours). NVMe CLI (Linux system) or Smartmontools tools are used uniformly to ensure data format consistency.

[0033] Intelligent Analysis Dimensions: Trend Analysis: Tracks the growth rate of bad block count and average write cycles (PE Cycle) to determine the hard drive's aging speed. Anomaly Identification: If the error correction count (ECC Error Count) increases more than 5 times in a single cycle, the temperature consistently exceeds 75°C, or there are at least 2 instances of "mutual interference exceeding the limit," the hard drive is marked as an "abnormal sample." Lifespan Prediction: Based on the growth rate of the average write cycles (PE Cycle) and the initial remaining lifespan, the estimated lifespan of the hard drive under current test conditions is calculated.

[0034] Step 5: Test Loop Judgment (Ensuring Test Sustainability) First, set the number of loops based on the test objectives (minimum 50 loops are recommended for mass production verification, and minimum 20 loops for performance evaluation): If the current loop count has not reached the preset count, return to Step 3 to continue the next round of comprehensive stress testing, while retaining historical SMART data for trend analysis; if the current loop count has reached the preset count, stop the loop and proceed to Step 6. Furthermore, if a hard drive displays "data error" or "abnormal sample" markers 3 or more times during the test, the test for that hard drive can be terminated early to analyze the cause.

[0035] Step 6: Test Report Generation (Data Visualization and Conclusion Output) includes: Test Overview: This includes test sample information (model, batch, quantity) and test environment parameters (temperature range, voltage level, number of cycles). Data Presentation: Trend charts show the average write / erase cycles (PE Cycle) and bad block count as a function of cycle count for each group of hard drives; comparative data illustrates the differences in anomaly rates, data error rates, and estimated lifespans between different voltage groups. Conclusions and Recommendations: Clearly define the hard drive's stability level under normal / critical voltage conditions (e.g., "Pass," "Needs Improvement," "Fail"); propose improvement directions for anomalous samples (e.g., optimizing power supply design, adjusting firmware cooling strategies), and indicate the interference-sensitive voltage range and anti-interference performance level.

[0036] Figure 2 yes Figure 1 The flowchart for step 2 is as follows: Figure 2 As shown, step 2 includes: step 201, dividing the solid-state drives into multiple groups according to voltage strategy, each group containing multiple solid-state drives and setting a "mutual interference" detection mechanism for multiple solid-state drives; step 202, setting voltage test points for each group of solid-state drives, wherein the voltage test points include at least rated voltage, voltage pull-off, and dynamic voltage fluctuation.

[0037] Through preprocessing, voltage grouping and special testing settings (including steps 201 and 202), cyclic stress testing, data acquisition and analysis and report generation, the stability, reliability and lifespan characteristics of solid-state drives under multiple conditions are verified, which is suitable for quality verification or product performance evaluation before mass production.

[0038] Step 201, "Grouping," divides the solid-state drives into three groups based on the two mainstream power supply types: 12V and 5V, to cover different power supply scenarios: Group A corresponds to normal power supply environments, with the 12V power supply range controlled at 11.4V-12.6V and the 5V power supply range controlled at 4.75V-5.25V; Group B corresponds to unstable power supply scenarios, with the 12V power supply range adjusted to 10.8V-11.3V and the 5V power supply range adjusted to 4.5V-4.74V; Group C corresponds to power fluctuation overload scenarios, with the 12V power supply range set at 12.7V-13.2V and the 5V power supply range set at 5.26V-5.5V.

[0039] Each sample group must contain at least 3 pieces to ensure statistical significance. Hard drives from the same batch must be evenly distributed across the groups to avoid test deviations due to batch differences. Power supply control must use a programmable DC power supply (such as Keysight N6705B) to precisely regulate the power supply voltage of each group, with the error strictly controlled within ±0.05V, providing a stable foundation for subsequent testing.

[0040] The core design purpose of the intra-group "mutual interference" detection mechanism is to meet the requirement of "setting up a mutual interference detection mechanism for multiple solid-state drives" in step 201, simulate the parallel operation scenario of multiple hard drives such as server hard drive arrays, and focus on verifying the impact of electromagnetic interference and power load fluctuations on hard drive stability, so as to ensure the reliable operation of hard drives in a cluster environment.

[0041] Specific implementation plan: Hardware layout: As required by step 201 "each group contains multiple solid-state drives", all hard drives in the same group are installed in a "closely parallel" manner, with the spacing controlled within 2cm (to restore the density of actual server hard drive bays), and all hard drives share the same power supply branch (drawing a single circuit from the programmable DC power supply) to avoid independent power supply and eliminate interference factors; each group needs to be equipped with an additional electromagnetic interference (EMI) monitor (such as R&S ESR26), with the monitor probe kept 1cm away from the hard drive casing, to record the interference signal strength in the frequency range of 10kHz-1GHz in real time and retain the original interference data.

[0042] Interference Detection Scenarios and Judgment Criteria: Parallel Read / Write Interference: The trigger condition is that all hard drives in the group simultaneously perform sequential read / write operations at 100% load using the FIO tool. Monitor the throughput fluctuation of a single drive and the maximum IO latency. If the throughput fluctuation exceeds 10%, or the IO latency exceeds 50ms and lasts for 10s, it is judged as "interference exceeding the standard." Record the load parameters and interference signal strength at the time of exceeding the standard. Power Supply Load Fluctuation Interference: The trigger condition is that 1-2 hard drives in the group are randomly selected, and read / write operations are suddenly started and stopped at 5-minute intervals. Monitor the voltage fluctuation value of the power supply branch and the number of hard drive power failures / restarts. If the voltage fluctuation exceeds ±0.1V, or there is one or more power failures, it is judged as "interference exceeding the standard." Simultaneously capture the power supply voltage waveform (using a Tektronix MDO3024 oscilloscope). Long-term parallel operation interference: The trigger condition is that all hard drives in the group continuously perform load operations for 72 hours. Monitor the increment of SMART error count and the number of data verification failures. If the error count increment exceeds 3 or the number of verification failures is not less than 1, it is judged as "interference exceeds the standard". Record the error type and occurrence time in detail.

[0043] Data recording requirements: Generate one "Mutual Interference Detection Report" per hour (echoing the implementation of step 201 "Detection Mechanism"). The report must include the EMI signal strength change curve, throughput fluctuation details, and error records. If "interference exceeds the standard", the power supply waveform diagram and load log at the time of the exceedance must be archived separately to provide a basis for subsequent analysis.

[0044] Voltage test point setup (corresponding to step 202): Following the requirement in step 202 to "set voltage test points, including at least rated voltage, voltage deviation, and dynamic voltage fluctuation," configure three types of test points for each group of hard drives. The specific parameters and execution logic are as follows: Test point parameter configuration: Rated voltage test point: According to the basic requirements of step 202, the standard voltage of the 12V power supply group is set to 12V (error ±0.05V), and the standard voltage of the 5V power supply group is set to 5V (error ±0.05V); the test time of this test point accounts for 40%, mainly simulating stable power supply scenarios such as desktop standby and normal operation of ordinary servers, to verify the basic performance of hard drives under standard voltage.

[0045] Voltage bias test points: Covering the "voltage bias" requirement in step 202, the 12V power supply group includes two types of bias: positive bias and negative bias. Positive bias is 12.6V (corresponding to rated voltage +5%) and 13.2V (corresponding to rated voltage +10%), and negative bias is 11.4V (corresponding to rated voltage -5%) and 10.8V (corresponding to rated voltage -10%). The 5V power supply group also includes positive bias and negative bias: positive bias is 5.25V (corresponding to rated voltage +5%) and 5.5V (corresponding to rated voltage +10%), and negative bias is 4.75V (corresponding to rated voltage -5%) and 4.5V (corresponding to rated voltage -10%). This test point accounts for 30% of the test time, simulating unstable power supply scenarios such as power adapter aging and mains voltage fluctuations, to verify the hard drive's voltage adaptability.

[0046] Dynamic voltage fluctuation test points: Meeting the requirements of step 202 "Dynamic voltage fluctuation", the fluctuation frequency of both the 12V and 5V power supply groups is controlled within 1kHz-10kHz, with the fluctuation amplitude of the 12V power supply group being ±0.5V (relative to the rated voltage) and the fluctuation amplitude of the 5V power supply group being ±0.2V (relative to the rated voltage). During the test, an alternating waveform mode of sine wave and square wave is used, accounting for 30% of the test time, to simulate the voltage fluctuation scenario caused by switching power supply ripple interference and simultaneous start-up and shutdown of multiple devices, verifying the hard drive's tolerance to dynamic power supply environment.

[0047] Test point execution logic: The single voltage test cycle is set to 2 hours, cycling in the order of "rated voltage (48 minutes) → positive bias (36 minutes) → negative bias (36 minutes) → dynamic fluctuation (36 minutes) → rated voltage (6 minutes, recovery transition)" to ensure uniform coverage of the three types of test points. When switching voltage, the slope of change must be controlled, with a rate not exceeding 0.5V / second (e.g., switching from 12V to 13.2V should take at least 2.4 seconds) to avoid sudden voltage changes damaging the hard drive hardware. During the operation of each test point, the "Power On Hours" and "Current Power Consumption" parameters of the hard drive must be read in real time via the NVMe CLI. The relationship between hard drive power consumption and operational stability under different voltage conditions is analyzed to provide data feedback for the test point settings in step 202.

[0048] The recommended cycle time for a single comprehensive stress test is 12 hours, which includes four core test modules and needs to be coordinated with the voltage test point in step 2 (step 202): the "transient voltage test" in this step belongs to "sudden pulse interference", and the "dynamic voltage fluctuation test point" belongs to "continuous periodic interference". The triggering time of the two should be staggered to avoid the test conditions from overlapping and causing the result deviation.

[0049] Temperature cycling test (environmental stress): The temperature range is cycled from -40℃ (low temperature storage) to 25℃ (normal temperature transition, 10 minutes) to 70℃ (high temperature operation, 2 hours) to 25℃ (transition, 10 minutes), with each temperature cycle lasting 4 hours. A high and low temperature chamber (such as ESPEC SH-241) is used to control the temperature, and the rate of change does not exceed 5℃ / minute to avoid sudden temperature changes that could damage the hard drive.

[0050] Load mode switching (read / write stress): The load mode is switched in the following order: "continuous sequential read / write (1 hour, 50% of the load) → random read / write (1 hour, 40% of the load, block size 4K) → idle state (0.5 hours, 10% of the load)". The read / write ratio is set to 1:1 using FIO or IOMeter configuration parameters to simulate a balanced usage scenario.

[0051] Transient voltage test (power supply stress): Voltage pulses are inserted during load mode operation—low voltage drop is a drop from the standard voltage to 80% of the standard voltage (lasting 100 milliseconds), and high voltage spike is a rise from the standard voltage to 120% of the standard voltage (lasting 50 milliseconds), triggered twice per hour; synchronously record whether the hard drive experiences power loss, restart, or data interruption.

[0052] Data integrity verification (result verification): After each load mode ends, read the verification file written in step 1, recalculate the MD5 or SHA256 value and compare it with the initial value; if the verification value is inconsistent, mark it as "data error" and record the temperature, voltage and load type when the error occurred.

[0053] Collect SMART data once after each comprehensive stress test cycle (12 hours), using either NVMe CLI (Linux system) or Smartmontools to ensure consistent data format.

[0054] Figure 3 yes Figure 1 The flowchart for step 3 is as follows: Figure 3As shown, step 3 includes: Step 301, setting the temperature range in the temperature chamber to -10℃ to 85℃, and performing temperature cycling on multiple groups of solid-state drives according to a preset curve, with each extreme temperature maintained for 10 hours; Step 302, performing 100% read / write load tests on multiple groups of solid-state drives in the high and low temperature segments of the temperature cycle, and verifying the firmware algorithm and power supply stability; Step 303, during the load test, applying voltage drops or spikes to the solid-state drive power supply through a programmable DC power supply to simulate power glitches caused by sudden loading / unloading of high-power devices such as CPUs and GPUs.

[0055] The duration of a single comprehensive stress test cycle is adjusted to 48 hours according to steps 301-303 (including temperature cycling, extreme temperature load, power interference, and conventional test modules). It is necessary to ensure that steps 301-303 are coordinated with the original load and voltage tests. Step 302's "100% read / write load" focuses on extreme temperature scenarios, and step 303's "power glitches" simulates interference from specific equipment, complementing the original "dynamic voltage fluctuation" and "transient voltage test".

[0056] In the temperature cycling parameter settings (corresponding to step 301), the temperature range and holding time are as follows: As required in step 301, the temperature range in the high and low temperature chamber (such as ESPEC SH-241) is set to -10℃~85℃, with the low temperature segment (-10℃) and the high temperature segment (85℃) each held for 10 hours, and the normal temperature transition segment (25℃) held for 2 hours. The total duration of a single temperature cycle is 24 hours (holding at -10℃ for 10 hours → raising the temperature to 25℃ for 2 hours → raising the temperature to 85℃ for 10 hours → lowering the temperature to 25℃ for 2 hours). Temperature change control: The heating rate is set to 3℃ / minute (it takes 11.7 minutes to heat from -10℃ to 25℃, and 20 minutes to heat from 25℃ to 85℃), and the cooling rate is set to 2℃ / minute (it takes 30 minutes to cool from 85℃ to 25℃, and 17.5 minutes to cool from 25℃ to -10℃) to avoid damage to the hard drive casing or interface caused by sudden temperature changes; a temperature sensor (accuracy ±0.1℃) must be placed inside the high and low temperature chamber to provide real-time feedback on the deviation between the actual temperature and the set value, ensuring that the deviation does not exceed ±1℃.

[0057] In the extreme temperature range read / write load test (corresponding to step 302), the timing and parameters of the load execution are as follows: As required by step 302, 100% read / write load tests are performed on all groups of solid-state drives only during the low temperature range (10 hours of holding at -10℃) and the high temperature range (10 hours of holding at 85℃) in step 301; the load parameters are configured using the FIO tool—the read / write mode is "mixed read / write" (50% read ratio, 50% write ratio), the block size is 8K, and the queue depth is 32, ensuring that the hard drive is running at full load, and the load duration is consistent with the extreme temperature holding time (10 hours each).

[0058] Firmware algorithm and power stability verification: Firmware algorithm verification: Using dedicated firmware logging tools provided by hard drive manufacturers (such as Samsung Magician Enterprise Edition, Intel SSD Toolbox), monitor the firmware's "temperature control strategy" (such as whether overheating throttling is triggered) and "error recovery mechanism" (such as whether ECC error correction is normal) in real time during load. If the firmware becomes unresponsive, the speed reduction is abnormal (such as a sudden drop in throughput of more than 50%), or the error recovery times out, it is marked as "firmware algorithm abnormal".

[0059] Power supply stability test: Use an oscilloscope (Tektronix MDO3024) to monitor the real-time voltage of the hard drive power supply interface and record the voltage fluctuation amplitude during the load (normal fluctuation should be ≤ ±0.03V); at the same time, read the power supply current change through a programmable DC power supply (Keysight N6705B). If the current suddenly increases by more than 20% of the rated value (e.g., for a 12V power supply hard drive with a rated current of 1A, a sudden increase of more than 1.2A) or the current is interrupted, mark it as "power supply stability abnormal".

[0060] In the power supply glitches simulation under load (corresponding to step 303), the interference application conditions and parameters are as follows: As required by step 303, during the execution of the "100% read / write load" in step 302 (i.e., the -10℃ and 85℃ load periods), a voltage drop or spike is applied to the solid-state drive power supply branch through a programmable DC power supply to simulate the interference scenario of sudden loading / unloading of high-power devices such as CPU and GPU; the specific parameters are as follows: Voltage drop: from the standard voltage of the current test group to 70% of the standard voltage (e.g., from 8.4V for the 12V group to 3.5V for the 5V group), lasting for 200 milliseconds; Voltage spike: from the standard voltage of the current test group to 130% of the standard voltage (e.g., from 15.6V for the 12V group to 6.5V for the 5V group), lasting for 50 milliseconds; Interference is triggered once per hour (10 times for 10 hours under low temperature load, 10 times for 10 hours under high temperature load), and the triggering time is the same as the "simulation of CPU / GPU". Loading signal synchronization (outputs a 5V trigger signal via a signal generator, which is linked with the power supply interference signal).

[0061] Post-interference status monitoring: After each voltage drop or spike, immediately read the hard drive's "Power Cycle Count" and "Hardware Error Count" via NVMe CLI. If the Power Cycle Count increases (indicating an unexpected hard drive restart) or the Hardware Error Count increases, record the interference type (drop / spike), temperature environment (-10℃ / 85℃), and error code. Simultaneously, re-execute the MD5 / SHA256 comparison of the file verification in step 1. If the verification fails, it is determined as "data corruption," and the correlation between power interference and data integrity is analyzed.

[0062] The routine test modules are executed in coordination, and the load mode is switched (non-extreme temperature period): During the normal temperature transition period (25℃, 4 hours in total) in step 301, the routine load mode is switched - "continuous sequential read and write (1 hour, accounting for 25%) → random read and write (2 hours, accounting for 50%, 4K block size) → idle state (1 hour, accounting for 25%)". The parameters are configured using the IOMeter tool to avoid excessive load outside of extreme temperatures, which may cause premature aging of the hard drive.

[0063] Transient voltage test (non-extreme temperature load period): During normal temperature load execution, insert transient voltages according to the original parameters (low voltage drop to 80% of standard voltage for 100ms, high voltage spike to 120% of standard voltage for 50ms), triggered once every 2 hours, staggered from the "power supply glitches" in step 303 to avoid interference superposition.

[0064] Data integrity verification: After each load module (including the extreme temperature load in step 302 and the normal load in 3.4.1) is completed, the verification file is read and the verification value is recalculated and compared with the initial value. If the verification is inconsistent, in addition to recording the temperature and voltage environment, it is necessary to additionally record whether it is accompanied by power interference or firmware abnormality in step 303, so as to provide a basis for subsequent fault location.

[0065] At each key node in step 3 (step 301 end of low temperature segment, end of high temperature segment, end of normal temperature transition segment, 3 times / cycle), SMART data is collected. In addition to the original parameters, “Temperature History” (maximum / minimum temperature history), “Power Cycle Count” (number of power-on cycles) and “Hardware Error Count” are collected. The NVMe CLI (Linux system) or Smartmontools tool is used to ensure data format consistency.

[0066] Trend Analysis: In addition to tracking the bad block count and the growth rate of the average erase / write cycles (PE Cycle), the analysis now includes the correlation between "Temperature History" and bad block growth (e.g., whether bad blocks surge after 85°C load) and the correspondence between "Power CycleCount" and power supply interference (step 303).

[0067] Anomaly Identification: Based on the original judgment criteria, the following are added: If "firmware algorithm anomaly" or "power stability anomaly" occurs ≥1 time in step 302, or "data corruption" occurs ≥1 time or "Hardware ErrorCount" increases ≥2 times in step 303, they are all marked as "abnormal samples".

[0068] Lifetime prediction: Combine the PE Cycle growth rate under extreme temperature load in step 302 (e.g., whether the PE growth at 85℃ for 10 hours is twice that at room temperature) to revise the lifetime prediction model and ensure that the prediction results are more in line with the actual use scenarios at high and low temperatures.

[0069] Before step 302, the method further includes: step 304, performing at least 3 random abnormal power failure tests on each solid-state drive at a specific temperature switching point, and randomly cutting off the power supply to the solid-state drive through a high-speed switch.

[0070] In the abnormal power-off test at the temperature switching point (corresponding to step 304), based on the temperature cycle process in step 301, the "specific temperature switching point" is defined as a dynamic process node where the temperature switches from the stable segment to the transition segment, or from the transition segment to the stable segment. Specifically, it includes 4 key nodes (within a single temperature cycle): Low temperature → Normal temperature switching point: After the end of the -10℃ stable segment, during the process of starting to rise to 25℃ (when the actual temperature rises to -5℃, it is considered the switching point triggering time); Normal temperature → High temperature switching point: After the end of the 25℃ transition segment, during the process of starting to rise to 85℃ (when the actual temperature rises to 50℃, it is considered the switching point triggering time); High temperature → Normal temperature switching point: After the end of the 85℃ stable segment, during the process of starting to drop to 25℃ (when the actual temperature drops to 50℃, it is considered the switching point triggering time); Normal temperature → Low temperature switching point: After the end of the 25℃ transition segment, during the process of starting to drop to -10℃ (when the actual temperature drops to -5℃, it is considered the switching point triggering time).

[0071] Power-down test hardware and parameter configuration: High-speed switch selection and connection: As required by step 304, a high-speed DC solid-state switch (such as Keysight N7744A, switching speed ≤10 microseconds) is used, connected in series in the power supply branch of each group of solid-state drives (located between the programmable DC power supply and the hard drive). "Random disconnection" is achieved through switch control software (such as Keysight PathWave) – each time triggered, the switch disconnection duration is randomly set to 100-200 milliseconds (simulating a brief power outage in a real-world scenario), avoiding a fixed duration that leads to uniform test results. Test frequency and distribution: Within a single 48-hour cycle, each solid-state drive undergoes 3 abnormal power-down tests, and the 3 tests must be evenly distributed across 4 switching points (e.g., the first test at the "low temperature → room temperature switching point", the second at the "high temperature → room temperature switching point", and the third at the "room temperature → low temperature switching point"). The same hard drive is not tested consecutively at the same switching point to ensure coverage of interference scenarios with different temperature change directions.

[0072] Power-down test execution and status verification, load coordination control: When the power-down test is triggered at the temperature switching point, the hard drive must be in a "light load" state - configure the read and write load (read ratio 60%, write ratio 40%, block size 4K, queue depth 8) through the FIO tool to avoid excessive damage caused by power-down under extreme load; the load start time must be 5 minutes earlier than the switching point trigger time to ensure that the hard drive is in a stable operating state when power-down occurs.

[0073] Power outage verification process: Power restoration: After the switch is disconnected for a random period of time, power is automatically restored (the voltage rise rate during restoration is controlled at 0.3V / millisecond to avoid voltage surges). The time from power restoration to the hard drive being recognized by the system again is recorded (normally ≤3 seconds). If it is not recognized after more than 10 seconds, it is marked as "Power failure to boot"; SMART data initial check: After power restoration, immediately read "Power Cycle Count" (number of power-on cycles, should increase by 1 compared to before the power outage), "Hardware Error Count" (hardware error count, normally no new errors), and "Unexpected Power Loss Count" (unexpected power loss count, should increase by 1 compared to before the power outage) via NVMe CLI. If there are abnormal counts or new errors, record the error code and the temperature at the switching point; Data integrity verification: After booting the hard drive, re-execute the MD5 / SHA256 comparison of the verification file written in step 1. If the verification values ​​are inconsistent, it is determined as "data corruption after power failure". Simultaneously record the temperature change rate at the time of power failure (e.g., from -10℃ to - The test measures the following parameters: speed at 5℃, duration of power failure, and stability reproduction test. If a hard drive experiences "boot failure" or "data corruption" at a certain switching point, the power failure test (without changing parameters) must be repeated once at the same switching point to confirm whether it is an occasional failure. If it occurs repeatedly, it is marked as a "switching point power failure sensitive sample".

[0074] The coordination with other modules requires that the power-down test and temperature control be synchronized: the temperature feedback signal of the high and low temperature chamber is linked with the high-speed switch trigger signal to ensure that the power-down is only performed when the temperature at the switching point reaches the set value (such as -5℃, 50℃), and the temperature deviation must be controlled within ±0.5℃; avoid other interference periods: the power-down test in step 304 must be staggered from the "power glitch" (extreme temperature stable period) in step 303 and the "100% read / write load" (extreme temperature stable period) in step 302, and should only be performed during the light load period during the temperature switching process to avoid multiple interferences that make fault location difficult.

[0075] Based on the existing key node data collection, a new SMART data collection is added for "within 3 minutes after each step 304 power failure test". The data collection focuses on recording "Unexpected Power Loss Count", "Power Cycle Count", and "CRC Error Count". At the same time, the data collection also includes the hard drive re-identification time and the temperature values ​​before and after the power failure, to ensure that the data can be correlated with the power failure scenario.

[0076] The analysis now includes the correlation between "Unexpected Power Loss Count" and "Number of Data Corruptions" (e.g., whether data corruption occurred after 3 power outages) and the difference in "Power-Off Startup Failure Rate" at different temperature switching points (e.g., whether startup failure is more likely at the "room temperature → low temperature switching point"). Based on the existing criteria, the following additions are made: In step 304, if "≥1 startup failure after power outage," "≥1 data corruption after power outage," or "Unexpected Power Loss Count" does not match the actual number of power outages (e.g., 3 actual power outages but the count only increases by 2), these will be marked as "abnormal samples." Based on the power outage test results in step 304, if a hard drive frequently experiences hardware errors after a power outage, the lifespan prediction model needs to be revised (e.g., shortening the estimated lifespan by 20%) to reflect the impact of power outage interference on the long-term reliability of the hard drive.

[0077] Figure 4 yes Figure 1 The flowchart for step 4 is as follows: Figure 4 As shown, step 4 includes: Step 401: After each loop, automatically run disk tools to check for bad blocks and use data verification tools to compare the hash values ​​of written data and read data; Step 402: Collect multi-dimensional SMART data of each solid-state drive in real time, including temperature, write volume, erase count, and number of remapped sectors, and combine it with historical data to perform trend modeling and anomaly warning analysis. Step 403: Use machine learning algorithms to perform correlation analysis on multi-dimensional SMART data, identify potential failure modes, and quantify the degree of reliability degradation.

[0078] In the bad block check and data verification after the cycle (corresponding to step 401), as required by step 401, after the 48-hour cycle is completed (the hard drive is still at the ambient temperature of 25°C in the high and low temperature chamber), the disk bad block detection tool is automatically run. For Linux systems, the badblocks tool is used (parameters configured as "non-destructive read / write test", block size is set to the physical sector size of the hard drive, usually 512 bytes or 4096 bytes), and for Windows systems, vendor-specific tools (such as Seagate SeaTools Enterprise Edition, Kingston SSD Manager) are used. The detection range covers the entire capacity of the hard drive to avoid missing bad blocks by only detecting some areas. During the detection process, the location of bad blocks (identified by LBA logical block address) and the type of bad blocks (temporary bad blocks / permanent bad blocks) are recorded in real time. If a new bad block is detected (compared with the initial record in step 1), it is marked as "new bad block added after cycle", and the number of new bad blocks and their distribution areas are counted (such as whether they are concentrated in the corresponding area of ​​a certain storage chip).

[0079] Data hash value comparison and verification: Verify file reading and hash calculation: Automatically call data verification tools (such as md5sum, sha256sum command-line tools, or the graphical tool HashTab) to read the full-capacity verification file written in step 1, and recalculate the MD5 value or SHA256 value of the file (consistent with the initial verification algorithm); if the verification file cannot be read completely due to test corruption, immediately record the starting LBA address of the reading failure, associate it with the detected bad block position, and determine whether the file corruption is caused by bad blocks.

[0080] Hash value comparison and result determination: The recalculated hash value is automatically compared with the initial hash value recorded in step 1. If they match, the data integrity is determined to be normal; if they do not match, the data is determined to be corrupted. The file fragments with mismatched hash values ​​are located using tools (accurate to 1MB fragments), and the cause of the corruption is analyzed in conjunction with the bad block detection results (caused by bad blocks / power interference / firmware error). All comparison results are automatically generated into a log file, which includes the comparison time, tool version, hash value details, and location of the corrupted fragment. The log must be archived in association with the corresponding hard drive number.

[0081] Step 4: Real-time acquisition and multi-dimensional analysis of SMART data (including steps 402-403) is upgraded from the original "SMART data acquisition and intelligent analysis" to a three-layer architecture of "real-time acquisition + trend modeling + machine learning analysis", fully implementing the requirements of steps 402 and 403.

[0082] In the multi-dimensional SMART data real-time acquisition (corresponding to step 402), as required by step 402, the focus is on four core dimensions: "temperature, write volume, erase count, and remapped sector count," while supplementing related parameters—the temperature dimension adds "peak temperature" and "temperature fluctuation amplitude" (the difference between the highest and lowest temperatures within an hour); the write volume dimension uses "total bytes written (TBW)" as the unit (accurate to GB level); the erase count dimension records "average PE cycle count" and "maximum PE cycle count per chip" (to avoid average data masking local overload); the remapped sector count dimension distinguishes between "number of remapped sectors" and "number of sectors to be remapped"; the acquisition frequency is upgraded from the original "critical node acquisition" to "real-time periodic acquisition," acquiring full parameters once every 5 minutes, and automatically executing them using timed scripts written through the nvme-cli tool (Linux) or Smartmontools tool (cross-platform) (such as Linux crontab scheduled tasks). The acquired data is stored in a database (such as InfluxDB, which supports time-series data storage) in real time to avoid data loss.

[0083] Historical data association and trend modeling: Historical data integration: The real-time data collected in the current cycle is associated with the historical cycle data (such as the previous 1st, 5th, and 10th cycles) according to the hard drive number to form a single disk full life cycle time series dataset. The data dimensions include "collection timestamp, temperature, TBW, PE cycle number, number of remapped sectors, and test stage identifier (such as low temperature segment / high temperature segment / power failure test segment)". Trend modeling method: A trend model is constructed using time series analysis algorithms. For "TBW - Time" and "PE cycle count - Time", a linear regression model is used (assuming that the write volume and time increase linearly under normal use), and the growth slope (such as the daily TBW growth value and the daily PE cycle count growth value) is calculated. For "remapped sector count - Time", an exponential regression model is used (because the growth of the number of remapped sectors may accelerate with hard drive aging), and the growth curve is fitted. The model fit should reach R²≥0.9 (R² is the coefficient of determination, the closer it is to 1, the better the model fits the actual data). If the fit is lower than 0.8, data anomalies (such as jump values ​​caused by a certain collection error) need to be investigated and the model corrected.

[0084] Anomaly warning threshold setting and triggering: Dynamic warning thresholds are set based on trend models—Temperature warning threshold is set to "exceeding the highest temperature of the same period in history by 10℃" or "temperature exceeding 75℃ for 3 consecutive collections"; Write volume warning threshold is set to "daily TBW growth exceeding twice the historical average daily growth"; PE cycle count warning threshold is set to "reaching 80% of the manufacturer's nominal lifespan PE cycle count"; Remapped sector count warning threshold is set to "more than 5 new sectors added in a single day" or "total number reaching 50% of the manufacturer's warranty threshold"; When real-time collected data triggers any of these thresholds, the system automatically generates warning information (including warning parameters, current value, threshold, and associated test stage), pushes it to testers via email or test management platform, and simultaneously suspends the next round of cycle testing for that hard drive until manual confirmation is received to decide whether to continue.

[0085] In the SMART data correlation analysis based on machine learning (corresponding to step 403), as required by step 403, an algorithm suitable for time series data and correlation analysis is selected—the correlation analysis uses the "Random Forest Algorithm" (which can quantify the impact weight of each SMART parameter on reliability), and the failure mode identification uses the "Long Short-Term Memory Network (LSTM)" (which can capture long-term dependencies in time series data, such as how long after a parameter becomes abnormal it is likely to fail). In the data preprocessing stage, the SMART data needs to be standardized (converting each parameter value to the 0-1 range to eliminate dimensional differences, such as converting temperature from -10℃ to 85℃ to 0~1, and TBW from 0 to the manufacturer's nominal value to 0~1), outlier removal (using the 3σ principle to remove extreme values ​​caused by acquisition errors), and feature engineering (constructing derived features, such as the "temperature × PE cycle count" interactive feature and the "remapping sector growth rate" feature) to ensure the quality of the algorithm input data.

[0086] Potential Failure Mode Recognition: Pattern Classification Training: The LSTM model is trained using the SMART dataset of historical failed hard drives (such as hard drive data due to excessive bad blocks, failure due to remapping sector exhaustion, and failure due to firmware crashes). The failure modes are classified into three categories: "storage media failure", "firmware function failure", and "power adaptation failure". During the model training process, 5-fold cross-validation is used (the dataset is divided into 5 parts, 4 parts are used for training and 1 part for validation, and the cycle is repeated 5 times) to ensure the model's generalization ability (accuracy of failure mode recognition for new hard drives ≥85%).

[0087] Real-time pattern recognition: Input the real-time SMART timing data of the current hard drive into the trained LSTM model. The model outputs the probability of the hard drive belonging to various failure modes (e.g., "30% probability of storage media failure, 10% probability of firmware function failure, 5% probability of power adaptation failure, and 55% probability of normal state"). If the probability of a certain failure mode exceeds 40% for three consecutive collections, it is judged as "identifying a potential failure mode" and the typical characteristics of the mode are output (e.g., "storage media failure" corresponds to "a surge in the number of remapped sectors + high number of PE cycles") to help testers locate risk points.

[0088] Quantification of Reliability Degradation: The degree of degradation is quantified using a "health score," calculated based on the parameter weights output by the random forest algorithm. The health score is set to a maximum of 100 points, with the initial state (after step 1) at 100 points. Points are deducted for each SMART parameter according to its weight, such as "20 points deducted for each PE cycle reaching 80% of lifespan," "15 points deducted for each new 5 remapped sectors added in a single day," "10 points deducted for each consecutive temperature exceeding 75℃," and "30 points deducted for each instance of data corruption." The weight values ​​are determined by the calculation results of the "parameter-failure correlation" using the random forest algorithm (higher correlation results in greater weight, e.g., remapped sector weight 0.3, temperature weight 0.2, and write volume weight 0.1). The health score is updated hourly, displaying the hard drive degradation trend in real time (e.g., a drop from 100 to 90 points indicates mild degradation, and a drop below 60 points indicates severe degradation). When the score falls below 50 points, a "reliability warning" is automatically triggered, suggesting stopping testing and evaluating whether the hard drive meets quality standards.

[0089] In addition to the existing termination conditions, add the following rules for steps 401-403: Step 401 Trigger Termination: If a hard drive detects "≥3 new bad blocks" or "data corruption" in two consecutive cycles, or detects "≥5 new bad blocks" in a single cycle, immediately terminate the test of the hard drive and analyze the cause of the bad blocks (such as storage chip quality problems or excessive test stress). Step 402 Trigger Termination: If the warning message is not cleared for 3 consecutive times (e.g., the temperature continues to exceed the threshold, the number of remapped sectors continues to increase), or the value of a certain parameter jumps to more than 50% of the historical maximum value (e.g., the number of PE loops suddenly increases by 100 times in a certain collection), pause the test to check for problems with the collection tool or hard drive hardware. Step 403 Trigger Termination: If the health score drops below 50 points, or the probability of a certain failure mode identified by the model exceeds 50% for 5 consecutive data collections, terminate the hard drive test and output a failure risk assessment report. The preset number of cycles remains unchanged (≥25 for mass production verification, ≥10 for performance evaluation), but it is necessary to ensure that each hard drive completes at least 5 cycles before termination to obtain sufficient reliability data.

[0090] If no termination conditions are triggered, the next 48-hour cycle test will automatically start after completing step 401 verification and step 4 data upload. Before the next cycle, the test environment needs to be reset (e.g., clear temporary logs, restore the high and low temperature chamber to the initial temperature -10℃, and reload load parameters) to ensure that the initial conditions are consistent for each cycle.

[0091] Following step 404, the method further includes: Step 404: When a group of solid-state drives (SSDs) experiences the same type of SMART warning in three consecutive test cycles, a deep diagnostic protocol is triggered to call the underlying logs to analyze the changes in its NAND write amplification rate and FTL mapping table; Step 405: If the growth rate of the number of remapped sectors exceeds 1.5 times the threshold, it is determined that the voltage strategy has a reliability risk; Step 406: Finally, the degradation curves and failure correlation matrices of all groups are summarized into a test report to provide a basis for decision-making for mass production of products, and high-risk voltage strategy combinations are marked. At the same time, the subsequent test plan is optimized by combining the coupling relationship between environmental stress and electrical response.

[0092] In the SMART alert-triggered deep diagnostics (corresponding to step 404), strictly following the requirements of step 404, when a solid-state drive (single or multiple drives) within a certain voltage group experiences the "same type of SMART alert" in three consecutive test cycles, the deep diagnostic protocol is automatically triggered. The "same type of alert" is defined as an alert triggered by the same SMART parameter (such as triggering the "more than 5 new remapped sectors per day" alert in three consecutive cycles, or triggering the "temperature exceeds 75℃" alert in three consecutive cycles), excluding mixed alerts across parameters (such as a temperature alert in the first cycle and a remapped sector alert in the second cycle). When triggered, the alert hard drive number, the voltage group to which it belongs, the number of consecutive alert cycles, and the alert parameter type must be recorded to ensure that the diagnostic object and scenario are clear.

[0093] Log extraction tools and scope: After triggering in-depth diagnostics, the hard drive manufacturer's underlying log tools (such as Samsung SSD Toolbox Enterprise Edition's "Log Extractor" and Intel Data Center Tool's "Log Capture") are used to extract the hard drive's "NAND operation log", "FTL (flash transition layer) operation log", and "power management log". The log extraction scope covers the entire time period of three consecutive warning cycles (a total of 144 hours) to avoid incomplete analysis caused by log fragmentation.

[0094] NAND Write Amplification Calculation and Analysis: Extract the "Total Physical NAND Write Bytes" (i.e., the total amount of data actually written to the NAND chip by the hard drive) and "User Logic Write Bytes" (i.e., the amount of write command data issued by the user layer during the test) from the NAND operation log. Calculate the write amplification for three consecutive cycles using the formula "Write Amplification = Total Physical NAND Write Bytes / User Logic Write Bytes". If the write amplification shows an increasing trend (e.g., 1.2 in the first cycle, 1.5 in the second cycle, and 1.8 in the third cycle), or if the write amplification in a certain cycle exceeds 1.2 times the manufacturer's stated value (e.g., the manufacturer states 1.5, but the actual value reaches 1.8), it is determined to be "NAND write amplification abnormality", and associated warning parameters (e.g., remapping sector warning may be due to excessive write amplification leading to accelerated NAND wear).

[0095] FTL Mapping Table Change Analysis: Extract key indicators such as "number of mapping table entries," "mapping table update frequency," and "number of mapping table error corrections" from the FTL runtime log. Under normal circumstances, the number of mapping table entries should match the logical capacity of the hard drive (e.g., approximately 240 million 4K sector mapping entries for a 1TB hard drive), and the update frequency should fluctuate with read / write load but not increase suddenly. If, in three consecutive cycles, the mapping table update frequency increases by more than three times compared to normal periods (e.g., from 1000 times per hour to 3000 times per hour), or the number of mapping table error corrections increases by ≥3 times, it is determined that "FTL mapping table is unstable." Analyze whether the SMART parameter warning is caused by mapping table abnormalities (e.g., mapping table errors may lead to an increase in the number of remapped sectors).

[0096] Deep Diagnostic Results Output: After the diagnosis is completed, a "Deep Diagnostic Report" is generated, which includes "Warning Parameters and Cyclic Correlation Table", "Textual Description of NAND Write Amplification Rate Change Curve", "Statistics of Key Indicators in FTL Mapping Table", and "Preliminary Location of Abnormal Causes" (such as "Remapping Sector Warning Originates from NAND Write Amplification Abnormality, It is Recommended to Check the Load Balancing Strategy Under This Voltage Group"). The report automatically associates with the corresponding voltage group, providing a basis for the voltage strategy risk assessment in the subsequent step 405.

[0097] First, determine the "remapped sector number threshold" (i.e., the warning threshold set above, such as 5 new sectors added per day). Then, calculate the "actual growth rate" as required in step 405. Select the remapped sector number data of all hard drives in a certain voltage group and calculate the average daily growth rate for three consecutive cycles (formula: (Nth cycle daily new sector number + N+1th cycle daily new sector number + N+2th cycle daily new sector number) / 3). Then, calculate the "ratio of actual growth rate to threshold" (actual growth rate / threshold). If the ratio of the actual growth rate to the threshold for a voltage group exceeds 1.5 times (e.g., the threshold is 5 units / day, and the actual average growth rate is 7.5 units / day or more), the voltage strategy corresponding to that voltage group is immediately determined to have a reliability risk. For example, if this situation occurs in group B (low voltage threshold 10.8V-11.3V / 4.5V-4.74V), it is marked as "low voltage threshold voltage strategy risk," and all subsequent cyclic tests of all hard drives in that group are simultaneously suspended to avoid further damage. The risk determination result needs to be correlated with a deep diagnostic report (e.g., whether there is NAND write amplification anomaly in the group) to clarify whether the risk is related to the underlying hardware operation.

[0098] If the termination condition is not triggered and there is no voltage strategy risk (step 405 no risk is determined), after completing the data analysis in step 4 and the in-depth diagnosis in step 404 (if triggered), the next round of 48-hour cycle testing will be automatically started; if a group is determined to have a risk (step 405), only the testing of that group will be suspended, while other risk-free groups will continue to be executed to ensure testing efficiency.

[0099] Following step 406, the method further includes: Step 407: Using a weighted fusion model to perform a health score on the test report, combining the degradation degree of key SMART parameters with historical failure data, and outputting the failure probability through a logistic regression algorithm; Step 408: When the failure prediction probability of a certain group of solid-state drives exceeds a set threshold for two consecutive test cycles, the system automatically marks the group of solid-state drives as high-risk devices and activates a pre-failure warning mechanism, recommending early replacement; Simultaneously, the system synchronously uploads the FTL logs, bad block distribution map, and voltage fluctuation records of the high-risk devices to the cloud analysis platform, and performs horizontal comparison with historical data of products in the same batch to further verify the consistency of failure trends; Step 409: If a group of solid-state drives exhibits similar degradation characteristics in the same test phase, the model will automatically update the reliability baseline of the group of solid-state drives and adjust the screening criteria in the mass production stage to ensure stable and reliable shipment quality.

[0100] In the weighted fusion model health score and failure probability calculation (corresponding to step 407), the weighted fusion model is constructed and the health score is calculated as follows: According to the requirements of step 407, the weighted fusion model is used to calculate the health score of a single disk. First, the core weight parameters are determined: five key indicators are selected: "number of remapped sectors (weight 0.3), number of PE cycles (weight 0.25), NAND write amplification rate (weight 0.2), temperature fluctuation range (weight 0.15), and number of mapping table errors (weight 0.1)" (the weights are determined based on the correlation analysis of historical failure data, and the higher the failure correlation, the greater the weight); each indicator is standardized (the actual value is converted into a score of 0-100, such as 20 points for reaching 80% of the lifespan and 50 points for reaching 50%), and then the final score is calculated according to "health score = Σ (standardized score of indicator × corresponding weight)" (the full score is 100 points, and the higher the score, the better the health); the health score is updated once per cycle and linked with the historical trend model to form a two-dimensional health assessment of "real-time score + trend change".

[0101] Logistic Regression Algorithm Failure Probability Output: Algorithm Training and Input Features: The logistic regression model is trained using historical test data (including normal and failed samples) of the same model of solid-state drive. The input features are "health score, remapped sector growth rate, write amplification rate change rate, and number of times the temperature exceeds the threshold". The output is "the probability of failure occurring within the next 3 cycles" (values ​​range from 0 to 1, where 0 represents no risk of failure and 1 represents certain failure). During model training, 10-fold cross-validation is used to ensure a prediction accuracy of ≥90%, and the model parameters are iteratively optimized periodically with new test data to avoid overfitting.

[0102] Real-time probability calculation and result application: After each loop, the above input features of the current hard drive are substituted into the trained logistic regression model, and the failure probability is automatically output. If the probability is <0.3, it is judged as "low risk" and the normal test continues. If 0.3≤probability<0.6, it is judged as "medium risk" and the SMART data acquisition frequency is increased (from once every 5 minutes to once every 2 minutes). If the probability is ≥0.6, it is judged as "high risk warning" and the pre-failure handling mechanism in subsequent step 408 is triggered.

[0103] Based on the original "cycle termination + voltage strategy risk assessment", a new step 408, "high-risk equipment marking and cloud upload", has been added to form a two-dimensional management of "strategy risk + equipment risk".

[0104] As required by step 408, if the failure probability of a certain solid-state drive (output of step 407) exceeds the set threshold (usually set to 0.6) for two consecutive test cycles, the system will automatically mark the drive as a "high-risk device" and immediately push a pre-failure warning (including device SN, current failure probability, and abnormal values ​​of key SMART parameters) to the test manager through the test management platform. At the same time, the next round of testing for the device will be suspended, and it is recommended to replace it in advance or perform offline deep testing. The warning information must be written to the device test log simultaneously, indicating the specific number of cycles and probability value that triggered the warning.

[0105] After marking high-risk devices, the system automatically collects the device's core data and uploads it to a cloud-based analytics platform (such as AWS IoT or a self-built private cloud). The data includes "complete FTL operation logs (last 3 cycles), detailed records of bad block distribution (including LBA addresses and their associated memory chips), raw voltage fluctuation data (power supply voltage waveforms for the last 3 cycles), and health score change curves." After receiving the data, the cloud platform automatically compares it with historical data from other devices in the same batch: if the high-risk device's "bad block distribution area, FTL error type, and voltage-sensitive period" are consistent with ≥3 devices in the same batch, it is determined to be a "batch-wide potential failure trend"; if only a single device is abnormal, it is determined to be an "individual failure risk," and comparison reports are generated and fed back to the testing end.

[0106] Before step 402, the method further includes: Step 404: After each test cycle ends and after each abnormal power outage test and power-on, read the pre-written verification file (e.g., 1GB in size) and calculate the hash value, then compare it with the hash value before writing.

[0107] Based on the original architecture, a new "verification file hash comparison" step (corresponding to step 404 mentioned by the user) is added, placed before step 402 (SMART data acquisition), forming a logical closed loop of "verification first → data acquisition → in-depth analysis" to ensure that data integrity is verified first.

[0108] In the file hash comparison verification (corresponding to the newly added step 404), the verification is automatically triggered at two key nodes as required: After each test cycle ends (i.e., the cycle is completed in 48 hours, and the hard drive is at a normal temperature of 25°C), the verification is started before performing step 401 bad block detection. After each step 304 abnormal power failure test (random power failure at the temperature switching point) is completed and the power is restored, the verification is started within 30 seconds after the hard drive is successfully recognized by the system. At this time, no other load or detection operations are performed to avoid data overwriting affecting the verification results.

[0109] Verification operation process: Verify file reading: Use a script to call a file reading tool (such as the Linux dd command or the Windows Copy command) to read the 1GB verification file written in the preprocessing stage in a "sector-level read" manner. During the reading process, record whether there is a reading delay (sector reading exceeding 100ms) or reading failure (returning an I / O error). If a reading exception occurs, mark the LBA address of the exception sector, and then analyze the cause by correlating it with the remapped sector information in the SMART data.

[0110] Hash value calculation and comparison: Using the same hash algorithm as the initial record (such as MD5 or SHA256), recalculate the hash value of the read 1GB verification file; automatically compare the new hash value with the initial hash value recorded in step 1—if they match completely, it is determined that "data integrity is normal", and a verification pass log is generated, which includes the execution time, trigger node (loop end / power failure recovery), and hash value details; if they do not match, it is determined that "data is corrupted", and the file segment with the hash mismatch is immediately located (accurate to a 1MB size range through segmented hash comparison), the LBA address range corresponding to the segment is recorded, and subsequent testing of the hard drive is suspended until the cause of the corruption is manually confirmed before deciding whether to continue.

[0111] Anomaly Handling and Data Association: If a read failure prevents the complete calculation of the hash value, the system combines bad block detection results (if verified after the loop ends) or SMART data before power loss (if verified after power loss recovery) to determine whether the failure is due to bad blocks, NAND media errors, or firmware read logic anomalies. For example, if the LBA address of the failed read overlaps with the address of a remapped sector, it can be preliminarily determined that "the remapped sector caused file fragment loss." All verification results (pass / fail, anomaly details) are automatically synchronized to the test database and stored in association with the corresponding hard drive number, test loop count, and power loss test parameters (such as power loss duration and switching point temperature), providing "data integrity labels" (normal / damaged) for subsequent machine learning analysis.

[0112] Figure 5 yes Figure 1 The flowchart for step 6 is as follows: Figure 5As shown, step 6 includes: Step 601, calculating the health score of the solid-state drive based on test data; Step 602, analyzing the correlation between SMART parameters and voltage and temperature stress, and screening out key parameters that have a significant impact on health; Step 603, predicting failure risk based on linear regression algorithm, providing the probability and time window of failure occurrence, and performing visual analysis on abnormal situations, with visual analysis intuitively displaying the health status of the solid-state drive; Step 604, generating a comprehensive evaluation report from the visual analysis, the comprehensive evaluation report including at least the health score, key parameter analysis, failure prediction, and visual analysis results.

[0113] Based on the original "report generation + solution optimization + baseline update", a new module "health score calculation - parameter correlation analysis - fault prediction - visualization report" (corresponding to steps 601-604) has been added, forming a complete report chain of "data integration → in-depth analysis → visualization output → decision support".

[0114] In the full-cycle health score calculation (corresponding to step 601), based on the full-cycle test data (SMART parameters, verification results, and fault warning records covering all cycles), the final health score is calculated according to the requirements of step 601. This step differs from the aforementioned "real-time health score"; it is the "final comprehensive score" output in the report. Scoring Dimensions and Weights Integration: The four core dimensions are integrated into a weighted system: “SMART parameter degradation (weight 0.4, including the number of remapped sectors, PE cycle count, etc.), data integrity (weight 0.3, including the pass rate of the check file hash and the number of read errors), environmental stress tolerance (weight 0.2, including the number of stable operations under extreme temperatures / voltages), and fault warning records (weight 0.1, including the number of times the historical fault probability exceeded the threshold). The scores for each dimension are taken from the real-time monitoring data in step 4 (e.g., SMART parameter degradation score = 100 - (number of remapped sectors / manufacturer threshold) × 100).

[0115] Final score calculation and grading: The final health score is calculated as "Final Health Score = Σ (Dimension Score × Corresponding Weight)", with a maximum score of 100 points. It is divided into five grades: 90-100 points (Excellent, no risk), 80-89 points (Good, only slight parameter fluctuations), 70-79 points (Pass, requires regular monitoring), 60-69 points (Risk, it is recommended to shorten the testing cycle), and <60 points (High risk, mass production prohibited). For example, if a hard drive has a SMART parameter degradation score of 85 points, a data integrity score of 90 points, an environmental stress tolerance score of 80 points, and a fault warning record score of 100 points, the final score is 85×0.4+90×0.3+80×0.2+100×0.1=85 points, which is judged as "Good".

[0116] SMART parameter and stress correlation analysis (corresponding to step 602): Analysis method and data source: The Pearson correlation coefficient method was used to analyze the linear correlation between each SMART parameter (number of remapping sectors, number of PE cycles, write amplification, etc.) and voltage stress (12V / 5V range in step 2 grouping) and temperature stress (-10℃ / 85℃ / 25℃ conditions in step 3). The data were taken from the full-cycle time series data collected in step 4 (each parameter contains at least 20 cycles of monitoring values). The closer the absolute value of the correlation coefficient is to 1, the stronger the correlation (>0.6 is strong correlation, 0.3-0.6 is moderate correlation, <0.3 is weak correlation).

[0117] Key parameter selection criteria and results: Parameters that are "strongly correlated" and "significantly affect the health score" are selected as key parameters. For example, "the correlation coefficient between the number of remapping sectors and the low-voltage strategy (Group B 10.8V-11.3V) is 0.72, and the correlation coefficient with the high-temperature segment (85℃) is 0.68, both of which are strongly correlated. Moreover, for every 10 additional parameters, the health score decreases by 8 points, so it is listed as a core key parameter; the correlation coefficient between amplification and the high-voltage strategy (Group C 12.7V-13.2V) is 0.65. For every 0.2 increase, the health score decreases by 5 points, so it is listed as an important key parameter." Finally, a "Key Parameter - Stress Correlation List" is output to clarify the sensitive stress type of each parameter (such as the number of remapping sectors being sensitive to "low voltage + high temperature").

[0118] Linear regression for fault risk prediction and visualization (corresponding to step 603): Linear regression model construction: The "degradation rate" (daily new / incremental value) of the key parameters selected above (such as the number of remapped sectors and write amplification) is used as the input feature, and "failure occurrence" (defined as health score < 60 or data corruption) is used as the output label. The linear regression model is trained using historical test data (full lifecycle data of at least 50 hard drives of the same model). The model outputs two core results: "probability of failure occurrence" (0-1, calculated based on the current parameter degradation rate) and "failure time window" (such as "failure may occur within the next 100-120 test hours", estimated based on the time it takes for the parameter to degrade to the threshold).

[0119] Visual Analysis Design: The health status is visually displayed through three types of charts (text descriptions of chart content, no tables): Health Score Trend Chart: The horizontal axis represents the number of test cycles (0-25 times), and the vertical axis represents the health score (0-100 points). The score value and grade lines for each cycle are marked (e.g., 70 points for passing, 60 points for risk), clearly showing the score decline trend (e.g., a hard drive starts at 100 points, drops to 90 points in the 10th cycle, and to 82 points in the 20th cycle, showing a slow overall decline); Parameter-Stress Correlation Chart: The horizontal axis uses "voltage level" (Group A / Group B / Group C), and the "temperature condition" uses different curves (-10℃ / 85℃ / 25℃). The vertical axis represents key parameter values ​​(e.g., number of remapped sectors), showing the changes in parameters under different stress combinations (e.g., under Group B +85℃ conditions, the parameter value is 3 times that of Group A +25℃). (Time axis of failure probability: The horizontal axis represents the estimated usage time (hours), and the vertical axis represents the failure probability (0-1). The time points when the probability exceeds 0.3 (medium risk) and 0.6 (high risk) are marked (e.g., if a hard drive has a probability of 0.3 at 800 hours and 0.6 at 1000 hours, the time window is marked as "entering the risk range at 800-1000 hours").

[0120] In the comprehensive evaluation report generation (corresponding to step 604), according to the requirements of step 604, the results of steps 601-603 are integrated to generate a structured comprehensive evaluation report. The core content includes four parts: Health score summary: List the final health scores and grade distribution of all tested hard drives (text description, such as "A total of 50 hard drives were tested, 15 were excellent (30%), 20 were good (40%), 10 were acceptable (20%), 3 were at risk (6%), and 2 were high risk (4%)"), and mark the typical characteristics of each grade of hard drive (such as excellent hard drives all using Group A voltage and having no verification failure records). Key parameter analysis: Present the list of key parameters selected above, explain the sensitivity stress of each parameter, the magnitude of its impact on the health score, and optimization suggestions (such as "The number of remapped sectors is a core key parameter, sensitive to low voltage + high temperature, it is recommended to avoid long-term superposition of Group B voltage and operating conditions above 85℃ during mass production"). Fault prediction results: Output the failure probability and time window by hard drive group (e.g., "Group A hard drives have an average failure probability of 0.15, and the estimated failure time window is 1200-1500 hours; Group B has an average failure probability of 0.4, and the estimated time window is 800-1000 hours"). High-risk hard drives with a probability exceeding 0.6 are individually labeled with their SN number and recommendations (e.g., "SN12345 has a failure probability of 0.72, and it is recommended to replace it immediately to avoid test interruption").

[0121] Visualized analysis results: The core information of the three types of charts is described in text to ensure clear understanding even without charts (e.g., "The health score trend chart shows that the slope of the A group hard drive score decline is only 0.5 points / cycle, which is much lower than the 1.2 points / cycle of the B group; the parameter-stress correlation chart shows that under the 85℃ operating condition, the write amplification rate of all groups is 0.3-0.5 higher than that at 25℃; the failure probability time axis shows that the C group hard drive enters the high-risk range as early as 600 hours").

[0122] This embodiment introduces innovative modules such as dynamic load simulation, rapid transient voltage testing, data persistence verification, and multi-dimensional fault prediction to construct a multi-stress coupled test environment involving temperature, voltage, load, timing, and transient interference. This more realistically simulates the actual working conditions of solid-state drives (SSDs) in harsh environments such as data centers, industrial control systems, and automotive electronics, overcoming the limitations of existing technologies that only focus on single-indicator testing. Through proactive fault injection and prediction mechanisms, combined with AI analysis of SMART data, effective prediction and early response to SSD faults are achieved, breaking through the passive detection mode of existing technologies that can only diagnose SSDs after a failure occurs. Through a deep data integrity measurement mechanism, introducing data hash verification and data validation tools, "silent data errors" that Diskinfo cannot detect can be discovered, establishing a black-box evaluation system for data reliability and effectively improving the data integrity level of SSDs. Through multi-disk parallel testing and mutual interference mechanisms, system-level compatibility issues when multiple SSDs operate under high load simultaneously are examined, providing reliability support for multi-device collaborative work in real-world application scenarios. By adopting the concept of full lifecycle management, and through the full monitoring and data collection and analysis of solid-state drives (SSDs), a comprehensive assessment of SSD performance and reliability is achieved, providing users with a more comprehensive and accurate SSD reliability assessment solution.

[0123] A second embodiment is proposed based on the first embodiment. In the second embodiment, step S1 is preprocessing and initial information recording. The host computer connects to all the solid-state drives under test via interfaces such as SATA / USB / NVMe. First, the system sends a secure erase command to each solid-state drive to perform a full erase to eliminate the influence of old data and ensure a consistent starting point for testing. Then, the smartctl command-line tool is used to read and record the initial SMART attribute information of each solid-state drive, including but not limited to: Media Wearout Indicator, Reallocated Sector Count, CRC Error Count, and Unsafe Shutdown Count. Simultaneously, the system generates a checksum file of a specific size (e.g., 1GB) and calculates its hash value H0 using a hash algorithm (such as SHA-256). This checksum file is written to the predetermined logical block address (LBA) area of ​​each solid-state drive, and the hash value H0 is stored in the system database along with the initial SMART information. Step S2: Test Grouping and Voltage Strategy Configuration. The pre-processed SSDs are divided into several test groups. Each group contains multiple SSDs to examine mutual interference when multiple drives are operating in parallel. In this example, 30 SSDs are divided into 5 groups of 6. Different voltage strategies are applied to each group of SSDs using a programmable DC power supply to simulate different power environments. The configuration strategy is as follows: Group 1: Rated voltage, such as 3.3V.

[0124] Group 2: Positive voltage bias, such as +5% (3.465V).

[0125] Group 3: Negative voltage bias, such as -5% (3.135V).

[0126] Group 4: Negative voltage bias, such as -8% (3.036V).

[0127] Group 5: Dynamic voltage fluctuations, varying between 3.0V and 3.6V with a specific frequency and waveform (such as a square wave).

[0128] Step S3: Perform the comprehensive stress test cycle.

[0129] The core of this step is to construct a test environment that couples multiple stresses, including temperature, voltage, load, and power failure. A complete test loop consists of the following sub-steps, with the following order and logical relationship: Temperature cycling and load application. The entire test platform is placed in a temperature chamber. The temperature is controlled to cycle between -10°C and 85°C, with a heating / cooling rate of, for example, 2°C / minute. The system is held at both extreme temperatures (85°C and -10°C) for a period of time (e.g., 10 hours). Throughout the temperature cycling and holding process, dynamically varying I / O loads are continuously applied to all SSDs under test via a load generator. Load modes include, but are not limited to: Sequential Write / Read Random Write / Read Mixed read / write (70% read / 30% write) Sustained 100% Random Write is used to test the performance and power stability of garbage collection (GC) algorithms.

[0130] Transient voltage interference injection. During load testing (S310), especially when the temperature reaches preset sensitive points (e.g., 0°C, 30°C, 60°C, 75°C), transient voltage interference is injected into the solid-state drives (SSDs) of a designated test group via a programmable DC power supply. For example, the voltage is dropped from 3.3V to 2.8V within 1 microsecond and maintained for 100 microseconds before recovering. The SSD's response during this process is recorded, such as whether a link disconnection, I / O error, or controller reset occurs.

[0131] Random power-down testing and data persistence verification. At specific temperature cycling points (e.g., before heating from -10°C, at a constant temperature of 25°C, and at a constant temperature of 70°C), a random power-down test is performed on each SSD. The power to a specified SSD is randomly cut off using a high-speed power switch, remaining in a power-off state for a period of time (e.g., 30 seconds), and then powered on again. After power-on, the system automatically performs data integrity verification: it reads the verification file written in step S100, calculates its new hash value H1, and compares it with the initial hash value H0. If H1 ≠ H0, a "silent data error" is recorded.

[0132] System-level mutual interference monitoring. During multi-disk parallel load testing (S310), the main control computer monitors indicators reflecting system-level compatibility in real time, such as: Fluctuations in the total throughput of each solid-state drive group.

[0133] I / O latency and its standard deviation for each solid-state drive.

[0134] Total current ripple of the programmable power supply output. These metrics are used to evaluate the mutual interference effects caused by multiple solid-state drives competing for bus bandwidth or power resources when working together under high load.

[0135] Step S4: Data Acquisition and Intelligent Analysis. After each test cycle, the system automatically executes: Data Acquisition: Read and record the current SMART attributes, data integrity verification results, and system-level performance metrics of all SSDs under test.

[0136] Intelligent Analysis and Health Scoring: Feature extraction: Extract key features from the collected data, such as "number of bad blocks", "CRC error count", "number of data verification failures after power failure", and "average write latency growth rate".

[0137] Model Prediction: The above features are input into a pre-trained failure prediction model. This model is trained using machine learning algorithms (such as Gradient Boosting Decision Tree - GBDT) based on historical large-scale SSD testing data up to failure. The model outputs a Health Index (HI) and a Failure Risk Probability (FRP).

[0138] Health Score Calculation: The health score is calculated using a comprehensive formula, for example: Score = max(0,100 - (W1×∆BadBlock+W2 × ∆CRCError+...+Wn × FRP)), where ∆BadBlock and others are the normalized deterioration values ​​of the feature parameters, W1...Wn are the weights obtained through model analysis, and FRP is the probability of failure risk. A lower score indicates a worse health status.

[0139] Step S5: Iteration and Report Generation. Repeat steps S300-S400 until the preset number of test cycles is reached or a large number of SSD failures occur. Finally, based on the data collected and analysis results from all cycles, the system generates a comprehensive test report. The report includes: Health score curves for each solid-state drive.

[0140] Correlation analysis of key SMART parameters with stress conditions (temperature, voltage).

[0141] Failure prediction results include estimated remaining lifespan.

[0142] Analysis conclusions regarding system-level mutual interference.

[0143] The overall results of data integrity verification include statistics on the occurrence rate of "silent data errors".

[0144] The beneficial effects of this embodiment are as follows: Through the above process, the present invention realizes the simulation of a real environment with multi-stress coupling, transforms passive detection into active AI prediction, and systematically evaluates the power adaptability, data integrity and system-level compatibility of SSDs through multi-disk parallel testing and meticulous data verification, thereby achieving comprehensive verification of the reliability of SSDs throughout their entire life cycle.

[0145] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0146] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0147] The above are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A data integrity comprehensive verification method of a solid state disk, characterized in that, The method comprises: Preprocessing the solid state disk, wherein the preprocessing mode at least includes: issuing a full disk erase instruction to the solid state disk, recording initial SMART information, and writing a verification file; Dividing the solid state disk into multiple groups according to a voltage strategy; Performing a comprehensive stress test cycle on the grouped solid state disks, wherein the comprehensive stress test cycle at least includes temperature cycling, load mode switching, transient voltage testing, and data integrity verification; Collecting SMART data of the solid state disk and intelligently analyzing the SMART data; Judging whether the test is completed, if the test times do not reach a preset number, returning to the previous step to continue the next round of test, if the preset number is reached, performing the next step; Generating a test report according to the SMART data and the intelligent analysis result.

2. The data integrity comprehensive verification method of a solid state disk according to claim 1, characterized in that, The step of dividing the solid state disk into multiple groups according to a voltage strategy comprises: Dividing the solid state disk into multiple groups according to a voltage strategy, each group containing multiple solid state disks and setting a "mutual interference" detection mechanism for the multiple solid state disks; Setting voltage test points for each group of solid state disks, wherein the voltage test points at least include rated voltage, voltage pull bias, and dynamic voltage fluctuation.

3. The data integrity comprehensive verification method of a solid state disk according to claim 1, characterized in that, The step of performing a comprehensive stress test cycle on the grouped solid state disks, wherein the comprehensive stress test cycle at least includes temperature cycling, load mode switching, transient voltage testing, and data integrity verification, comprises: Setting a temperature range of-10℃~85℃ in a temperature box, and performing temperature cycling on the multiple groups of solid state disks according to a preset curve, each extreme temperature being maintained for 10 hours; During the high-temperature and low-temperature sections of the temperature cycling, performing 100% read-write load testing on the multiple groups of solid state disks, and verifying firmware algorithm and power stability; During the load testing, applying voltage drop or spike to the power supply of the solid state disk through a programmable direct current power supply to simulate power glitches caused by sudden loading / unloading of CPU and GPU high-power devices.

4. The data integrity comprehensive verification method of a solid state disk according to claim 3, characterized in that, Before the step of performing 100% read-write load testing on the multiple groups of solid state disks during the high-temperature and low-temperature sections of the temperature cycling, and verifying firmware algorithm and power stability, the method further comprises: At a specific temperature switching point, performing at least 3 random abnormal power-off tests on each solid state disk by randomly cutting off the power supply of the solid state disk through a high-speed switch.

5. The method of claim 1, wherein, The step of collecting SMART data of the solid state disk and intelligently analyzing the SMART data comprises: After each cycle, automatically running a disk tool to check bad blocks, and using a data verification tool to compare the hash values of the written data and the read data; Using a machine learning algorithm to perform correlation analysis on multi-dimensional SMART data, identify potential failure modes, and quantify the degree of reliability degradation.

6. The data integrity comprehensive verification method of a solid state disk according to claim 5, wherein, After the step of using a machine learning algorithm to perform correlation analysis on multi-dimensional SMART data, identify potential failure modes, and quantify the degree of reliability degradation, the method further comprises: When the same type of SMART warning occurs in a group of solid state disks for three consecutive test cycles, triggering a deep diagnosis protocol, and calling underlying logs to analyze the NAND write amplification rate and FTL mapping table changes; If the growth rate of the remapping sector number exceeds 1.5 times the threshold value, it is determined that the voltage strategy has a reliability risk; Finally, all groups of degradation curves and failure correlation matrices are summarized into a test report to provide a decision basis for product mass production, and high-risk voltage strategy combinations are marked, and the coupling relationship between environmental stress and electrical response is combined to optimize the subsequent test scheme.

7. The data integrity comprehensive verification method of a solid state disk according to claim 6, wherein, After the step of finally summarizing all groups of degradation curves and failure correlation matrices into a test report to provide a decision basis for product mass production, and marking high-risk voltage strategy combinations, and combining the coupling relationship between environmental stress and electrical response to optimize the subsequent test scheme, the method further comprises: A weighted fusion model is used to perform health scoring on the test report, and the degradation degree of key SMART parameters and historical failure data are combined to output a failure probability through a logistic regression algorithm; When the failure prediction probability of a group of solid state disks exceeds the set threshold value for two consecutive test cycles, the system automatically marks the group of solid state disks as high-risk equipment and starts a pre-failure warning mechanism, suggesting early replacement, and at the same time, the system synchronously uploads the FTL log, bad block distribution graph and voltage fluctuation record of the high-risk equipment to the cloud analysis platform, and combines the historical data of the batch of products for horizontal comparison to further verify the consistency of the failure trend; If a group of solid state disks exhibits similar degradation characteristics at the same test stage, the model will automatically update the reliability baseline in the degradation curve and adjust the screening criteria at the mass production stage to ensure stable and reliable product quality.

8. The data integrity comprehensive verification method of a solid state disk according to claim 7, wherein, Before the step of using a machine learning algorithm to perform correlation analysis on multi-dimensional SMART data, identifying potential failure modes and quantifying the reliability degradation degree, the method further comprises: After each test cycle ends, and after each abnormal power failure test and re-powering, the pre-written verification file is read and the hash value is calculated, which is compared with the hash value before writing. 9.The method of claim 1, wherein, The step of generating a test report according to the SMART data and the intelligent analysis result comprises: Calculating the health score of the solid state disk based on the test data; Analyzing the correlation between the SMART parameters and the voltage and temperature stress, and selecting key parameters that have a greater impact on health; Based on the linear regression algorithm, the failure risk is predicted, the probability and time window of failure occurrence are given, and the abnormal situation is visually analyzed, and the visual analysis intuitively displays the health status of the solid state disk; A comprehensive evaluation report is generated by visual analysis, and the comprehensive evaluation report at least includes health score, key parameter analysis, failure prediction and visual analysis result. A comprehensive evaluation report is generated by visual analysis, and the comprehensive evaluation report at least includes health score, key parameter analysis, failure prediction and visual analysis result.

Images