Chip aging test board and health state early warning method and system thereof

Abstract

This invention relates to the field of chip testing technology, and particularly to a chip aging test board and its health status early warning method and system. The board includes: an aging test board comprising a substrate and multiple test sockets mounted on the substrate; a protective plate fixed to the back of the substrate, the protective plate having through holes corresponding to the positions of each test socket; a support plate comprising a base plate and multiple monitoring support columns disposed on the base plate; wherein each monitoring support column integrates a uniaxial pressure sensor for collecting forces perpendicular to the support direction; and a data acquisition unit electrically connected to all uniaxial pressure sensors for synchronously acquiring pressure signals generated by each test socket under downward pressure during each loading and unloading process. This invention, through the integration of sensors and innovative signal processing algorithms, transforms each loading and unloading action into a detection of the test socket connection system, enabling real-time, online, and non-invasive sensing of the mechanical health status of the test sockets.

Description

Technical Field

[0001] This invention relates to the field of chip testing technology, and in particular to a chip aging test board and its health status early warning method and system. Background Technology

[0002] In chip aging test systems, high-density pin chips need to be precisely mounted in test sockets, which are secured to the aging test board with nuts to establish an electrical connection. To meet the demands of mass production, automated loading and unloading equipment (Loader) uses a robotic arm to frequently press and release the test sockets during the testing process to load and unload chips. During this process, the periodic downward pressure applied by the robotic arm is transmitted through the test socket to the fixing nut between it and the aging test board. Because the aging test board is primarily a multilayer printed circuit board with limited structural rigidity, long-term exposure to such dynamic impact loads can cause the nut preload to gradually decrease, resulting in "nut loosening." Loosening directly increases the electrical contact resistance between the test socket and the aging test board, degrades signal integrity, and ultimately leads to misinterpretations of test results. Furthermore, the elastic contacts inside the test socket (such as probes and springs) can also fatigue under long-term cyclic stress, leading to a decrease in contact force.

[0003] Currently, maintenance strategies for this issue mainly rely on periodic manual inspections (such as using a torque wrench to check nut torque) or post-inspection after systematic deviations in test results occur. Periodic inspections carry the risk of insufficient or excessive maintenance and cannot detect sudden deterioration occurring between inspections. Post-inspection, on the other hand, wastes testing resources and carries potential batch quality risks. Furthermore, the existing aging test board design, such as the use of a single back support plate, primarily aims to enhance static stiffness and lacks the ability to sense local dynamic mechanical states. Therefore, the industry lacks an effective technical means to sense the mechanical health status of the test socket connection system in real-time, online, and non-intrusively during testing and to provide early warnings of performance degradation. Summary of the Invention

[0004] The main objective of this invention is to provide a chip aging test board and its health status early warning method and system, aiming to solve the technical problems mentioned in the background art.

[0005] This invention proposes a chip aging test board, comprising:

[0006] An aging test board includes a substrate and a plurality of test sockets mounted on the substrate. A protective plate is fixed to the back of the substrate, and the protective plate is provided with through holes corresponding to the positions of each of the test sockets.

[0007] A support plate includes a base plate and a plurality of monitoring support columns disposed on the base plate. The positions of the monitoring support columns correspond to the through holes, such that when the aging test plate is installed on the support plate, the top of each monitoring support column passes through the corresponding through hole and contacts the back of the test socket.

[0008] Each of the monitoring support columns integrates a uniaxial pressure sensor for collecting forces perpendicular to the support direction; the bottom of each monitoring support column is equipped with an elastic height adjustment element.

[0009] The data acquisition unit is electrically connected to all the uniaxial pressure sensors and is used to synchronously acquire the pressure signals generated by each of the test sockets when it is subjected to downward pressure during each loading and unloading process.

[0010] This invention also discloses a health status early warning method for a chip aging test board, applied to a chip aging test board, comprising:

[0011] The current pressure signal of the target test socket during real-time loading and unloading is collected. The current pressure signal is time-aligned with the pre-stored standard health pressure waveform template, and the morphological difference signal after alignment is calculated.

[0012] The energy integral of the morphological difference signal is obtained within three time windows: the early stage of the pressure rise, the pressure peak holding period, and the pressure fall period, in order to calculate the stiffness trend factor that reflects the stiffness change trend of the target test socket.

[0013] Calculate the difference between the integral values ​​of the morphological difference signal at the pressure rise edge and pressure fall edge, and calculate the hysteresis anomaly factor reflecting the mechanical hysteresis characteristics.

[0014] The incremental health index of each historical test period is obtained. The relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods is combined to calculate the incremental health index of the current test period. The standardized health index is calculated based on the cumulative sum of the incremental health index.

[0015] The current health status of the target test socket is determined based on the health interval threshold of the standardized health index, and the remaining service life of the target test socket is predicted based on the time-series change trend of the standardized health index. A graded early warning decision is generated by combining the status confidence.

[0016] The present invention is further configured such that the steps of obtaining the incremental health index of each historical test period, combining the relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculating the incremental health index of the current test period, and calculating the standardized health index based on the cumulative sum of the incremental health index include:

[0017] Obtain the stiffness trend factor and hysteresis anomaly factor corresponding to the current test cycle, as well as the stiffness trend factor and hysteresis anomaly factor corresponding to the previous test cycle.

[0018] The incremental health index for the current testing period is calculated by weighting the relative change rate of the stiffness trend factor in the current testing period and the stiffness trend factor in the previous testing period, as well as the relative change rate of the hysteresis anomaly factor in the current testing period and the hysteresis anomaly factor in the previous testing period.

[0019] The incremental health index of each historical test period is obtained, and the cumulative health index is calculated by combining the incremental health index of the current test period with the index decay weighted summation method.

[0020] The cumulative health index is normalized to obtain a standardized health index that characterizes the health of the target test socket connection structure.

[0021] The present invention is further configured such that the step of calculating the incremental health index for the current test period by weighting the relative change rate of the stiffness trend factor of the current test period and the stiffness trend factor of the previous test period, and the relative change rate of the hysteresis anomaly factor of the current test period and the hysteresis anomaly factor of the previous test period, includes:

[0022] Obtain the current test task, obtain the chip's package type and power consumption level based on the current test task, determine the load type identifier, and adjust the weighting coefficients of the weighted calculation based on the load type identifier;

[0023] Obtain the historical maintenance records of the target test socket, obtain the main failure modes of the target test socket based on the historical maintenance records, and adjust the weighting coefficients of the weighted calculation based on the main failure modes;

[0024] The adjusted weighting coefficients will be reused in the weighted calculation of the incremental health index for the current and subsequent testing periods.

[0025] The present invention is further configured such that the steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the status confidence level include:

[0026] Preset dynamic health range thresholds, including a first lower health threshold, a second lower health threshold, and a fault threshold;

[0027] Obtain historical health index statistics of similar test sockets to initialize the health interval threshold, and adaptively update the health interval threshold based on the minimum value of the standardized health index of the most recent multiple health test cycles during online operation;

[0028] The standardized health index is compared with the health interval threshold, and the current health status is determined according to the preset health interval threshold crossing rules.

[0029] The present invention is further configured such that the steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the status confidence level include:

[0030] Extract time-series data of the standardized health index over the most recent testing periods;

[0031] The slope of the standardized health index is obtained by performing a linear fit on the time series data.

[0032] If the slope of change is less than zero, the number of remaining test cycles to reach the fault threshold is predicted by dividing the difference between the standardized health index and the second lower health threshold by the absolute value of the slope of change.

[0033] The present invention is further configured such that the steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the status confidence level include:

[0034] The confidence level of the current health status is calculated based on the frequency of the standardized health index crossing the health interval threshold of the most recent preset number of times.

[0035] A comprehensive risk score is calculated by weighting the standardized health index, the state confidence level, and the normalized number of remaining test cycles.

[0036] Establish a risk accumulator to accumulate the portion of the comprehensive risk score that exceeds a preset basic threshold.

[0037] Multiple warning level thresholds are set based on the value of the risk accumulator. When the value of the risk accumulator exceeds one of the warning level thresholds, the corresponding warning action is triggered.

[0038] If the overall risk score for multiple consecutive preset test cycles is lower than the preset risk accumulation threshold after the warning action is triggered, the risk accumulator is controlled to decay at a preset rate, and the warning level is adaptively downgraded accordingly.

[0039] Preferably, the step of obtaining the standard health pressure waveform template includes:

[0040] When the target test socket is in a healthy state, its pressure signal is collected during multiple standard loading and unloading processes.

[0041] Based on the moment when the pressure signal first exceeds a preset quiet threshold, all collected pressure signals are time-aligned.

[0042] The average of the aligned pressure signals is calculated to obtain a quasi-healthy pressure waveform template unique to the target test socket.

[0043] Preferably, the present invention further includes preprocessing the acquired current pressure signal, including:

[0044] The original pressure signal is filtered to suppress high-frequency noise, resulting in a filtered signal.

[0045] The peak value of the filtered signal is detected. If the peak value does not fall within the preset reasonable force value range, the acquisition is determined to be invalid and discarded, and an effective pressure signal is obtained.

[0046] Based on the difference between the effective pressure signal and the standard healthy pressure waveform template in the undeformed section, an amplitude correction coefficient is calculated and applied to correct the effective pressure signal.

[0047] Use the corrected effective pressure signal as the current pressure signal.

[0048] This invention also discloses a health status early warning system for a chip aging test board, comprising:

[0049] The signal sensing module is used to collect the current pressure signal of the target test socket during the real-time loading and unloading process, align the current pressure signal with the pre-stored standard health pressure waveform template, and calculate the morphological difference signal after alignment.

[0050] The signal processing module is used to acquire the energy integral of the morphological difference signal in three time windows: the early stage of the pressure rise, the pressure peak holding period, and the pressure fall period, so as to calculate the stiffness trend factor that reflects the stiffness change trend of the target test socket.

[0051] The feature extraction module is used to calculate the difference between the integral values ​​of the morphological difference signal at the pressure rising edge and the pressure falling edge, and to calculate the hysteresis anomaly factor reflecting the mechanical hysteresis characteristics.

[0052] The health assessment module is used to obtain the incremental health index of each historical test period, combine the relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculate the incremental health index of the current test period, and calculate the standardized health index based on the cumulative sum of the incremental health index.

[0053] The early warning decision module is used to determine the current health status of the target test socket based on the health interval threshold of the standardized health index, predict the remaining service life of the target test socket based on the time-series change trend of the standardized health index, and generate a graded early warning decision based on the status confidence level.

[0054] The beneficial effects of this invention are as follows: By integrating sensors and innovative signal processing algorithms, this invention transforms each loading and unloading action into a detection of the test socket connection system, enabling real-time, online, and non-invasive perception of its mechanical health status. By exploring the microscopic evolution of pressure waveform morphology and establishing a mapping relationship with specific physical degradation mechanisms such as nut loosening and contact fatigue, early diagnosis and trend prediction of latent degradation processes are achieved, transforming the maintenance mode from passive response to proactive prediction and improving test reliability. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the assembly of an aging test plate and a support plate according to an embodiment of this application.

[0056] Figure 2 This is a schematic diagram of the back structure of an aging test plate according to an embodiment of this application.

[0057] Figure 3 This is a schematic diagram of a support plate structure according to an embodiment of this application.

[0058] Figure 4 This is a cross-sectional view of the aging test plate and support plate assembly structure according to an embodiment of this application.

[0059] Figure 5 This is a schematic diagram of the internal structure of a monitoring support column according to an embodiment of this application.

[0060] Figure 6 This is a flowchart of a method according to an embodiment of this application.

[0061] Figure 7 This is a system flowchart of an embodiment of this application.

[0062] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0063] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0064] In the current automated aging test process for high-density chips, test sockets are mounted on the aging test board using fasteners such as nuts. The robotic arm of the automated loading and unloading equipment needs to frequently press down on the sockets to load or unload chips. During this process, cyclic mechanical loads can lead to two main progressive mechanical failure modes: 1. The preload of the fixing nut decreases due to vibration and stress relaxation, i.e., "nut loosening"; 2. Fatigue occurs in the precision elastic contacts inside the socket (such as probes and springs), resulting in a decrease in contact force.

[0065] Current technologies cannot detect the degradation of the aforementioned mechanical state in real time during testing. Maintenance strategies primarily rely on two methods: one is preventative manual inspection based on fixed time intervals (e.g., using a torque wrench to check torque), and the other is post-incident troubleshooting after systematic deviations in electrical performance are observed. Both methods have significant drawbacks: the setting of regular inspection cycles lacks scientific basis, potentially leading to insufficient maintenance (failure to detect sudden degradation within the interval) or excessive maintenance (unnecessary intervention on connections still in good condition, wasting time and potentially accelerating wear); post-incident troubleshooting implies that the fault has already occurred, usually rendering the entire batch of chip test data invalid, wasting test time, and even potentially damaging high-value chips due to poor contact, resulting in significant quality and economic losses. For example, aging tests of long-cycle, high-reliability chips, such as automotive-grade, industrial control, and aerospace chips, require high-temperature or high-low temperature aging tests exceeding 1000 hours. These tests are lengthy and involve high-value chips. This system can serve as a comprehensive safeguard, monitoring the mechanical stability of the test fixture in real time to ensure long-cycle... Successful testing on the first attempt avoids wasting time and money due to fixture problems. For example, in testing high-density, high-pin-count chips such as CPUs, GPUs, and FPGAs, the dedicated test sockets are extremely complex and precise, with high unit prices and stringent requirements for contact reliability and signal integrity. This invention can independently and continuously monitor the status of each expensive socket, providing early warnings for any minor contact degradation, thus achieving key protection for core test assets. For example, in fully automated, unmanned test production lines (lights-out factories), predictive maintenance is crucial for ensuring unmanned operation in production lines with high labor costs or requiring 24-hour continuous operation. This system can automatically diagnose and locate sockets in poor health, issue early warnings, and even link with the Production Management System (MES) to generate maintenance work orders.

[0066] Therefore, this application provides an effective technical means to monitor the mechanical health status of the test socket connection structure online, in real time, and non-invasively during continuous chip testing, and to provide early warning and trend prediction of its performance degradation.

[0067] like Figures 1-5As shown, this application provides a chip aging test board, comprising:

[0068] An aging test plate 1 includes a substrate 120 and a plurality of test sockets 150 mounted on the substrate 120. A protective plate 110 is fixed on the back of the substrate 120. The protective plate 110 is provided with through holes corresponding to the positions of each of the test sockets 150.

[0069] The support plate 2 includes a base plate 210 and a plurality of monitoring support columns 230 disposed on the base plate. The positions of the monitoring support columns 230 correspond to the through holes, such that when the aging test plate 1 is installed on the support plate 2, the top end of each monitoring support column 230 passes through the corresponding through hole and contacts the back of the test socket 150.

[0070] Each of the monitoring support columns 230 integrates a single-axis pressure sensor for collecting forces perpendicular to the support direction; the bottom of the monitoring support column 230 is provided with an elastic height adjustment element.

[0071] The data acquisition unit is electrically connected to all the uniaxial pressure sensors and is used to synchronously acquire the pressure signals generated by each of the test sockets 150 when it is subjected to downward pressure during each loading and unloading process.

[0072] It should be noted that, as shown in the attached document Figure 1 As shown, the aging test plate 1 is precisely positioned between the aging test plate 1 and the support plate 2 by engaging with the positioning pin 220 and the positioning hole at a specific position on the substrate 120, thereby fixing the aging test plate 1 in place. After the aging test plate 1 is fixed, the monitoring support column 230 on the support plate 2 passes through the through hole on the protective plate 110 and contacts the reinforcing pad 140 on the back of the test socket 150 installed on the substrate 120, thus providing support for the test socket 150. Figure 4 As shown; an adjusting shim 250 is designed between the monitoring support column 230 and the base plate 210 to eliminate the assembly gap between the monitoring support column 230 and the reinforcing shim 140, ensuring good contact between the two. Figure 5As shown; the adjusting shim 250 is made of rubber and polytetrafluoroethylene elastic material, or a spring-based structural component can also be used; the monitoring support column 230 is internally designed with a sensor 240, including but not limited to a pressure sensor, or a displacement sensor. When the aging test board 1 is loading and unloading chips in the Loader equipment, the pressure generated by the robotic arm pressing the test socket 150 is transmitted to the monitoring support column 230 through the reinforcing shim 140. The sensor 240 monitors and records the pressure value data, or the displacement deformation amount. The value recorded by the sensor 240 is uploaded to the Loader equipment server, which can monitor the changes in parameters in real time and determine whether the test socket nut 130 has become loose. Through a chip aging test board health status early warning method, the service life of the test socket 150 can be evaluated, and a warning can be given for the test socket 150 that has become defective, prompting the user to repair or replace it in time, so as to avoid bringing the defective test socket 150 into the subsequent production testing stage.

[0073] like Figure 6 As shown, the present invention also discloses a health status early warning method for a chip aging test board, applied to a chip aging test board, comprising:

[0074] S1, Collect the current pressure signal of the target test socket during the real-time loading and unloading process, align the current pressure signal with the pre-stored standard health pressure waveform template, and calculate the morphological difference signal after alignment.

[0075] S2, obtain the energy integral of the morphological difference signal in three time windows: the early stage of the pressure rise, the pressure peak holding period, and the pressure fall period, in order to calculate the stiffness trend factor that reflects the stiffness change trend of the target test socket.

[0076] Stiffness trend factor calculation formula:

[0077]

[0078] In the formula, This represents a stiffness trend factor, which is dimensionless. An increase in its value indicates a decrease in the stiffness of the connected structure. This represents the absolute integral value of the morphological difference signal within the early time window of the pressure rise, i.e. ,in[ This represents the early time window of the pressure rise, such as the stage where the pressure rises from the resting threshold (e.g., 10% of the maximum value) to approximately 40% of the peak pressure. This represents the absolute integral value of the morphological difference signal within the time window of the pressure peak holding period, i.e. ,in[ This refers to the main load phase within the pressure peak holding period, where the pressure fluctuates near the peak value (e.g., above 85% of the peak value). This represents the absolute integral value of the morphological difference signal within the time window of the pressure decrease period, i.e. ,in[ This refers to the time window during the pressure decline period, such as the stage when the pressure drops from near the peak to the quiescent threshold.

[0079] S3, calculate the difference in integral values ​​of the morphological difference signal at the pressure rise edge and pressure fall edge, and calculate the hysteresis anomaly factor reflecting the mechanical hysteresis characteristics.

[0080] Formula for calculating hysteresis anomaly factor:

[0081]

[0082] In the formula, This represents the hysteresis anomaly factor, which is dimensionless. An increase in its value indicates abnormal energy dissipation or a change in hysteresis characteristics. This represents the algebraic integral value of the morphological difference signal over the entire pressure rise interval. This represents the algebraic integral value of the morphological difference signal over the entire pressure drop interval. This represents the absolute integral value of the morphological difference signal over a complete push-release cycle. 1. A small positive number added to prevent the denominator from being zero is a preset constant;

[0083] In the calculation process of the method of this invention, the analysis of the pressure pattern difference signal relies on its precise definition and division in several key time intervals. These intervals serve different feature extraction purposes based on their different physical meanings in a complete pressing-releasing action: "Complete pressing-releasing cycle" refers to the entire time span from the start moment when the pressure signal first exceeds the preset quiet threshold to the end moment when the pressure signal is finally released and falls below the quiet threshold again. This interval defines the complete mechanical action process of a single loading and unloading action and is the basis for calculating the total energy benchmark and judging the integrity of the cycle. "Complete pressure rising edge interval" specifically refers to the time period from the start moment of the cycle to the moment when the pressure signal reaches its global peak. This interval completely describes the entire process of the test socket connection structure bearing load from a relaxed state until the load reaches its maximum value. Its time span and signal pattern directly reflect the overall dynamic response characteristics of the system, especially the mechanical behavior of the loading path. "Complete pressure falling edge interval" refers to the time period from the pressure peak moment to the end moment of the cycle. This interval describes the complete process of the load being released from its maximum value to zero. Its signal characteristics are closely related to the unloading elasticity, internal friction, and possible plastic deformation recovery mechanism of the system. It should be noted that the above three "complete intervals" are mainly used to calculate the hysteresis anomaly factor. The core of the hysteresis anomaly factor is to evaluate the area of ​​the force-displacement hysteresis loop formed by the rising and falling edges. This area is a direct measure of the mechanical energy dissipation of the connected structure in one cycle. Therefore, by performing algebraic integration on the complete rising and falling paths, this energy dissipation characteristic can be accurately captured.

[0084] S4. Obtain the incremental health index of each historical test period, combine the relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculate the incremental health index of the current test period, and calculate the standardized health index based on the cumulative sum of the incremental health index.

[0085] S5. Determine the current health status of the target test socket based on the health interval threshold of the standardized health index, predict the remaining service life of the target test socket based on the time-series change trend of the standardized health index, and generate a graded early warning decision based on the status confidence.

[0086] As described in steps S1-S5 above, this invention aims to solve the technical problem in the background art of "the inability to monitor the mechanical health status of the test socket connection structure in real time during testing, and the inability to provide early predictive warnings of its performance degradation." By converting the raw stress signals sensed by the hardware into a quantitative assessment of the health of the connection structure and executable hierarchical warning commands, it achieves a shift from periodic preventive maintenance to condition-based predictive maintenance.

[0087] The method for establishing the standard health pressure waveform template is as follows: when the target test socket is in a healthy state, its pressure signals are collected during multiple standard loading and unloading processes; using the moment when each pressure signal first exceeds a preset quiet threshold as a benchmark, all collected pressure signals are time-aligned; the average of the aligned pressure signals is calculated to obtain a unique quasi-healthy pressure waveform template for the target test socket, thereby establishing an individualized and dynamic evaluation benchmark and forming an effective online judgment basis. By learning the characteristic pressure waveform of each test socket in a healthy state, the evaluation of all subsequent real-time signals is transformed into a comparison with this individualized baseline, eliminating the inherent signal offset caused by individual manufacturing tolerances of the test socket, small installation angle differences, and different chips being tested, allowing the analysis to focus on signal morphology changes purely caused by changes in the health state of the mechanical structure. The time alignment operation (which can use a dynamic time warping algorithm) further eliminates the overall scaling or translation effect of the signal on the time axis caused by small fluctuations in the speed of each operation of the Loader robot. After this step, the calculated morphological difference signal is significant because, after removing time jitter and individual inherent characteristics, it reflects the waveform distortion of the current mechanical connection state relative to its own health baseline.

[0088] In this invention, the calculation of the stiffness trend factor and hysteresis anomaly factor relies on the analysis of the absolute energy integral of the morphological difference signal within three specific time windows: the early pressure rise window, the pressure peak holding period window, and the pressure fall period window. This feature is designed for the main failure mode of "nut loosening." The physical mechanism is as follows: when the nut preload decreases, leading to a reduction in the overall axial stiffness of the connection structure, the system exhibits a "softer" characteristic under the same downforce. The pressure signal rises more slowly, and energy accumulation in the early rise stage increases relatively. Simultaneously, stiffness anomalies cause the energy distribution in the loading, holding, and unloading stages to become more uneven or exhibit specific change patterns. The stiffness trend factor, through a combination of ratio and energy distribution uniformity term, can sensitively capture this systematic drift in the waveform morphology related to stiffness; its increased value indicates a decrease in stiffness. In step S3, the formula design directly targets failure modes that increase energy dissipation due to "contact fatigue" and changes in internal friction. Its physical mechanism originates from the hysteresis effect in mechanics: in a complete load-unload cycle, the area enclosed by the loading and unloading paths represents mechanical energy loss. When contact fatigue, wear, or internal friction increases, the area of ​​this hysteresis loop increases, which is reflected in the signal as an increase in the absolute value of the algebraic difference between the integral of the rising edge and the integral of the falling edge. The hysteresis anomaly factor calculates this difference, and the denominator is normalized using the absolute energy integral of the entire cycle to eliminate the influence of accidental fluctuations in the magnitude of a single operating force, thus obtaining a stable index reflecting the degree of abnormal energy dissipation. The extraction of these two characteristic factors maps abstract waveform difference signals to specific physical parameters directly related to the failure mechanisms described in the background art, providing an interpretable and quantitative basis for diagnosing degradation types and quantifying degradation degrees.

[0089] Then, the method performs quantitative fusion and trend modeling of health status. By using the time series of stiffness trend factor and hysteresis anomaly factor, a comprehensive health index that can reflect the long-term degradation trend is constructed, providing a stable and reliable input for the final threshold judgment and early warning.

[0090] Ultimately, continuous standardized health indices are transformed into discrete maintenance decisions, thereby enabling real-time, online, and non-intrusive sensing of the mechanical health status of the test socket connection system during testing, and providing early warnings of its performance degradation.

[0091] It should be noted that the core of calculating the stiffness trend factor is to indirectly reflect the stiffness characteristics of the connection structure through the energy distribution change of the morphological difference signal at different mechanical stages. The energy integral of the morphological difference signal is a quantification of the cumulative effect of signal intensity over time. The energy integral in the early stage of the pressure rise corresponds to the mechanical response of the structure at the initial stage of the load, the energy integral during the pressure peak holding period corresponds to the energy dissipation of the structure under stable load, and the energy integral during the pressure fall period corresponds to the mechanical recovery characteristics of the structure during the unloading process. When the stiffness of the test socket connection structure decreases, the mechanical response at the initial stage of the load will slow down, resulting in a slower energy integral in the early stage of the rise. The relative increase in energy integral during the peak holding period and the relatively stable energy dissipation during the peak holding period lead to an increase in the ratio between the two. At the same time, the decrease in stiffness will make the energy distribution of the three stages tend to be uniform, and the ratio of the minimum to the maximum value among the three will increase, thus reducing the result of (1 minus this ratio). However, the increase in the ratio of the energy integral during the early rising period to that during the peak holding period usually dominates the entire calculation result, ultimately leading to an increase in the stiffness trend factor. Under healthy conditions, the value of this factor is relatively stable and at a low level. When stiffness degradation phenomena such as loose nuts occur, this factor will show a significant upward trend. Therefore, its change can be used to judge the degree of structural stiffness degradation.

[0092] It should be noted that the calculation of the hysteresis anomaly factor is based on the principle of hysteresis effect in mechanical systems. During loading and unloading, due to factors such as internal friction and material fatigue, the loading and unloading paths cannot completely coincide. The resulting hysteresis loop area corresponds to the mechanical energy dissipation in a single cycle. The absolute value of the difference between the algebraic integrals of the morphological difference signal at the rising and falling edges is the quantification of this hysteresis loop area. The absolute integral of the complete cycle serves as a normalization benchmark, eliminating the influence of force fluctuations in a single operation, allowing the factor to stably reflect the degree of energy dissipation anomaly. When the elastic contact inside the test socket is fatigued, worn, or the contact friction increases, the mechanical energy dissipation increases, the hysteresis loop area expands, and the hysteresis anomaly factor value rises. In a healthy test socket, this factor value usually remains in a low range. As the degree of contact degradation intensifies, the factor gradually increases. Therefore, its change can be used to characterize the degradation state of the contact.

[0093] In one embodiment, the steps of obtaining the incremental health index of each historical test period, combining the relative change rates of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculating the incremental health index of the current test period, and calculating the standardized health index based on the cumulative sum of the incremental health indices include:

[0094] S41, obtain the stiffness trend factor and hysteresis anomaly factor corresponding to the current test cycle, as well as the stiffness trend factor and hysteresis anomaly factor corresponding to the previous test cycle.

[0095] S42. The incremental health index for the current test period is calculated by weighting the relative change rate of the stiffness trend factor in the current test period and the stiffness trend factor in the previous test period, as well as the relative change rate of the hysteresis anomaly factor in the current test period and the hysteresis anomaly factor in the previous test period.

[0096] Incremental health index calculation formula:

[0097] In the formula, Indicates the current test period ( The incremental health index (corresponding to the current testing period) is a scalar that can be positive or negative; a negative value indicates a decline in health. Indicates the first The stiffness trend factor is calculated over each test cycle. Indicates the first The hysteresis anomaly factor calculated over each test cycle Indicates the first The stiffness trend factor is calculated over each test cycle. Indicates the first The hysteresis anomaly factor calculated over each test cycle No. The incremental health index is calculated over each testing period. , and Let represent the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively, such that the sum of the three equals 1, and The specific values ​​are optimized based on historical data or set according to experience. and Small positive numbers are added to prevent the denominator from being zero;

[0098] S43, obtain the incremental health index of each historical test period, and calculate the cumulative health index by combining the incremental health index of the current test period and using the index decay weighted summation method.

[0099] The formula for calculating the cumulative health index is:

[0100] In the formula, Indicates the first The cumulative health index for each testing period is a floating-point number; a downward trend indicates a decline in health status. This indicates the initial health index (which can be preset to 0). Indicates the first Incremental health index over each testing period This represents the decay time constant, a positive preset constant used to control the decay rate of the influence of historical increments. This represents the exponentially decaying weighting factor, used to assign greater weight to recent increments;

[0101] S44, the cumulative health index is normalized to the [0,1] interval through an S-shaped function mapping to obtain a standardized health index characterizing the health of the target test socket connection structure.

[0102] As described in steps S41-S44 above, the present invention systematically synthesizes the time series of characteristic factors reflecting stiffness and hysteresis anomalies into a comprehensive health scale that can stably characterize the long-term degradation process of the connection structure, thereby transforming discrete, instantaneous physical characteristic observations into a continuous, standardized health index with historical memory and trend characterization capabilities.

[0103] Specifically, it provides the stiffness trend factor and hysteresis anomaly factor corresponding to the current test period, as well as the stiffness trend factor and hysteresis anomaly factor corresponding to the previous test period. It places the independent features of each test period in the context of time series, which prepares the data for the transformation from static feature evaluation to dynamic trend tracking. This allows the system to focus not only on the absolute value of the current feature when performing health assessment, but also on its change relative to the most recent historical state. This is a prerequisite for capturing the core characteristic of slow degradation.

[0104] Next, based on the aforementioned feature values, and the relative change rates of the stiffness trend factor and the hysteresis anomaly factor between the current and previous cycles, the incremental health index for the current test cycle is calculated through weighted average. Using "relative change rate" instead of absolute feature values ​​as input reduces sensitivity to sensor absolute calibration accuracy and long-term drift, allowing the system to focus on capturing minute, directional trend changes. Secondly, by introducing historical incremental terms, the model is given a "memory" function. The third weighting coefficient controls the strength of the influence of historical information on the current assessment. This mechanism effectively smooths out random noise or transient interference that may be included in a single measurement, ensuring that the output of the incremental health index better reflects the true health change trend over a period of time, rather than being dominated by fluctuations in a single operation. This significantly improves the stability and robustness of the condition assessment. Finally, the first and second weighting coefficients provide an adjustable fusion of the contributions of two different physical degradation modes (stiffness decrease and energy dissipation anomaly), allowing the model to be optimized according to the emphasis of the actual failure mechanism.

[0105] Subsequently, the incremental health index for each historical test period is obtained, and combined with the incremental health index of the current period, a cumulative health index is calculated through an exponentially decaying weighted summation. The calculation principle is as follows: the initial health index is added to the weighted sum of all incremental health indices from the first period to the current period; where the weight of each historical incremental health index is a factor that decays exponentially with increasing time distance from the current period. This step constructs an index reflecting the cumulative effect of degradation throughout the entire life cycle, giving greater weight to recent health increments, while the impact of long-term increments decays exponentially with time distance. The decay time constant controls the memory length of historical impacts, meaning that the recent operating state has a much higher predictive value for current health than the state several months ago. Through this weighting method, the cumulative health index not only accumulates all historical health change information but also highlights the dominant role of recent trends, enabling the index to respond sensitively to the latest performance degradation while not neglecting the long-term degradation background, thus providing an ideal time series basis for predicting remaining service life.

[0106] In summary, standardized health indices can transform the mechanical health status of test sockets into continuously measurable, traceable, and predictable key performance parameters, thereby solving the problem of delayed and blind maintenance decisions caused by the lack of quantitative and continuous status trend data.

[0107] In one embodiment, the step of calculating the incremental health index for the current test period by weighting the relative change rate of the stiffness trend factor of the current test period with that of the previous test period, and the relative change rate of the hysteresis anomaly factor of the current test period with that of the previous test period, includes:

[0108] S421, Obtain the current test task, obtain the chip's package type and power consumption level according to the current test task, determine the load type identifier, and adjust the weight coefficient of the weighted calculation according to the load type identifier;

[0109] For example, if the load type identifier corresponds to a type of chip (corresponding to high power consumption, large size chip), then the second weight coefficient associated with the hysteresis anomaly factor is reduced; if the load type identifier corresponds to a type of chip (corresponding to low power consumption, small size chip), then the first weight coefficient associated with the stiffness trend factor is reduced.

[0110] S422, Obtain the historical maintenance record of the target test socket, obtain the main failure mode of the target test socket based on the historical maintenance record, and adjust the weight coefficient of the weighted calculation based on the main failure mode;

[0111] For example, if historical maintenance records show that the primary failure mode of the target test socket is loose nuts, the first weighting coefficient is further increased; if the primary failure mode is internal contact fatigue of the socket, the second weighting coefficient is further increased.

[0112] S423, the adjusted weighting coefficients are reused for the weighted calculation of the incremental health index in the current and subsequent testing periods.

[0113] For example, based on the adjusted first and second weighting coefficients, when satisfying Under the constraints, a new set of first weight coefficients, second weight coefficients, and third weight coefficients are determined for the calculation of the incremental health index in the current and subsequent testing periods.

[0114] As described in steps S421-S423 above, this invention introduces an intelligent weight adjustment mechanism based on actual working conditions and historical experience. The weighting coefficients of the weighted calculation are dynamically adjusted according to two dimensions: "current test task load type" and "historical failure modes of the target test socket." This aims to address how to make the health assessment model not only universal but also adaptively optimize its diagnostic focus to more accurately reflect the most likely degradation risks under specific conditions, thereby improving the targeting and accuracy of early warnings.

[0115] Since high-power, large-size chips typically mean that test sockets need to withstand greater downforce and thermal loads, their mechanical connections (such as nuts) experience greater stress. Therefore, the decrease in stiffness caused by "nut loosening" may be a more dominant or predominant failure mode. In this case, by reducing the focus on hysteresis factors and correspondingly increasing the weight of sensitivity to changes in stiffness factors, the standardized health index can be made more focused on monitoring such risks. Conversely, for low-power, small-size chips, the downforce is smaller and the mechanical stress of the nuts is relatively lower, but the minute fatigue of the precision contacts inside the socket may become the primary factor affecting contact reliability. In this case, by reducing the focus on stiffness factors and increasing the sensitivity to hysteresis factors that reflect contact friction and energy dissipation, the system can more effectively capture such degradation. This mechanism enables the method to predict and optimize based on the current physical characteristics of the test object, achieving preliminary adaptive operation.

[0116] Furthermore, even for sockets of the same model, subtle differences in manufacturing batches, installation processes, or usage history can lead to different actual weak points. Some sockets may experience repeated nut loosening, while others may have contacts that wear out more easily. Traditional methods cannot utilize this valuable field experience data. This invention analyzes the historical maintenance data of specific sockets to identify their individual failure tendencies and dynamically strengthens the monitoring weight of corresponding failure characteristics (stiffness or hysteresis) in the health index model accordingly. This allows the system to remember the failure modes of each socket and conduct more targeted analysis based on this, enhancing the ability to predict individual-specific degradation risks.

[0117] In one embodiment, the steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the state confidence level include:

[0118] S51, preset dynamic health interval thresholds, including the first lower health limit threshold. Second health lower limit threshold and fault threshold ,in ;

[0119] For example, the first health lower limit threshold and the second health lower limit threshold correspond to the lower limit threshold of the green health interval and the lower limit threshold of the yellow health interval, respectively. The green health interval and the yellow health interval can be used to represent the healthy interval and the interval where there is a health risk.

[0120] S52, Obtain historical health index statistics for similar test sockets to initialize the health interval threshold, and during online operation, adjust the data based on the most recent... The minimum value of the standardized health index for each health testing cycle is used to slowly and adaptively update the health interval threshold.

[0121] Taking the first health lower limit threshold as an example, the update formula is:

[0122]

[0123] In the formula, Indicates the first The first health lower limit threshold updated after each test cycle Indicates the first The first health lower limit threshold for each test cycle The learning rate is a small preset positive constant ( ∈[0.01,0.1]), used to control the degree of influence of new information on the first health lower limit threshold. Indicates the current test period ( The standardized health index (corresponding to the current testing period) Indicates recent The minimum value of the standardized health index over a period of time. This represents the window size used to calculate the minimum value, and is a preset positive integer value. The learning rate controls the strength of the influence of new observations on the threshold, while the most recent... The minimum value of the standardized health index over a health testing cycle represents the best health level observed recently.

[0124] S53, compare the standardized health index with the health interval threshold, and determine the current health status according to the preset health interval threshold crossing rules.

[0125] As described in steps S51-S53 above, this invention introduces dynamic threshold management and a series of life prediction mechanisms based on time-series trends to transform the continuously changing standardized health index calculated in the preceding steps into clear, hierarchical conclusions of the current health status, quantified remaining risk time, and robust early warning instructions, thereby enabling executable maintenance decisions.

[0126] The initialization of the first health lower limit threshold, the second health lower limit threshold, and the fault threshold relies on the historical health index statistics of similar test sockets. This sets a reasonable initial reference standard for sockets of different models or batches. Its online update mechanism allows the status judgment standard to adapt to the individual characteristics of each socket and the slow performance baseline drift that may be caused by environmental factors, aging, etc. during long-term operation. By slowly bringing the threshold closer to the best performance recently observed, the system avoids two types of errors that may occur when using fixed thresholds: first, misjudging sockets with normal performance but slightly low individual baselines as abnormal due to excessively high threshold settings (false alarms); second, failing to report early degradation due to excessively low threshold settings or failure to track slow performance decline (false alarms). This adaptability ensures the objectivity and accuracy of the status judgment benchmark.

[0127] For example, the health interval threshold crossing rule: when the standardized health index falls below the first health lower limit threshold M times consecutively, the status changes from 'healthy' to 'warning'. Introducing the judgment condition of "M consecutive times", rather than a single measurement result, is a key anti-interference design. It effectively filters out the brief fluctuations of the index caused by instantaneous signal noise, accidental fluctuations in a single operation, or other random interference, ensuring that the judgment of state transition is based on continuous and stable trend changes, thereby improving the stability and reliability of system output and avoiding the waste of maintenance resources due to misjudgment.

[0128] It should be noted that the determination of a health testing cycle must meet two conditions: first, the current standardized health index is not lower than the first lower health threshold of the previous cycle; second, the state confidence is not lower than 0.8. The state confidence is calculated by the frequency of the standardized health index crossing the health interval threshold in the last 20 measurements, i.e., the confidence equals 1 minus the ratio of the number of crossings to 20. When the number of crossings in the last 20 measurements does not exceed 2, it indicates that the current state is stable, and the confidence is not lower than 0.9, which is determined as a high-confidence health cycle, and its standardized health index can be used for threshold updates. If the number of crossings exceeds 5, it indicates that the state fluctuates greatly, and the confidence is not higher than 0.75. This cycle is not included in the health cycle statistics. This rule can ensure that the data source used for threshold updates is stable health state data, avoiding distortion of threshold updates due to state fluctuations.

[0129] In one embodiment, the steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the state confidence level include:

[0130] S54, Extract the time-series data of the standardized health index over the most recent W test periods;

[0131] S55, perform linear fitting on the time series data to obtain the slope of the standardized health index;

[0132] S56, if the slope of change is less than zero, then the remaining number of test cycles to reach the fault threshold is predicted by dividing the difference between the standardized health index and the second lower health threshold by the absolute value of the slope of change, using the following formula:

[0133]

[0134] In the formula, This indicates the predicted number of remaining test cycles, and is a positive number. This represents the second lower health threshold. This represents the standardized health index for the current testing period. This represents the absolute value of the slope of change obtained by linearly fitting the standardized health index over the most recent W test periods, and represents the average rate of deterioration. When the fitted slope... When < 0, use the above formula to calculate, if If ≥0, then RUL is set to infinity or a very large preset value.

[0135] As described in steps S54-S56 above, the present invention can transform time series data of standardized health indices into quantitative estimates of the future failure time of equipment, thereby improving health status monitoring from diagnosing the present to predicting the future, thus achieving early and predictive warnings.

[0136] The choice of using "the most recent W cycles" instead of all historical data is based on the following: the performance degradation of mechanical components is a gradual process that may be accompanied by phased accelerations. The historical state in the distant past is less relevant to predicting the deterioration trend in the near future, while excessively long historical data may contain early states that are no longer representative, which may dilute or distort the current trend signal. By focusing on recent data, the system can more sensitively capture the current performance degradation trajectory, thereby making the prediction results more timely and accurate. The parameter W, as a configurable integer preset value, allows the system to adjust and optimize according to the aging characteristics of the actual components (such as the rate of degradation) and the density of test cycles.

[0137] This invention chooses a linear model as the basis for preliminary prediction because it is computationally simple, has intuitive physical meaning, and has good approximation in many slow linear degradation scenarios. Even if the actual degradation path is not completely linear, the slope obtained by linear fitting can serve as an effective representation of the current "average degradation rate" and provide core input for the next step of extrapolation prediction.

[0138] The formula for calculating the remaining test cycles defines the predicted endpoint, namely reaching the failure threshold. This is a clear trigger point strongly correlated with maintenance actions. Secondly, the formula quantifies the remaining "health margin" between the current healthy state and the failure-triggered state. Finally, the denominator (the absolute value of the slope of change) represents the current "rate" at which health is consumed. Dividing the remaining "health margin" by the "rate" of consumption naturally yields the "time" (in test cycles) required to exhaust the margin under the assumption of maintaining the current consumption rate. This prediction model directly transforms the status monitoring data (current health index, historical trend) into time information crucial for operational decisions, namely the remaining service life. When the slope of change obtained from linear fitting is non-negative, the system determines that there is no deterioration trend. Therefore, the predicted value of the remaining service life is set to a maximum value or marked as "infinity". This indicates that the failure risk is very low under the current trend, and there is no need to trigger early warning based on prediction, thereby avoiding unnecessary maintenance scheduling.

[0139] It should be noted that the remaining lifetime prediction needs to be based on the actual engineering scenario and set boundary conditions: First, when the current standardized health index is lower than the second lower health threshold, it indicates that the test socket has entered the warning state. At this time, there is no need to calculate the remaining lifetime, and the second-level warning should be triggered first to prompt maintenance personnel to intervene in time. Second, when the slope of the linear fit is in the range of [-0.01, 0), it indicates that the health status is slowly deteriorating. In order to reserve sufficient maintenance preparation time, the calculated remaining lifetime is multiplied by a safety factor of 1.2. Third, when the slope is less than -0.1, it indicates that the health status is rapidly deteriorating. In order to avoid failure due to untimely maintenance, the remaining lifetime is multiplied by a coefficient of 0.8, and the first-level warning is triggered for expedited processing. These boundary conditions make the remaining lifetime prediction results more in line with the actual operation and maintenance, avoiding unnecessary prediction calculations and adjusting the response strategy according to the rate of deterioration, thereby improving the practicality of the prediction results.

[0140] In one embodiment, the steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the state confidence level include:

[0141] S57, Calculate the confidence level of the current health status determination based on the frequency of the standardized health index crossing the health interval threshold of the most recent preset number of times;

[0142] It's important to note that frequent fluctuations around the threshold indicate an unstable or critical state, resulting in low confidence. Conversely, lower confidence indicates higher confidence. State confidence adds a "credibility" label to the state determination result. This allows system or maintenance personnel to understand the reliability of the current judgment. Warnings with low confidence can be subject to more cautious review, while warnings with high confidence can be responded to quickly, thereby optimizing the decision-making process and improving the efficiency and security of human-machine collaboration.

[0143] S58. A comprehensive risk score is calculated by weighting the standardized health index, the state confidence level, and the normalized number of remaining test cycles.

[0144] Comprehensive risk scoring formula:

[0145]

[0146] In the formula, This represents the overall risk score; a higher value indicates a higher risk. This represents the current standardized health index. This indicates the state confidence level for the current testing period. This represents the number of remaining test cycles after normalization. , and These represent the fourth, fifth, and sixth weighting coefficients, respectively, which are preset constants; the sum of the three equals 1. Small normal numbers added to prevent the denominator from being zero

[0147] S59, Establish a risk accumulator to accumulate the portion of the comprehensive risk score that exceeds a preset basic threshold;

[0148] The risk accumulator update formula is:

[0149]

[0150] In the formula, Indicates the first The value of the risk accumulator updated after each test period ranges from [0,1]. Indicates the first The value of the risk accumulator for each test period. This represents the overall risk score for the current testing period. This represents the risk accumulation threshold, a preset constant used to determine what level of risk warrants attention. The accumulation rate, a small preset normal value, is used to determine how long a risk needs to persist before triggering an alarm. (Setting the risk accumulation threshold and accumulation rate is a multi-objective trade-off process, and its specific value needs to comprehensively consider multiple constraints of the actual production scenario: First, it depends on the fault tolerance of the testing process; the higher the product value and the smaller the allowable test error, the lower the threshold and the faster the rate should be to achieve more sensitive monitoring; second, it is constrained by the noise level of the signal; the greater the interference such as production line vibration, the higher the threshold and the slower the rate should be to effectively filter random disturbances and improve robustness; at the same time, the maintenance response cost directly affects the parameter selection; the higher the cost of manual inspection or downtime, the higher the threshold should be.) The higher the rate, the slower it should be to avoid unnecessary maintenance. Furthermore, the gradual nature of the fault itself determines the direction of rate adjustment. For slowly developing faults, such as nut creep loosening, the rate should be slower to capture long-term trends, while for potentially sudden faults, such as snap-fit ​​brittle fracture, the rate should be set faster to achieve rapid response. The quantification of all parameters is based on historical data statistics. Threshold benchmarks are set by analyzing the fluctuation range of comprehensive risk scores under a large number of healthy states. Simulation tests are used to find a balance between early warning time and false alarm rate to determine the optimal cumulative rate, enabling the system to achieve a balance of reliability, economy, and timeliness in complex engineering environments. The update is a recursive calculation used to limit the cumulative value to 1, whereby a small amount is added each time based on the portion of the comprehensive risk score that exceeds a preset basic threshold.

[0151] S510, Multiple warning level thresholds are set according to the value of the risk accumulator. When the value of the risk accumulator exceeds one of the warning level thresholds, a warning action of the corresponding level is triggered. The warning action includes different levels from logging, marking as a concern, generating a planned maintenance work order to immediate shutdown intervention.

[0152] S511, if the comprehensive risk score of multiple consecutive preset test cycles after the warning action is triggered is lower than the preset risk accumulation threshold, then the risk accumulator is controlled to decay at a preset rate, and the warning level is adaptively downgraded accordingly.

[0153] As described in steps S57-S511 above, this invention aims to address the problems of early warning mechanisms being susceptible to false alarms due to transient interference and the lack of dynamic management after early warning. By introducing state judgment credibility assessment, multi-factor risk fusion calculation, a recursive decision mechanism with memory function, and a hierarchical response and adaptive recovery strategy that is deeply integrated with operation and maintenance practices, a highly reliable early warning decision system is constructed.

[0154] The confidence level of the current health status is calculated based on the frequency with which the standardized health index crosses the health interval threshold in the most recent preset number of observation windows. Specifically, the calculation logic is as follows: within the most recent preset length of observation window, the number of times the standardized health index crosses the health interval threshold (such as the first lower health limit threshold) is counted. The higher the frequency of crossing, the more the health index fluctuates around the threshold, and the more unstable and unreliable the current status judgment (e.g., "healthy" or "warning") is, resulting in a lower calculated confidence level. Conversely, if the health index remains stable on one side of the threshold and rarely crosses it, then... The higher the confidence level of the status determination, the more reliable the status determination becomes. This step quantifies the "certainty" of the status determination result to address the issue that not all "below the threshold" signals are equally reliable. For example, the "warning" status of a socket whose health index fluctuates drastically around the threshold is far less reliable than that of another socket whose health index is consistently and stably below the threshold. Introducing status confidence as an input dimension for subsequent decision-making enables the system to distinguish between deterministic risks and uncertain fluctuations, thereby providing a basis for more prudent and accurate decision-making and effectively preventing hasty or erroneous warnings due to unstable data.

[0155] By weighting and calculating a comprehensive risk score based on the standardized health index, the state confidence level, and the normalized number of remaining test cycles, the system achieves the fusion of multiple independent assessment dimensions into a single risk decision scale. This step combines three core risk factors through a specific weighted summation formula: first, "current health deviation" directly reflects the absolute degree of poor health; second, "state uncertainty" reflects the reliability of the judgment itself; and third, "time urgency" reflects the estimated proximity of the failure. By assigning configurable weight coefficients to these three factors, the system can adjust the proportion of attention to different risk dimensions according to the actual operation and maintenance strategy, and finally calculate a comprehensive risk score within a certain range (e.g., between 0 and 1), realizing a three-dimensional, multi-criteria risk assessment.

[0156] The risk accumulator-based hierarchical early warning triggering and execution mechanism enables the verification of risk over time. Instantaneous high-risk scores (potentially caused by a single disturbance) contribute minimally to the accumulator's increase due to the limited portion exceeding the threshold and its multiplication by a small rate factor, thus not immediately triggering a high-level alarm. Only when the risk situation persists and the comprehensive risk score repeatedly or continuously exceeds the basic threshold will the accumulator value be steadily and slowly increased. This effectively filters out accidental impulse interference, ensuring that the final early warning signal represents a time-verified, real, and continuous risk trend, greatly enhancing the system's anti-interference capability and the weight of the early warning. Based on this stable cumulative value, the system outputs action measures: multiple warning level thresholds are preset (e.g., 0.3, 0.6, 0.9). When the value of the risk accumulator exceeds a certain level threshold, a preset warning action strictly corresponding to that level is triggered. These actions are tiered, ranging from low-impact actions such as logging and marking as important, to medium-impact actions such as generating a planned maintenance work order and notifying the management system, and high-impact actions such as triggering an immediate shutdown command and issuing an emergency alarm. Through the above steps, different levels of risk can be accurately matched with different costs and different urgency of operation and maintenance responses, achieving optimal allocation of maintenance resources and less disturbance to the operation and maintenance process.

[0157] Since the health status of equipment can be improved through effective maintenance interventions (such as tightening nuts) or stress redistribution, rather than deteriorating in a one-way manner, when an early warning is triggered, if subsequent continuous monitoring finds that the risk source has been eliminated (the comprehensive risk score remains below the cumulative threshold), the system controls the value of the risk accumulator to gradually decay at a preset rate. When the decayed cumulative value falls below the next level of early warning threshold, the system automatically downgrades the early warning level until it fully restores normal monitoring status. This mechanism avoids the system from labeling repaired or temporarily stable equipment with permanent faults, ensuring real-time synchronization between the early warning status and the actual risk of the equipment.

[0158] For example, the specific execution logic of the warning downgrade is as follows: the number of consecutive preset test cycles is consistent with the health threshold update window (preset to 30). That is, after the warning is triggered, the comprehensive risk score must be lower than the risk accumulation threshold (preset to 0.3) for 30 consecutive test cycles before the risk accumulator decay mechanism is activated. The decay rate is set to 1 / 2 of the accumulation rate. If the accumulation rate is 0.05, the risk accumulator value decays by 0.025 each time the cycle is updated until it drops to 0 or below the next level warning threshold. For example, when the risk accumulator value is 0.6 (level 2 warning threshold), if the comprehensive risk score is lower than 0.3 for 30 consecutive cycles, the value can drop to 0 after 24 cycles of decay, and the warning level is downgraded from level 2 to normal monitoring status. If the comprehensive risk score is not lower than the threshold in any cycle during the decay process, the decay is stopped immediately and the risk accumulation logic is restored. This design ensures that the warning downgrade is synchronized with the actual risk status and avoids premature downgrade leading to risk omission.

[0159] In one embodiment, the invention further includes preprocessing the acquired current pressure signal, including:

[0160] S11. Apply a moving average filter or a low-pass digital filter to filter the original pressure signal to suppress high-frequency noise and obtain the filtered signal.

[0161] S12, detect the peak value of the filtered signal. If the peak value does not fall within the preset reasonable force value range, the acquisition is determined to be invalid and discarded, and an effective pressure signal is obtained.

[0162] S13, Based on the difference between the effective pressure signal and the standard healthy pressure waveform template in the undeformed section, calculate and apply an amplitude correction coefficient to correct the effective pressure signal and compensate for the slow drift of the sensor gain.

[0163] S14, use the corrected effective pressure signal as the current pressure signal.

[0164] As described in steps S11-S14 above, this set of steps aims to address the real-world engineering problem that in actual industrial testing environments, the raw pressure signals directly acquired are inevitably mixed with various interferences, noises, and the slow drift of the sensor itself. If used directly for subsequent precise condition assessment without processing, it will severely degrade the accuracy of feature extraction and may even lead to misjudgment.

[0165] The application of a moving average filter or a low-pass digital filter to filter the original pressure signal to suppress high-frequency noise is the first step in addressing interference from electronic noise, high-frequency mechanical vibration components, and other sources of interference in the signal. Specifically, in the data acquisition unit or initial processing module, the original pressure signal sequence acquired at a high sampling rate is processed through a moving average window or a low-pass digital filter with a specific cutoff frequency. The moving average filter outputs a new sequence by taking the arithmetic mean of multiple consecutive sampling points, effectively smoothing random high-frequency fluctuations. The low-pass digital filter, on the other hand, directly attenuates high-frequency noise through digital algorithms (such as FIR or IIR filters) to a value higher than the set threshold. The technical contribution of this step, which determines the cutoff frequency of the signal, lies in its ability to remove high-frequency interference components from the signal that are unrelated to the mechanical degradation of the test socket. In pressure signals, changes in mechanical characteristics such as stiffness and damping of the connection structure are reflected, mainly manifested as changes in waveform shape at low or mid-low frequencies. Circuit thermal noise, environmental electromagnetic interference, and far-field mechanical vibrations usually appear as high-frequency noise superimposed on the useful signal. If not suppressed, these high-frequency noises will introduce significant random errors in subsequent calculations of morphological difference signals and integration operations, blurring or even masking the true degradation trend. Therefore, this filtering operation can improve the signal-to-noise ratio and ensure the accuracy of subsequent feature extraction.

[0166] After filtering, the global peak value of the pressure signal during this pressing cycle is immediately detected and compared with a preset reasonable force range (the lower and upper limits are set according to the specific test socket model, chip type, and Loader robot parameters). If the peak force value is abnormally low (for example, it may be due to the chip not being fully placed, poor instantaneous contact of the sensor, or insufficient pressing stroke of the robot) or abnormally high (for example, it may be due to abnormal jamming or abnormal impact on the sensor), then the mechanical process of this loading and unloading action is determined to be inconsistent with standard working conditions, and the pressure signal collected in this way cannot represent the normal mechanical connection structure. The responses are therefore actively marked as invalid by the system and discarded, not entering the subsequent analysis process. In actual production line operation, occasional non-standard operations or momentary failures are difficult to completely avoid. If these abnormal data are mixed with normal data for analysis, it will seriously pollute the health baseline, causing violent and meaningless fluctuations in the health index calculation and triggering false warnings. By setting physically reasonable force value boundaries and performing real-time verification, the system can automatically filter out these bad samples, ensuring the consistency, reliability and comparability of the dataset input to the core state assessment model in a physical sense, thereby avoiding system malfunctions caused by abnormal data acquisition.

[0167] After confirming the valid signal, a segment of the pressure waveform is selected that is considered stable even under healthy conditions and is not easily affected by structural degradation (e.g., the flat baseline area before pressure rise, or a period of time far from the main loading / unloading process). The average amplitude of the valid pressure signal in this segment is compared with the average amplitude of the same segment of the standard healthy pressure waveform template, and a ratio is calculated as the "amplitude correction coefficient". Subsequently, all points in the entire valid pressure signal sequence acquired this time are multiplied by this coefficient, thereby completing the amplitude recalibration. This step combats the slow gain drift problem that is unavoidable for pressure sensors in industrial environments. Factors such as temperature changes and material aging may affect the sensor's sensitivity (i.e., the ratio between the output signal and the actual pressure). As the changes occur slowly over time, without compensation, this slow drift can be misinterpreted as a continuous trend of pressure level changes in subsequent calculations of morphological difference signals. This can interfere with the accurate calculation of stiffness factors and other parameters. By using a "reference segment" in the health template that should remain physically constant as a reference, the system can estimate and offset the current sensor gain deviation in real time. This ensures that the corrected signal is aligned with the reference established during template learning in terms of amplitude scale. This guarantees the comparability of signals collected at different times in terms of absolute amplitude, allowing the health index model to focus on waveform morphological differences caused by changes in the mechanical structure state, rather than amplitude differences caused by sensor characteristic drift. This improves the stability and assessment accuracy of the system during long-term operation.

[0168] like Figure 7 As shown, the present invention also discloses a health status early warning system for a chip aging test board, comprising:

[0169] The signal sensing module is used to collect the current pressure signal of the target test socket during the real-time loading and unloading process, align the current pressure signal with the pre-stored standard health pressure waveform template, and calculate the morphological difference signal after alignment.

[0170] The signal processing module is used to acquire the energy integral of the morphological difference signal in three time windows: the early stage of the pressure rise, the pressure peak holding period, and the pressure fall period, so as to calculate the stiffness trend factor that reflects the stiffness change trend of the target test socket.

[0171] The feature extraction module is used to calculate the difference between the integral values ​​of the morphological difference signal at the pressure rising edge and the pressure falling edge, and to calculate the hysteresis anomaly factor reflecting the mechanical hysteresis characteristics.

[0172] The health assessment module is used to obtain the incremental health index of each historical test period, combine the relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculate the incremental health index of the current test period, and calculate the standardized health index based on the cumulative sum of the incremental health index.

[0173] The early warning decision module is used to determine the current health status of the target test socket based on the health interval threshold of the standardized health index, predict the remaining service life of the target test socket based on the time-series change trend of the standardized health index, and generate a graded early warning decision based on the status confidence level.

[0174] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0175] The above description is merely a preferred embodiment of the present invention and does 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 health state early warning method of a chip aging test board, characterized in that, This is applied to a chip aging test board, the chip aging test board comprising: An aging test plate (1) includes a substrate (120) and a plurality of test sockets (150) mounted on the substrate (120). A protective plate (110) is fixed on the back of the substrate (120), and the protective plate (110) is provided with through holes corresponding to the positions of each of the test sockets (150). The support plate (2) includes a base plate (210) and a plurality of monitoring support columns (230) disposed on the base plate. The position of the monitoring support column (230) corresponds to the through hole, such that when the aging test plate (1) is installed on the support plate (2), the top end of each monitoring support column (230) passes through the corresponding through hole and contacts the back of the test socket (150). Each of the monitoring support columns (230) integrates a sensor (240) for collecting forces perpendicular to the support direction; the bottom of the monitoring support column (230) is provided with an elastic height adjustment element; The data acquisition unit is electrically connected to all sensors (240) and is used to synchronously acquire the mechanical signals generated by each of the test sockets (150) when subjected to downward pressure during each loading and unloading process; The health status early warning method includes the following steps: The current pressure signal of the target test socket during real-time loading and unloading is collected. The current pressure signal is time-aligned with the pre-stored standard health pressure waveform template, and the morphological difference signal after alignment is calculated. The energy integral of the morphological difference signal is obtained within three time windows: the early stage of the pressure rise, the pressure peak holding period, and the pressure fall period, in order to calculate the stiffness trend factor that reflects the stiffness change trend of the target test socket. Calculate the difference between the integral values ​​of the morphological difference signal at the pressure rise edge and pressure fall edge, and calculate the hysteresis anomaly factor reflecting the mechanical hysteresis characteristics. The incremental health index of each historical test period is obtained. The relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods is combined to calculate the incremental health index of the current test period. The standardized health index is calculated based on the cumulative sum of the incremental health index. The current health status of the target test socket is determined based on the health interval threshold of the standardized health index, and the remaining service life of the target test socket is predicted based on the time-series change trend of the standardized health index. A graded early warning decision is generated by combining the status confidence.

2. The health state early warning method of a chip burn-in test board according to claim 1, wherein, The steps of obtaining the incremental health index of each historical test period, combining the relative change rates of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculating the incremental health index of the current test period, and calculating the standardized health index based on the cumulative sum of the incremental health indices include: Obtain the stiffness trend factor and hysteresis anomaly factor corresponding to the current test cycle, as well as the stiffness trend factor and hysteresis anomaly factor corresponding to the previous test cycle. The incremental health index for the current testing period is calculated by weighting the relative change rate of the stiffness trend factor in the current testing period and the stiffness trend factor in the previous testing period, as well as the relative change rate of the hysteresis anomaly factor in the current testing period and the hysteresis anomaly factor in the previous testing period. The incremental health index of each historical test period is obtained, and the cumulative health index is calculated by combining the incremental health index of the current test period with the index decay weighted summation method. The cumulative health index is normalized to obtain a standardized health index that characterizes the health of the target test socket connection structure.

3. The health state early warning method of a chip burn-in test board according to claim 2, characterized in that, The step of calculating the incremental health index for the current testing period by weighting the relative change rate of the stiffness trend factor in the current testing period and the stiffness trend factor in the previous testing period, and the relative change rate of the hysteresis anomaly factor in the current testing period and the hysteresis anomaly factor in the previous testing period, includes: Obtain the current test task, obtain the chip's package type and power consumption level based on the current test task, determine the load type identifier, and adjust the weighting coefficients of the weighted calculation based on the load type identifier; Obtain the historical maintenance records of the target test socket, obtain the main failure modes of the target test socket based on the historical maintenance records, and adjust the weighting coefficients of the weighted calculation based on the main failure modes; The adjusted weighting coefficients will be reused in the weighted calculation of the incremental health index for the current and subsequent testing periods.

4. The health status early warning method for a chip aging test board according to claim 1, characterized in that, The steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the status confidence include: Preset dynamic health range thresholds, including a first lower health threshold, a second lower health threshold, and a fault threshold; Obtain historical health index statistics of similar test sockets to initialize the health interval threshold, and adaptively update the health interval threshold based on the minimum value of the standardized health index of the most recent multiple health test cycles during online operation; The standardized health index is compared with the health interval threshold, and the current health status is determined according to the preset health interval threshold crossing rules.

5. The health status early warning method for a chip aging test board according to claim 4, characterized in that, The steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the status confidence include: Extract time-series data of the standardized health index over the most recent testing periods; The slope of the standardized health index is obtained by performing a linear fit on the time series data. If the slope of change is less than zero, the number of remaining test cycles to reach the fault threshold is predicted by dividing the difference between the standardized health index and the second lower health threshold by the absolute value of the slope of change.

6. The health status early warning method for a chip aging test board according to claim 5, characterized in that, The steps of determining the current health status of the target test socket based on the health interval threshold of the standardized health index, predicting its remaining service life based on the time-series change trend of the standardized health index, and generating a graded early warning decision based on the status confidence include: The confidence level of the current health status is calculated based on the frequency of the standardized health index crossing the health interval threshold of the most recent preset number of times. A comprehensive risk score is calculated by weighting the standardized health index, the state confidence level, and the normalized number of remaining test cycles. Establish a risk accumulator to accumulate the portion of the comprehensive risk score that exceeds a preset basic threshold. Multiple warning level thresholds are set based on the value of the risk accumulator. When the value of the risk accumulator exceeds one of the warning level thresholds, the corresponding warning action is triggered. If the overall risk score for multiple consecutive preset test cycles is lower than the preset risk accumulation threshold after the warning action is triggered, the risk accumulator is controlled to decay at a preset rate, and the warning level is adaptively downgraded accordingly.

7. The health status early warning method for a chip aging test board according to claim 1, characterized in that, The steps for obtaining the standard health stress waveform template include: When the target test socket is in a healthy state, its pressure signal is collected during multiple standard loading and unloading processes. Based on the moment when the pressure signal first exceeds a preset quiet threshold, all collected pressure signals are time-aligned. The average of the aligned pressure signals is calculated to obtain a quasi-healthy pressure waveform template unique to the target test socket.

8. The health status early warning method for a chip aging test board according to claim 1, characterized in that, It also includes a step of preprocessing the current pressure signal: The original pressure signal is filtered to suppress high-frequency noise, resulting in a filtered signal. The peak value of the filtered signal is detected. If the peak value does not fall within the preset reasonable force value range, the acquisition is determined to be invalid and discarded, and an effective pressure signal is obtained. Based on the difference between the effective pressure signal and the standard healthy pressure waveform template in the undeformed section, an amplitude correction coefficient is calculated and applied to correct the effective pressure signal. Use the corrected effective pressure signal as the current pressure signal.

9. A health status early warning system for a chip aging test board, characterized in that, include: The signal sensing module is used to collect the current pressure signal of the target test socket during the real-time loading and unloading process, align the current pressure signal with the pre-stored standard health pressure waveform template, and calculate the morphological difference signal after alignment. The signal processing module is used to acquire the energy integral of the morphological difference signal in three time windows: the early stage of the pressure rise, the pressure peak holding period, and the pressure fall period, so as to calculate the stiffness trend factor that reflects the stiffness change trend of the target test socket. The feature extraction module is used to calculate the difference between the integral values ​​of the morphological difference signal at the pressure rising edge and the pressure falling edge, and to calculate the hysteresis anomaly factor reflecting the mechanical hysteresis characteristics. The health assessment module is used to obtain the incremental health index of each historical test period, combine the relative change rate of the stiffness trend factor and the hysteresis anomaly factor in adjacent test periods, calculate the incremental health index of the current test period, and calculate the standardized health index based on the cumulative sum of the incremental health index. The early warning decision module is used to determine the current health status of the target test socket based on the health interval threshold of the standardized health index, predict the remaining service life of the target test socket based on the time-series change trend of the standardized health index, and generate a graded early warning decision based on the status confidence level.

Images