Method and system for eMMC potential defect detection based on temperature rise rate and current fingerprint

Abstract

This application relates to a method and system for detecting potential defects in eMMCs based on temperature rise rate and current fingerprint. The method includes: subjecting the target eMMC to controlled read / write load disturbance excitation to obtain detection process data; performing thermal response deconvolution processing on the detection process data to obtain current time-series data and temperature rise rate time-series data; performing thermoelectric coupling dynamic feature analysis on the current time-series data and temperature rise rate time-series data to obtain current fingerprint features and phase map features; performing deviation analysis on the current fingerprint features and phase map features to obtain abnormal deviation; and determining defects based on the abnormal deviation to obtain the potential defect detection result of the target eMMC. This method can improve system reliability and maintenance foresight.

Description

Technical Field

[0001] This application relates to the field of intelligent detection technology, and in particular to a method and system for detecting potential defects in eMMC based on temperature rise rate and current fingerprint. Background Technology

[0002] Traditional technologies for detecting potential defects in eMMC typically employ a combination of production line testing (such as read / write erase cycles, bad block scanning, interface timing verification, and ECC / CRC error statistics), aging screening (high and low temperature and power-on stress testing), operational health monitoring (based on SMART-based lifetime counts, erase / write cycles, remapped block counts, write failures / read retries, etc.), and log / abnormal behavior analysis (boot failures, mount anomalies, I / O latency jitter, disk drop resets, etc.) to identify and warn of risks such as storage unit degradation, controller malfunctions, firmware instability, and poor soldering connections. However, traditional detection methods often rely on fixed thresholds and offline testing, making it difficult to accurately identify early latent defects and progressive failures in complex scenarios, resulting in insufficient system reliability and maintenance foresight. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for detecting potential defects in eMMC based on temperature rise rate and current fingerprint, which can improve system reliability and maintenance foresight, in order to address the above-mentioned technical problems.

[0004] In a first aspect, this application provides a method for detecting potential defects in eMMC based on temperature rise rate and current fingerprint, including: The target eMMC is subjected to controlled read / write load perturbation stimulation to obtain detection process data; The detection process data is subjected to thermal response deconvolution processing to obtain current time series data and temperature rise rate time series data; Thermoelectric coupling dynamic feature analysis is performed on the current time series data and the temperature rise rate time series data to obtain current fingerprint features and phase map features; Deviation analysis is performed on the current fingerprint features and the phase map features to obtain the abnormal deviation. The abnormal deviation is used to determine the defect, and the potential defect detection results of the target eMMC are obtained.

[0005] Secondly, this application also provides an eMMC potential defect detection system based on temperature rise rate and current fingerprint, including: a server and a terminal; The server is used to perform controlled read / write load perturbation stimulation on the target eMMC to obtain detection process data; the detection process data is acquired through the terminal, and the terminal transmits it to the server through the network. The server is used to perform thermal response deconvolution processing on the detection process data to obtain current time-series data and temperature rise rate time-series data. The server is used to perform thermoelectric coupling dynamic feature analysis on the current time series data and the temperature rise rate time series data to obtain current fingerprint features and phase map features. The server is used to perform deviation analysis on the current fingerprint features and the phase map features to obtain the abnormal deviation. The server is used to determine the defects based on the abnormal deviation and obtain the potential defect detection results of the target eMMC.

[0006] The aforementioned method and system for detecting potential defects in eMMC based on temperature rise rate and current fingerprinting compensates for the thermal inertia and broadening effects in the temperature measurement link by subjecting the target eMMC to controlled read / write load disturbance excitation and combining it with thermal response deconvolution processing. This allows for the acquisition of current timing data and temperature rise rate timing data that are more accurately aligned with the excitation event. Furthermore, thermoelectric coupling dynamic feature analysis is used to simultaneously extract current fingerprint features and phase map features, characterizing the dynamic response differences of the eMMC from the perspective of thermal and electrical domain coupling mechanisms, rather than relying solely on single indicators such as absolute temperature, average current, or read / write error counts. Further, by performing deviation analysis on the current fingerprint and phase map features, early abnormal offsets can be identified more sensitively than in a healthy response state. Defect determination is then performed based on the abnormal deviation, enabling early detection and risk warning of potential defects before read / write errors or obvious bad block anomalies appear in the target eMMC. Therefore, this method features high detection sensitivity, strong resistance to operating condition fluctuations, low false alarm rate, online deployment capability, and benefits for improving system reliability and maintenance foresight. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is an application environment diagram of an eMMC potential defect detection method based on temperature rise rate and current fingerprint in one embodiment; Figure 2 This is a flowchart illustrating a potential defect detection method for eMMC based on temperature rise rate and current fingerprint in one embodiment. Figure 3 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0010] This application provides a method for detecting potential defects in eMMC based on temperature rise rate and current fingerprint, which can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on other network servers. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0011] In one exemplary embodiment, such as Figure 2 As shown, a method for detecting potential defects in eMMC based on temperature rise rate and current fingerprint is provided, and this method is applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 202 to 210. Wherein:

[0012] Step 202: Perform controlled read / write load perturbation on the target eMMC to obtain detection process data.

[0013] Step 204: Perform thermal response deconvolution processing on the detection process data to obtain current time series data and temperature rise rate time series data.

[0014] Step 206: Perform thermoelectric coupling dynamic characteristic analysis on the current time series data and the temperature rise rate time series data to obtain current fingerprint characteristics and phase map characteristics.

[0015] Step 208: Perform deviation analysis on the current fingerprint features and phase map features to obtain the abnormal deviation.

[0016] Step 210: Defect judgment is performed on the abnormal deviation to obtain the potential defect detection results of the target eMMC.

[0017] The target eMMC is an embedded multimedia memory device that requires potential defect detection.

[0018] The controlled read / write load disturbance excitation is a diagnostic read / write load applied to the target eMMC by the master controller according to preset timing and access rules, in order to induce its thermo-electric dynamic response.

[0019] The detection process data consists of detection-related signal data and timing marker data collected or recorded during the execution of controlled read / write load disturbance excitation.

[0020] Thermal response deconvolution processing is a process that compensates for temperature-related signals based on temperature measurement links or thermal response models to reduce the effects of thermal inertia and response broadening.

[0021] Current timing data is data that characterizes the relationship between the supply current and time during the detection process of the target eMMC.

[0022] Temperature rise rate time series data are data that characterize the relationship between the rate of temperature change of the target eMMC and time.

[0023] Thermoelectric coupling dynamic characteristic analysis is a process that combines the analysis of the coupling changes between current time series data and temperature rise rate time series data to extract characteristics that characterize dynamic response.

[0024] Current fingerprint features are a set of features extracted from current time-series data or thermoelectric coupling current response components to characterize the dynamic characteristics of the target eMMC current.

[0025] Phase map features are a set of features extracted from the geometric structure, topological structure, or evolutionary morphology of the thermoelectric coupling response in the phase map trajectory.

[0026] Deviation analysis is an analytical process that compares current fingerprint characteristics and phase map characteristics with a health response reference state to assess the degree of deviation.

[0027] Anomaly deviation is a result obtained from deviation analysis and is used to quantify the degree to which the current response state of the target eMMC deviates from the healthy response state.

[0028] Defect assessment is a process of determining whether a target eMMC has potential defect risks based on the degree of abnormal deviation and preset rules.

[0029] Potential defect detection results are outputs of defect determination that characterize whether the target eMMC has early defects and its risk status.

[0030] Specifically, the main control processor in the server sends a preset diagnostic stimulus sequence to the target eMMC during its idle service window, power-on self-test window, or maintenance time slot to form a controlled read / write load disturbance stimulus. The preset diagnostic stimulus sequence may include at least one read / write burst sequence, read / write alternating sequence, address dispersion / concentration access sequence, and data mode switching sequence, used to induce observable coupled dynamic responses of the target eMMC in the thermal and electrical domains. During stimulus execution, the start and end times of the stimulus, the stimulus type identifier, the access address range, the data mode identifier, and the timing markers of each stimulus segment are recorded to form detection process data corresponding to the controlled read / write load disturbance stimulus.

[0031] The power supply current sampling signal and temperature sampling signal corresponding to the excitation timing mark are extracted from the detection process data. Based on the thermal inertia model of the temperature measurement link, thermal response deconvolution processing is performed on the temperature sampling signal to reduce the impact of temperature sensor response hysteresis, encapsulation thermal diffusion low-pass effect, and sampling link broadening effect on temperature rise dynamics. During thermal response deconvolution, the current sampling signal is synchronously sliced ​​and time-aligned according to the excitation event boundaries to obtain current timing data. Simultaneously, the temperature rise rate is reconstructed from the deconvolution-compensated temperature response (e.g., through piecewise fitting, differential reconstruction, or filtered differentiation) to obtain temperature rise rate timing data.

[0032] Thermally locked phase-locked reference data is constructed based on the temperature rise rate time-series data. This reference data is then used to perform synchronous demodulation analysis on the current time-series data to extract thermoelectric coupled current response components. Further processing, including frequency shift coding, inter-frequency mapping, coherence enhancement, and inverse mapping demodulation, is employed to suppress non-thermally coupled disturbance components and enhance the current response components related to the dominant thermal response frequency band. Based on this, coupled fingerprint feature analysis is performed on the thermoelectric coupled current response component data. First, phase map features are obtained through phase trajectory skeleton extraction and topological hierarchical analysis. Then, inverse mapping and fingerprint aggregation correction are performed on the current response details using these phase map features to obtain the current fingerprint features. The phase map features may include trajectory structure hierarchy, loop structure features, and curvature distribution features, while the current fingerprint features may include dynamic amplitude features, frequency band energy features, time-series ripple features, and response detail morphology features.

[0033] Current fingerprint features and phase map features are input into the deviation analysis calculation model, and coupled manifold expansion is performed on them to obtain manifold offset representation data. Then, based on the health response reference model corresponding to the target eMMC (the model is constructed from thermoelectric coupling response samples formed by a healthy eMMC under multiple operating conditions), inverse constraint matching analysis is performed on the manifold offset representation data to generate matched residual spectrum data. During the inverse constraint matching analysis, residual spectra reflecting the structural differences between the current response and the healthy response can be obtained through offset anchor point expansion, inverse constraint probe construction, reverse activation matching, constraint conflict recombination, and spectral decomposition. Finally, the matched residual spectrum data is re-injected into the constraint space of the health response reference model for consistency verification and mismatch convergence evaluation, outputting the abnormal deviation degree characterizing the degree to which the current state of the target eMMC deviates from the healthy state.

[0034] The defect assessment module assesses potential defects in the target eMMC based on abnormal deviation and preset assessment rules. These preset rules may include at least one of the following: static threshold assessment rules, segmented threshold assessment rules, trend accumulation assessment rules, and multiple detection consistency assessment rules. When the abnormal deviation exceeds the assessment threshold for the corresponding operating condition, or when the abnormal deviation shows a continuous upward trend over multiple consecutive detection cycles and meets stability conditions, the target eMMC is assessed as having a potential defect risk, and the potential defect detection result is output. This potential defect detection result may include an anomaly flag, risk level, deviation mode flag, and recommended re-inspection status, providing early warning before read / write errors or obvious bad block anomalies appear in the target eMMC.

[0035] The aforementioned method for detecting potential defects in eMMC based on temperature rise rate and current fingerprinting compensates for thermal inertia and broadening effects in the temperature measurement link by subjecting the target eMMC to controlled read / write load disturbance excitation and combining it with thermal response deconvolution processing. This allows for the acquisition of current timing data and temperature rise rate timing data that are more accurately aligned with the excitation event. Furthermore, thermoelectric coupling dynamic feature analysis is used to simultaneously extract current fingerprint features and phase map features, characterizing the dynamic response differences of the eMMC from the perspective of thermal and electrical domain coupling mechanisms, rather than relying solely on single indicators such as absolute temperature, average current, or read / write error counts. Further, by performing deviation analysis on the current fingerprint and phase map features, early abnormal offsets can be identified more sensitively than in a healthy response state. Defect determination is then performed based on the abnormal deviation, enabling early detection and risk warning of potential defects before read / write errors or obvious bad block anomalies appear in the target eMMC. Therefore, this method features high detection sensitivity, strong resistance to operating condition fluctuations, low false alarm rate, online deployment capability, and benefits for improving system reliability and maintenance foresight.

[0036] In an exemplary embodiment, thermoelectric coupling dynamic feature analysis is performed on current time-series data and temperature rise rate time-series data to obtain current fingerprint features and phase map features, including steps 302 to 306. Wherein:

[0037] Step 302: Based on the time series data of temperature rise rate, construct a phase-locked reference for the dominant frequency band of thermal response to obtain thermal phase-locked reference data.

[0038] Step 304: Based on the thermal lock-in reference data, perform synchronous demodulation analysis on the current timing data to obtain thermoelectric coupling current response component data.

[0039] Step 306: Perform coupling fingerprint feature analysis on the thermoelectric coupling current response component data to obtain current fingerprint features and phase map features.

[0040] The thermal response dominant frequency band is the frequency range in the time series data of temperature rise rate that is related to the controlled read / write load disturbance excitation and has stable energy accumulation and repetitive response characteristics.

[0041] Phase-locked reference construction is a process of generating a thermal domain reference signal or reference template for synchronous demodulation based on the frequency and phase information of the dominant thermal response frequency band.

[0042] Thermal lock-in reference data is constructed from time-series data of temperature rise rate and is used to characterize the relationship between the reference frequency and the reference phase in the dominant frequency band of thermal response.

[0043] Synchronous demodulation analysis is an analytical process that uses thermally locked phase-locked reference data to extract phase correlations and suppress interference in current time series data in order to separate thermally coupled current response components.

[0044] Thermoelectric coupling current response component data is current response data extracted from current time series data that is coherent with the dominant frequency band of thermal response and characterizes the dynamic relationship of thermo-electric coupling.

[0045] Coupled fingerprint feature analysis is an analytical process that extracts phase map structural features and current detail features from thermoelectric coupled current response component data to obtain current fingerprint features and phase map features.

[0046] Specifically, the temperature rise rate time-series data is preprocessed to reduce the impact of sampling noise, boundary abrupt changes, and unstable drift on frequency band identification. Preprocessing may include detrending, segmented smoothing, outlier suppression, and excitation event boundary alignment. Combined with the time-series markers corresponding to controlled read / write load disturbance excitations, frequency domain analysis or time-frequency joint analysis is performed on the preprocessed temperature rise rate time-series data to identify the dominant thermal response frequency bands exhibiting stable energy accumulation, continuous phase evolution, and repetitive response characteristics across multiple excitation cycles. After determining the dominant thermal response frequency bands, the corresponding center frequency, bandwidth range, initial reference phase value, phase tracking trajectory, and cycle boundary information are extracted. Thermal phase-locked reference data is then constructed based on the center frequency, bandwidth range, initial reference phase value, phase tracking trajectory, and cycle boundary information. This thermal phase-locked reference data can take the form of a single-frequency phase-locked reference, a multi-phase phase-locked reference, or a multi-dominant frequency band composite phase-locked reference.

[0047] Using thermally locked phase-locked reference data as the demodulation benchmark, phase alignment and synchronous demodulation are performed on the current time series data to extract current response components coherently related to the dominant thermal response frequency band. First, the current time series data is mapped to a reference frequency band and divided into phase windows based on the thermally locked phase-locked reference data. Then, frequency shift coding and inter-frequency mapping analysis are used to convert the current variation components related to the dominant thermal response frequency band to the difference frequency domain, forming candidate difference frequency response data. Next, coherent enhancement processing is performed on the candidate difference frequency response data, such as phase stability spectrum construction, coherent gated resampling, anti-phase residual cancellation, and coherent superposition correction, to enhance the coupled response components consistent with the thermally locked phase-locked reference and suppress power supply ripple, random noise, and non-thermally coupled disturbance components. Finally, inverse mapping demodulation is performed on the enhanced difference frequency response to return it to the coupled response representation space corresponding to the original current time series, obtaining the thermoelectric coupled current response component data.

[0048] A coupled response trajectory or phase trajectory representation is constructed based on thermoelectric coupling current response component data, and a phase trajectory skeleton is extracted from this trajectory to obtain coupled phase skeleton data reflecting the main coupling evolution path. On this basis, topological hierarchical analysis is performed on the coupled response morphology corresponding to the coupled phase skeleton data to extract phase map features. Phase map features may include trajectory loop structure features, branch structure features, curvature hierarchy features, trajectory thickness features, and trajectory evolution stability features, used to characterize the changes in the thermoelectric coupling dynamic response at the structural and topological levels. Based on the phase map features and thermoelectric coupling current response component data, a reverse mapping analysis is performed on the current response details, mapping the coupled trajectory structural information back to the current response detail level to obtain candidate current fingerprint data. Fingerprint aggregation correction is then used to constrain and fuse the candidate current fingerprint data within different time periods, different excitation cycles, or different phase windows, ultimately obtaining the current fingerprint features. This "phase map features first, then current fingerprint features" analysis path improves the stability of current detail feature extraction and sensitivity to early weak anomalies under structural-level feature constraints.

[0049] In this embodiment, a thermally locked phase-locked reference is first constructed based on the temperature rise rate time-series data. Then, the current time-series data is synchronously demodulated and analyzed using this thermally locked phase-locked reference. This allows for the priority separation and enhancement of current coupling components coherent with the dominant frequency band of the thermal response, thereby reducing the impact of non-thermal coupling disturbances, power supply ripple, and random noise on the feature extraction results. On this basis, coupled fingerprint feature analysis is further performed on the thermoelectric coupled current response component data to obtain current fingerprint features and phase map features. This is beneficial for improving the structurality and distinguishability of feature expression, thereby enhancing the ability to identify early weak anomalies and improving the stability of subsequent deviation analysis.

[0050] In an exemplary embodiment, based on thermally locked phase-locked reference data, synchronous demodulation analysis is performed on the current timing data to obtain thermoelectric coupling current response component data, including steps 402 to 408. Wherein:

[0051] Step 402: Frequency shift coding is performed on the thermal phase-locked reference data to obtain thermal reference frequency shift template data.

[0052] Step 404: Based on the thermal reference frequency shift template data, perform heterogeneous frequency mapping analysis on the current time series data to obtain candidate difference frequency response data.

[0053] Step 406: Perform coherent enhancement processing on the candidate difference frequency response data to obtain the corrected difference frequency response data.

[0054] Step 408: Perform inverse mapping demodulation on the corrected difference frequency response data to obtain thermoelectric coupling current response component data.

[0055] Frequency shift coding is a process that involves shifting the dominant frequency band reference frequency and phase information in a thermal phase-locked reference according to a preset mapping rule and then expressing it in a template-based manner.

[0056] The thermal reference frequency shift template data is a reference template data formed by frequency shift encoding, which contains frequency shift mapping relationship and phase constraint relationship and is used for inter-frequency mapping.

[0057] Heterogeneous frequency mapping analysis is an analytical process that uses thermal reference frequency shift template data to perform frequency domain / time domain transformation on current time series data under template constraints, so that thermally coupled related components are clustered and characterized in the difference frequency domain.

[0058] Candidate difference frequency response data are difference frequency response data obtained from heterogeneous frequency mapping analysis, located in the difference frequency domain, and containing thermally coupled related response components.

[0059] Coherent enhancement processing is a process that uses phase consistency and periodic repeatability to screen candidate difference frequency response data, suppress pseudo-coherence, and enhance effective coherent components.

[0060] The corrected difference frequency response data is the difference frequency response data obtained after coherent enhancement processing, which has more stable phase and amplitude and suppresses pseudo-coherent components.

[0061] Inverse mapping demodulation is a process that uses the frequency shift and phase mapping relationship of thermal reference frequency shift template data to map the corrected difference frequency response data back to the original current response representation space in order to recover the coupled response components.

[0062] Specifically, reference frequency, reference phase, phase window boundary, and period marker information corresponding to the dominant frequency band of the thermal response are extracted from the thermal phase-locked reference data. One or more frequency shift mapping target intervals are constructed for the dominant frequency band of the thermal response according to preset frequency shift coding rules, so that components coherent with the thermal response in the subsequent current response can be more stably aggregated in the difference frequency domain. During the frequency shift coding process, frequency shift parameters can be assigned to different dominant frequency bands, and the frequency shift components are time-aligned, phase-offset compensated, and template fragmented organized in conjunction with the phase window boundary, thereby forming thermal reference frequency shift template data with frequency mapping relationships, phase constraint relationships, and period correspondence relationships. The thermal reference frequency shift template data retains the phase evolution information of the dominant frequency band of the thermal response and converts it into a templated expression form that is convenient for cross-frequency mapping processing.

[0063] Using thermal reference frequency shift template data as the heterofrequency mapping benchmark, phase alignment and mapping transformation are performed on current time series data under template constraints, enabling the response components coherent with the thermally locked phase reference in the current time series data to be centrally characterized in the difference frequency domain. In implementation, the current time series data can first be segmented and rearranged or phase rearranged according to the phase window and period marker in the thermal reference frequency shift template data. Then, heterofrequency transformation is performed on the rearranged current components according to the frequency shift mapping relationship. Cross-cycle folding, difference frequency aggregation, or candidate peak cluster extraction are performed on the mapping results within multiple excitation cycles to form candidate difference frequency response data. Simultaneously, for current components that do not meet the template phase consistency, period repeatability, or frequency shift matching conditions, their influence can be reduced through gating suppression, weight weakening, or elimination, thereby obtaining candidate difference frequency response data that can more centrally characterize the thermo-electric coupling-related current response components.

[0064] The phase coherence, amplitude stability, and repetitive response characteristics of candidate difference frequency response data across multiple excitation cycles, multiple phase windows, or multiple candidate difference frequency components are evaluated to identify effective difference frequency response components that maintain coherence with a thermally locked phase reference. Based on this, coherent enhancement processing is performed on the candidate difference frequency response data. This process may include at least one or a combination of phase stability spectrum construction, coherent gated resampling, inverse residual cancellation, and coherent superposition correction. Phase stability spectrum construction characterizes the phase stability of each candidate difference frequency component across multiple cycles; coherent gated resampling prioritizes high-coherence-confidence sampling segments; inverse residual cancellation weakens pseudo-coherent components caused by power supply ripple, random noise, or non-thermal coupling disturbances; and coherent superposition correction aligns and cumulatively enhances the amplitude and phase coherence of the selected effective difference frequency responses. Through this processing, the signal-to-noise ratio and stability of the candidate difference frequency response data can be significantly improved while preserving the thermo-electric coupling master response information, ultimately yielding corrected difference frequency response data.

[0065] Based on the pre-established frequency shift coding relationship, phase mapping relationship, and period correspondence in the thermal reference frequency shift template data, inverse frequency recovery and phase reconstruction are performed on the corrected difference frequency response data, so that the corrected response components in the difference frequency domain are back-mapped to the time-series representation space where the original current response resides. During the inverse mapping demodulation process, the corrected difference frequency response data is first recovered to each phase window and period segment according to the template segment correspondence. Then, combined with the excitation timing marker, the recovered results are time-rearranged, amplitude-scaled, and phase-biased to ensure that the back-mapped response components are consistent with the original current time-series data in terms of time position, phase relationship, and amplitude characteristics. If necessary, boundary smoothing and segment splicing consistency correction can also be performed on the back-mapped response components to reduce the discontinuities caused by template switching and window reconstruction, ultimately obtaining the thermoelectric coupling current response component data.

[0066] In this embodiment, by frequency-shift encoding the thermal phase-locked reference data and constructing thermal reference frequency-shift template data, and then performing inter-frequency mapping analysis, coherent enhancement processing, and inverse mapping demodulation on the current time series data based on this template, the coupled response components related to the dominant frequency band of the thermal response in the original current time series can be aggregated, purified, and stably restored in the difference frequency domain. This improves the separation degree and anti-interference capability of the thermal-electric coupling response component extraction without directly relying on the full-band characteristics of the original current. Compared with conventional direct filtering or synchronous detection methods, this processing link is more conducive to suppressing frequency band overlap interference, periodic pseudo-response, and response drift caused by template switching, thereby improving the accuracy, consistency, and cross-condition applicability of subsequent coupled fingerprint feature analysis.

[0067] In an exemplary embodiment, based on thermal reference frequency shift template data, heterogeneous frequency mapping analysis is performed on current time series data to obtain candidate difference frequency response data, including steps 502 to 510. Wherein:

[0068] Step 502: Perform phase surface encoding on the thermal reference frequency shift template data to obtain template phase surface data.

[0069] Step 504: Based on the template phase surface data, perform surface index mapping on the current time series data to obtain surface-mapped current data.

[0070] Step 506: Perform differential frequency energy folding processing on the surface-mapped current data to obtain candidate differential frequency energy maps.

[0071] Step 508: Perform peak cluster connectivity analysis on the candidate difference frequency energy map to obtain candidate difference frequency component data.

[0072] Step 510: Perform mapping stability analysis on the candidate difference frequency component data to obtain candidate difference frequency response data.

[0073] Phase surface coding is a process of encoding the frequency shift, phase, and period relationships in thermal reference frequency shift template data into a surface-based indexed representation according to preset rules.

[0074] Template phase surface data is surface template data obtained by phase surface encoding, used to characterize the correspondence between frequency shift index, phase evolution, and period position.

[0075] Surface index mapping is a process that maps sampling points in current timing data to the surface coordinate domain according to the phase and period relationship based on template phase surface data.

[0076] Surface-mapped current data is obtained by surface index mapping and represents the distribution of current response on the template phase surface.

[0077] Differential frequency energy folding is a process of cross-cycle merging, rearranging and aggregating the differential frequency related energy in surface-mapped current data to form a unified differential frequency representation.

[0078] Candidate difference frequency energy maps are graphical data obtained after difference frequency energy folding processing, used to characterize the location and intensity distribution of energy accumulation in the difference frequency domain.

[0079] Peak cluster connectivity analysis is an analytical process that merges the connectivity of local peaks and their neighborhoods in candidate difference frequency energy maps and identifies peak cluster structures.

[0080] Candidate difference frequency component data are data extracted from peak cluster connected domain analysis, which characterize the position, energy, and morphological parameters of candidate difference frequency response components.

[0081] Mapping stability analysis is an analytical process that evaluates the consistency and stability of candidate difference frequency component data under multiple periods or multiple mapping conditions.

[0082] Specifically, frequency shift mapping parameters, reference phase parameters, phase window boundaries, period marker information, and template segment identifiers corresponding to different dominant frequency bands are extracted from the thermal reference frequency shift template data. These parameters undergo unified time base calibration and phase continuity preprocessing to eliminate phase jumps and period boundary misalignments at template segment switching points. According to preset phase surface coding rules, discrete template parameters are mapped to the surface coordinate domain according to the joint coordinate relationship of "frequency shift index—phase position—period number," constructing a template phase surface characterizing the frequency shift mapping relationship and phase evolution relationship. This can be achieved using continuous surface fitting, piecewise surface splicing, or gridded surface coding methods. During the coding process, boundary smoothing, phase offset compensation, and surface coherence constraint processing can be applied to the surface sub-regions corresponding to multiple dominant frequency bands to obtain template phase surface data, which exhibits consistent index availability and stable phase reference characteristics during cross-frequency band and cross-period mapping.

[0083] Using template phase surface data as indexing rules and mapping constraints, a joint relocation of sampling points in the current time series data—based on "time position, period position, and phase position"—is performed, mapping the current sample values ​​in the original time series domain to the surface coordinate units corresponding to the template phase surface. In implementation, the current time series data is first segmented and sliced ​​according to the excitation timing marker, template period boundary, and phase window boundary. Then, surface index mapping is performed based on the template phase position, frequency shift index, and period coordinates corresponding to each sampling point, ensuring that the current responses coherent with the thermal reference frequency shift template within multiple excitation cycles are aligned and clustered in the surface space. For mapping holes and discontinuous regions caused by sampling jitter, boundary misalignment, or local missing data, corrections can be made through neighborhood interpolation, surface mesh compensation, gating suppression, or low-weight backfilling to reduce the impact of incoherent components and outliers on the surface mapping results. The final surface-mapped current data is used to characterize the distribution structure, local energy accumulation state, and cross-cycle consistency of the current response on the template phase surface.

[0084] To address the distribution of surface-mapped current data within the template phase surface, normalized folding and difference-frequency domain rearrangement are performed on the response energies under different periods, phase windows, and frequency shift indices. This aims to aggregate the thermally coupled related response energies, originally scattered across multiple periods and segments, into a unified difference-frequency representation space. During the aggregation calculation, the surface-mapped current data undergoes periodic normalization and index merging based on the template period boundaries and frequency shift mapping relationships. Then, accumulation, weighted averaging, or local smoothing are applied to energy components at the same or adjacent difference-frequency positions to enhance stable difference-frequency energy regions that recur across periods, while suppressing discrete energy distributions caused by isolated noise points and random drift. If necessary, the folding weights can be adjusted based on phase window confidence or mapping quality indices to reduce the impact of low-confidence mapping segments on the overall folding result, ultimately yielding candidate difference-frequency energy maps.

[0085] Local energy peaks are identified in the candidate difference frequency energy map, and these peaks are used as seed points to expand the energy distribution of their neighborhoods, forming candidate peak cluster regions. Connectivity analysis is then performed on these candidate peak cluster regions, merging peaks that are continuous or correlated in spatial location, energy distribution, and morphological evolution. Noise peaks, isolated spurious peaks, and unstable peak clusters are eliminated, split, or downweighted based on preset constraints (such as energy threshold, connected region area, peak cluster compactness, peak cluster extension direction, and inter-peak distance). After peak cluster screening, difference frequency position parameters, peak energy parameters, peak cluster area parameters, morphological description parameters, and correlation parameters with the template phase surface are extracted from the retained connected region peak clusters to form candidate difference frequency component data.

[0086] To address the recurrence of candidate difference frequency components across multiple excitation cycles, phase windows, mapping batches, or template segments, a comprehensive evaluation is conducted on the difference frequency position drift, energy intensity fluctuation, phase correlation consistency, and peak cluster structure retention of each candidate difference frequency component. This evaluation aims to determine whether the component belongs to the effective difference frequency response component formed by stable mapping of the thermal reference frequency shift template. Typically, the comprehensive evaluation involves calculating position stability, energy stability, phase consistency, and morphological stability indices. Candidate difference frequency components are then screened, reweighted, sorted, or merged according to preset stability rules. Components that appear only in a few cycles, exhibit significant position drift, or undergo drastic morphological changes are identified as pseudo-mapping components and suppressed. Finally, candidate difference frequency components that meet the conditions for multidimensional stability indices are retained, and candidate difference frequency response data is generated accordingly.

[0087] In this embodiment, by further encoding the thermal reference frequency shift template data into template phase surface data, and based on this, performing surface index mapping, difference frequency energy folding, peak cluster connectivity analysis, and mapping stability analysis on the current time series data, the dispersed and noise-sensitive thermal coupling response information in the original time series domain can be converted into a difference frequency energy distribution and component-level representation with spatial structure constraints. This introduces phase continuity, peak cluster structure, and cross-cycle stability constraints simultaneously during the candidate difference frequency response extraction stage. Compared to methods that rely solely on single-point peak values ​​or single difference frequency energy thresholds for screening, this approach is more effective in suppressing false detections caused by local abnormal peaks, isolated pseudo-responses, and mapping drift, improving the structural integrity, stability, and repeatability of the candidate difference frequency response data, and providing higher-quality input for subsequent coherent enhancement processing and coupling feature extraction.

[0088] In an exemplary embodiment, coherent enhancement processing is performed on the candidate difference frequency response data to obtain corrected difference frequency response data, including steps 602 to 608. Wherein:

[0089] Step 602: Construct the phase stability spectrum of the candidate difference frequency response data to obtain phase stability spectrum data.

[0090] Step 604: Based on the phase stability spectrum data, coherently gated resampling is performed on the candidate difference frequency response data to obtain gated difference frequency response data.

[0091] Step 606: Perform inverse residual cancellation on the gated differential frequency response data to obtain residual corrected differential frequency response data.

[0092] Step 608: Perform coherent superposition correction on the residual correction difference frequency response data to obtain the corrected difference frequency response data.

[0093] Phase stability is the degree to which the phase of a candidate difference frequency response remains consistent and drift is controlled across multiple excitation cycles, phase windows, or time segments.

[0094] Spectrum construction is the process of organizing phase stability into a spectral distribution representation in the dimensions of difference frequency index, time segment, or candidate component.

[0095] Phase stability spectrum data is data obtained by constructing a spectrum and used to characterize the phase stability distribution and strength relationship of candidate difference frequency responses.

[0096] Coherent gated resampling is a process that uses phase stability spectrum data to gate and resample candidate difference frequency responses in order to prioritize the retention of components with high coherence confidence.

[0097] Gated difference frequency response data is difference frequency response data obtained after coherent gating resampling, which suppresses low coherence components and highlights high coherence components.

[0098] Anti-phase residual cancellation is a process of constructing residual components that are out of phase with the residual pseudo-coherence or background perturbation components and then superimposing and canceling them out.

[0099] The residual-corrected difference frequency response data is the difference frequency response data obtained after the inverse residual is canceled, and the pseudo-coherent residual is further weakened.

[0100] Coherent superposition correction is a process of aligning amplitudes, weighting and superimposing, and correcting consistency of effective coherent components in residual correction difference frequency response data.

[0101] Specifically, candidate difference frequency response data are hierarchically grouped according to excitation period, phase window, difference frequency component index, and / or time segment. For each group, phase sequence, phase change rate, and phase drift trajectory are extracted to characterize the phase preservation characteristics of the candidate difference frequency response under multi-cycle repetitive excitation conditions. Then, stability is calculated on the phase sequence and phase drift trajectory. The stability calculation may include at least one or a combination of phase dispersion statistics, phase drift amplitude statistics, periodic phase consistency assessment, and inter-window phase continuity assessment. It may also incorporate joint weighting based on difference frequency position, response energy intensity, and peak cluster morphology parameters to enhance the contribution of high-confidence difference frequency components to the stability assessment results. Based on this, the phase stability results corresponding to each difference frequency position, each candidate component, or each time segment are organized into a spectral representation according to a preset index relationship, forming phase stability spectrum data.

[0102] Based on the phase stability score, phase dispersion distribution, and difference frequency position correlation information of each candidate difference frequency component in the phase stability spectrum data, a corresponding coherent gating strategy is generated, and gated resampling processing is performed on the candidate difference frequency response data. Specifically, difference frequency response segments with high phase stability and good period repeatability are assigned higher retention weights or increased sampling density, while difference frequency response segments with low phase stability, significant drift, or insufficient repeatability are subjected to reduced-weight sampling, sparse sampling, or gated suppression. During the resampling process, alignment can be performed using periodic boundaries and phase window boundaries to maintain consistency in the time position and difference frequency index of the gated difference frequency response, ultimately obtaining the gated difference frequency response data.

[0103] For the gated difference frequency response data after coherent gated resampling, the remaining non-thermally coupled pseudo-coherent components, background disturbance residuals, and residual difference frequency components introduced by periodic power supply ripple or sampling jitter are further identified. Based on their recurrence characteristics across multiple excitation cycles and phase windows, corresponding residual estimation components are constructed. Inverse-phase residual components with opposite phases or satisfying cancellation conditions are generated according to the amplitude, phase, and effective range of the residual estimation components. These inverse-phase residual components are then segmented and superimposed with the gated difference frequency response data to cancel out residual responses that do not conform to the thermally locked phase-locked reference coherence characteristics. During the cancellation process, differentiated cancellation strengths are set for different difference frequency components using phase stability spectrum data and gate weight information. Over-cancellation suppression constraints, local amplitude lower limit constraints, and boundary smoothing processing can be introduced to avoid erroneous weakening of the effective thermoelectrically coupled difference frequency response or the generation of new discontinuities at segment boundaries. After inverse-phase residual cancellation, residual-corrected difference frequency response data is obtained.

[0104] Amplitude alignment is performed on effective response segments of the residual-corrected difference frequency response data in different excitation periods, phase windows, and difference frequency component channels to eliminate the influence of periodic phase shift, amplitude scale differences, and local time misalignment on the superposition results. After alignment, superposition weights are assigned to each response segment according to phase stability spectrum data, coherent gated resampling results, and reliability information after residual cancellation. Coherent superposition is then performed to enhance the stable thermoelectric coupling difference frequency response components, while noise and spurious responses are reduced through mutual cancellation of random residuals. After superposition, amplitude normalization, phase offset correction, local discontinuity smoothing, and segment splicing consistency correction are performed on the results to ensure that the output data meets the inverse mapping demodulation requirements in terms of difference frequency response shape, phase continuity, and period consistency, ultimately yielding the corrected difference frequency response data.

[0105] In this embodiment, by first constructing a phase stability spectrum from the candidate difference frequency response data, then implementing coherent gating resampling based on the phase stability spectrum, and further combining anti-phase residual cancellation and coherent superposition correction, the "stable coherent components" and "pseudo-coherent residuals" can be processed separately at the difference frequency response level, thereby avoiding the effective response attenuation problem caused by relying solely on single-stage threshold filtering or simple averaging. Among these, the phase stability spectrum provides a quantifiable basis for gating and weighting, anti-phase residual cancellation is beneficial for targeted suppression of periodic pseudo-responses, and coherent superposition correction is beneficial for enhancing the real coupling components that recur across periods. Therefore, the purity, phase consistency, and periodic stability of the corrected difference frequency response data can be further improved, providing higher quality input for subsequent inverse mapping demodulation and coupling feature extraction.

[0106] In an exemplary embodiment, coupled fingerprint feature analysis is performed on the thermoelectric coupled current response component data to obtain current fingerprint features and phase map features, including steps 702 to 708. Wherein:

[0107] Step 702: Extract the phase trajectory skeleton from the thermoelectric coupling current response component data to obtain the coupled phase skeleton data.

[0108] Step 704: Based on the coupled phase skeleton data, perform topological hierarchical analysis on the coupled response morphology to obtain phase map features.

[0109] Step 706: Based on the phase diagram features and thermoelectric coupling current response component data, perform reverse mapping analysis on the current response details to obtain candidate current fingerprint data.

[0110] Step 708: Perform fingerprint aggregation and correction on the candidate current fingerprint data to obtain current fingerprint features.

[0111] Phase trajectory skeleton extraction is a process of extracting the skeleton structure that characterizes the main phase evolution path from the phase trajectory corresponding to the thermoelectric coupling current response component data.

[0112] Coupled phase skeleton data is obtained by extracting the phase trajectory skeleton and is used to characterize the dominant phase structure and key transition relationships of thermoelectric coupling response.

[0113] The coupled response morphology is the structural morphology, hierarchical relationship, and evolution mode exhibited by the thermoelectric coupled current response components in the phase trajectory space.

[0114] Topological hierarchical analysis is an analytical process that divides the coupled response morphology into hierarchical levels and extracts structural features based on the connectivity, branching structure, and closed-loop structure of coupled phase skeleton data.

[0115] Current response details are fine-grained response information in thermoelectric coupled current response components that reflect local amplitude changes, ripple morphology, response transitions, and periodic microstructures.

[0116] Back-mapping analysis is an analytical process that back-maps coupled structure information to the current response detail level under phase map feature constraints in order to recover and extract detailed features.

[0117] Candidate current fingerprint data are candidate fingerprint data obtained through reverse mapping analysis and used to characterize the detailed features of current response.

[0118] Fingerprint aggregation correction is a process that aligns, weights, fuses, and corrects the consistency of multiple candidate current fingerprint data to form stable current fingerprint features.

[0119] Specifically, a phase trajectory representation is constructed based on thermoelectric coupling current response component data, uniformly mapping the phase evolution process in each excitation cycle, phase window, or response segment to the phase trajectory coordinate space. Skeleton extraction processing is performed on the phase trajectory, which may include at least one or a combination of trajectory denoising, local smoothing, trajectory refinement, main path preservation, and redundant branch suppression to extract the central skeleton structure representing the main coupling evolution trend from the original trajectory containing minor fluctuations and local perturbations. During the extraction process, skeleton nodes and segments are screened and corrected by incorporating information on trajectory curvature changes, phase inflection point distribution, trajectory connectivity, and trajectory density to avoid the generation of pseudo-skeletons caused by noise spikes, transient jitter, or local anomalies, ultimately obtaining coupled phase skeleton data.

[0120] The skeleton structure is initially layered based on the number of connected branches, closed-loop structure distribution, primary / secondary branch relationships, and node degree information of the coupled phase skeleton data. Then, each layer is further refined by combining skeleton segment curvature, path length, local thickness mapping, and branch merging / splitting positions, thereby distinguishing stable dominant coupling paths, locally perturbed branches, and transitional response structures. Simultaneously, the connection relationships between different topological layers, the evolutionary order between layers, and the recurring topological patterns across periods are statistically analyzed to extract feature parameters reflecting the stability, complexity, and evolutionary laws of the coupled response structure. Finally, phase map features are obtained, which may include closed-loop structure features, branch structure features, layer depth features, curvature layer features, and topological evolution stability features.

[0121] Based on the hierarchical structure, closed-loop structure, branch structure, and curvature hierarchy information in the phase map features, the structural positioning intervals and phase alignment references for current response details are determined. Then, combined with thermoelectric coupling current response component data, back-mapping reconstruction is performed on the amplitude variations, local ripple morphology, response transition characteristics, and periodic details within each structural positioning interval. During the back-mapping process, phase constraint compensation, timing rearrangement, and local amplitude correction are applied to detail deviations caused by sampling jitter, local noise, or segment misalignment to ensure that the recovered current response details are consistent with the coupling evolution structure represented by the phase map. Finally, candidate current fingerprint data is obtained, which is used to characterize the current dynamic detail features extracted under the structural constraints of the phase map.

[0122] Candidate current fingerprint data are grouped and quality evaluated based on the structural hierarchy information and inverse mapping quality information corresponding to the phase map features. Then, time alignment, phase alignment, amplitude normalization, and local noise suppression are performed on each group of candidate current fingerprint data to eliminate feature shifts caused by sampling bias, segment misalignment, and local perturbations. Next, aggregation weights are assigned based on the phase consistency, energy stability, structural integrity, and cross-cycle repeatability of the candidate current fingerprint data. Multiple groups of candidate current fingerprint data are then weighted, fused, and their consistency corrected. Local morphological smoothing and abnormal segment suppression are applied to the aggregation results to avoid contamination of the final features by low-quality segments, ultimately yielding the current fingerprint features.

[0123] In this embodiment, phase map features are obtained by first extracting the phase trajectory skeleton and performing topological hierarchical analysis on the thermoelectric coupling current response component data. Then, under the constraints of the phase map structure, the current response details are reverse-mapped and corrected by fingerprint aggregation to obtain current fingerprint features. The extraction of current detail features is based on the prior constraints of the coupled response topology, thereby avoiding the problems of local noise, transient jitter and fragment misalignment when directly extracting features from the original response details. Compared with feature extraction methods based solely on time-domain or frequency-domain statistics, this scheme is more conducive to maintaining the consistency between current detail features and thermoelectric coupling evolution structure, improving the structural interpretability, cross-cycle repeatability and ability to distinguish weak anomalous patterns of current fingerprint features, and providing more stable and discriminative feature input for subsequent deviation analysis.

[0124] In an exemplary embodiment, deviation analysis is performed on the current fingerprint features and phase map features to obtain the abnormal deviation, including steps 802 to 806. Wherein:

[0125] Step 802: Couple manifold expansion is performed on the current fingerprint features and phase map features to obtain manifold offset representation data.

[0126] Step 804: Based on the manifold offset representation data, perform inverse constraint matching analysis on the health response reference model corresponding to the target eMMC to obtain the matching residual spectrum data.

[0127] Step 806: Perform inverse constraint backfeed verification on the matched residual spectrum data to obtain the abnormal deviation degree.

[0128] Coupled manifold expansion is a process of mapping and expanding the coupling relationship between current fingerprint features and phase map features in manifold space to form a representation that can characterize the offset relationship.

[0129] Manifold offset representation data is obtained by expanding coupled manifolds and is used to characterize the offset of the current sample relative to the healthy coupled response manifold in terms of direction, magnitude, and local structure.

[0130] The health response reference model is a reference model for deviation analysis built based on the thermoelectric coupling response characteristics of healthy eMMC under various operating conditions.

[0131] Inverse constraint matching analysis is an analytical process that uses manifold offset representation data as constraint input to perform reverse matching on a health response reference model in order to obtain the matching residuals corresponding to the current sample.

[0132] Matching residual spectrum data is obtained from inverse constraint matching analysis and is used to characterize the mismatch distribution between the current sample and the health response reference model at different constraint dimensions or structural levels.

[0133] Inverse constraint backfeed verification is a process of backfeeding the matched residual spectrum data into the constraint space of the healthy response reference model for consistency verification and convergence verification.

[0134] Specifically, the current fingerprint and phase map features are preprocessed to remove noise and outliers, and then standardized and normalized so that they can be compared in a unified manifold space. Manifold learning methods (such as Principal Component Analysis (PCA) or t-SNE algorithm) are used to embed these features into a high-dimensional manifold space. Manifold unfolding projects these features from the original space to a lower-dimensional space to capture the nonlinear relationships between them, resulting in manifold offset representation data. This data, by preserving the global structure and local feature information of the data, can reveal the coupling relationship between the current fingerprint and phase map, helping to discover potential patterns and biases between features.

[0135] Using manifold offset representation data as input for inverse constraint matching, and combining it with the pre-stored health state manifold distribution, feature constraint relationships, and operating condition stratification parameters in the health response reference model corresponding to the target eMMC, the matching region of the current sample in the health response constraint space is searched and filtered. Then, using the offset direction, offset magnitude, and local structural relationships in the manifold offset representation data as constraints, inverse matching is performed on the health response reference model, transforming the health model from a "compared object" into a "reversely activated and back-substituted generation object," thereby generating the candidate health response matching result closest to the current sample. In this process, the matching error is decomposed according to the constraint type (such as current fingerprint constraint, phase map structure constraint, and operating condition consistency constraint), and the mismatch terms under different constraint levels are reorganized and spectrally organized to form matching residual spectrum data.

[0136] For the matched residual spectrum data, it is backfeeded into the constraint space corresponding to the healthy response reference model according to the preset constraint level, residual component type, or spectral range. Inverse constraint backfeeding verification is then performed to verify whether the residual spectrum can be explained, absorbed, or converged within the healthy constraint system. During the backfeeding process, component weights, structural consistency constraints, and convergence threshold constraints are applied to different residual spectrum components. The impact of each residual component on the closed-loop stability of the healthy model is evaluated through iterative backfeeding, constraint reprojection, and consistency verification. This distinguishes between normal deviations that can be explained by operating condition fluctuations and measurement disturbances, and abnormal deviations that are difficult to converge within the healthy constraint space. Then, based on the degree of residual convergence after backfeeding, the strength of constraint conflicts, the persistence of spectral mismatch, and the range of structural mismatch, the matched residual spectrum data is comprehensively verified and quantified to obtain the degree of abnormal deviation.

[0137] In this embodiment, manifold expansion is performed by coupling current fingerprint features and phase map features to obtain manifold offset representation data. Then, based on this manifold offset representation data, inverse constraint matching analysis and inverse constraint refeed verification are performed on the health response reference model corresponding to the target eMMC. The anomaly identification process is transformed from the traditional direct distance comparison into a structured deviation discrimination process of "manifold offset - constraint matching - refeed verification". This process can not only characterize the magnitude of feature deviation, but also characterize the mismatch pattern of deviation in constraint level, structural dimension and spectral distribution. Therefore, compared with a single distance threshold or simple classification judgment method, this scheme is more conducive to distinguishing between interpretable deviations caused by operating condition fluctuations and non-convergent deviations caused by potential defects, and improves the discrimination accuracy, robustness and interpretability of anomaly deviation calculation.

[0138] In an exemplary embodiment, based on manifold offset representation data, an inverse constraint matching analysis is performed on the health response reference model corresponding to the target eMMC to obtain matching residual spectrum data, including steps 902 to 908. Wherein:

[0139] Step 902: Expand the manifold offset representation data by offset anchor points to obtain inverse constraint probe data.

[0140] Step 904: Based on the inverse constraint probe data, perform reverse activation matching analysis on the health response reference model to obtain candidate health response back-substitution data.

[0141] Step 906: Perform constraint conflict reorganization on the candidate health response back-substitute data to obtain constraint conflict residual data.

[0142] Step 908: Perform spectral decomposition on the constraint conflict residual data to obtain the matching residual spectral data.

[0143] Offset anchor point unwrapping is a process of identifying key offset anchor points from manifold offset representation data and generating a probe set covering the key offset structure according to preset unwrapping rules.

[0144] The inverse constraint probe data is the inverse constraint input data obtained by expanding the offset anchor points and used to drive the health response reference model for inverse activation matching analysis.

[0145] Reverse activation matching analysis is an analytical process that uses inverse constraint probe data as input to selectively activate and reverse match the health response reference model in order to generate candidate health response interpretation results.

[0146] The candidate health response back-substitution data is generated by back-substitution from the health response reference model through reverse activation matching analysis, and is the candidate health response data that is closest to the current target eMMC response.

[0147] Constraint conflict reorganization is a process of reconstructing and grouping mismatched terms in candidate health response back-substitution data that cannot meet the constraints of the health model according to conflict type, intensity, or structural relationship.

[0148] Constraint conflict residual data is residual data obtained by constrain conflict reorganization and used to characterize the mismatch part of the candidate health response back-substitution data within the health constraint system.

[0149] Spectral decomposition is a process of representing and decomposing conflicting residual data into spectral data and components according to a preset decomposition dimension to form matching residual spectral data.

[0150] Specifically, the manifold migration representation data undergoes migration structure analysis to identify key migration regions, local extrema, structural inflection points, and hierarchical boundary points that represent deviations from healthy manifold characteristics. These key points are then designated as migration anchor points. Next, migration anchor point expansion is performed around each anchor point according to preset expansion rules. These rules can be based on the manifold's local neighborhood structure, migration direction continuity, migration magnitude gradient, and hierarchical topology, expanding the anchor points' neighborhoods, extending paths, or generating anchor point clusters to form a set of inverse constraint probes covering the key migration structures. During the expansion process, redundancy removal constraints, inter-probe conflict constraints, and structural integrity constraints are applied to the probe data to avoid overly dense probe distribution, conflicting probe directions, or omissions of key migration paths; ultimately, inverse constraint probe data is obtained.

[0151] Using inverse constraint probe data as inverse activation input, selective activation is performed on the response subspace, constraint level, or operating condition hierarchical unit in the health response reference model according to the offset anchor point position, offset direction, offset magnitude, and structural hierarchy information corresponding to the probe, to determine the candidate regions of health response most relevant to the current target eMMC response. Then, inverse matching is performed within the activated candidate regions, applying the inverse constraint probe data as inverse constraint conditions to the health response reference model, transforming the model from a "health response prior library" into a "back-substitution generator driven by inverse constraints," and generating candidate health response segments or candidate health response states that are closest to the inverse constraint probe data in terms of structural relationships and offset trends. During this process, operating condition consistency constraints, topology consistency constraints, and current fingerprint / phase map coupling constraints are combined to filter, reduce weights, or reorder candidate matching paths to suppress false matches caused by operating condition mismatches or local noise. Finally, the selected candidate health response segments are back-substituted and spliced ​​or the states are back-substituted and recombined to obtain candidate health response back-substitution data.

[0152] To establish the correspondence between candidate health response back-substitution data and inverse constraint probe data, manifold migration representation data, and the constraint set of the health response reference model, consistency verification of the back-substitution results at different constraint levels is performed to identify conflict terms in current fingerprint constraints, phase map topology constraints, operating condition consistency constraints, and structural evolution constraints. The identified conflict terms are grouped and reorganized according to conflict type, conflict intensity, effective range, and structural correlation, reconstructing the mismatch information originally scattered across different feature dimensions, time segments, or topology levels into an analyzable set of constraint conflict residuals. During the reorganization process, weak conflict terms caused by local noise, transient disturbances, or boundary reconstruction errors are reduced in weight and suppressed, while strong conflict terms that repeatedly occur across levels or persist on critical structural paths are enhanced and marked to avoid excessive influence of non-critical errors on subsequent analysis results. Finally, constraint conflict residual data are obtained.

[0153] Based on the organizational structure of the constraint conflict residual data, a residual decomposition index is established. Then, at least one of the following methods is performed on each conflict residual component: frequency domain decomposition, time-frequency decomposition, hierarchical decomposition, or component decomposition, to extract the energy distribution, persistence characteristics, and structural correlation characteristics of different conflict sources in the spectral bands. At the same time, the spectral decomposition results are weighted and sorted by conflict intensity weight, structural importance weight, and operating condition consistency weight to highlight the key conflict spectral bands that are more sensitive to the determination of abnormal deviations, and to suppress low-confidence residual spectral components caused by occasional disturbances, finally obtaining matched residual spectral data.

[0154] In this embodiment, by further converting the manifold offset representation data into inverse constraint probe data, and performing reverse activation matching analysis, constraint conflict recombination, and spectral decomposition on the health response reference model based on this inverse constraint probe data, the deviation information originally scattered in the feature space can be transformed into a mismatch characterization process with a chain structure of "probe-driven—back-substitution interpretation—conflict extraction—spectral expression". This not only improves the sensitivity of the health model matching process to the response of key offset structures, but also allows the mismatch information to be output in the form of constraint conflict residuals and matching residual spectra in a structured manner. Compared with the method of directly performing overall distance fitting or coarse-grained residual calculation on manifold offset, this scheme is more conducive to locating the source of deviation, distinguishing the conflict contribution of different constraint levels, and improving the refinement and interpretability of subsequent inverse constraint backfeed verification of abnormal deviation calculation.

[0155] In an exemplary embodiment, based on the inverse constraint probe data, a reverse activation matching analysis is performed on the health response reference model to obtain candidate health response back-substitution data, including steps 1002 to 1008. Wherein:

[0156] Step 1002: Activate path encoding is performed on the inverse constraint probe data to obtain inverse path template data.

[0157] Step 1004: Based on the reverse path template data, perform hidden layer state reverse wake-up analysis on the health response reference model to obtain candidate hidden layer response state data.

[0158] Step 1006: Perform state folding inversion and recombination on the candidate hidden layer response state data to obtain candidate healthy response back-substitution data.

[0159] Step 1008: Perform inverse constraint reprojection verification on the candidate health response back-substitute data to obtain candidate health response back-substitute data that meets the inverse constraint matching conditions.

[0160] Activation path encoding encodes the offset information in the inverse constraint probe data according to path rules, forming path template data that describes the activation direction, magnitude, and hierarchical relationship.

[0161] The reverse path template data is template data obtained through activation path encoding, used to describe the reverse activation path from the target eMMC offset feature to the health response reference model.

[0162] Hidden state reverse wake-up analysis is based on reverse path template data to reverse activate the hidden state of the health response reference model, and analyzes the activation status of state nodes and response paths at each layer.

[0163] Candidate hidden state response data is generated during the hidden state reverse wake-up analysis process, and is the candidate hidden state data that is closest to the target eMMC response.

[0164] State folding inversion and recombination involves folding and recombinating candidate hidden response state data to integrate multiple scattered state information into a consistent healthy response back-substitution data.

[0165] The candidate health response back-substitution data is obtained through state folding inversion and recombination, and serves as a reference model for interpreting the current target eMMC response.

[0166] Inverse constraint reprojection verification is a process of backprojecting candidate health response data back to the constraint space of the health response reference model for consistency verification and correction.

[0167] Specifically, the probe positions, offset directions, offset magnitudes, structural hierarchy identifiers, and inter-probe relationships in the inverse constraint probe data are analyzed. Based on the constraint hierarchy structure and response state organization of the health response reference model, each inverse constraint probe is mapped to a corresponding inverse activation path element. The path elements are sequentially encoded, branched encoded, and hierarchically encoded according to preset activation path encoding rules to form inverse path template data that represents the relationship between "probe trigger position—constraint propagation direction—candidate response path." During the encoding process, conflicting, redundant, or low-confidence path elements are merged, downweighted, or removed, and key path nodes are prioritized to generate inverse path template data.

[0168] The reverse path template data is used as the input for reverse activation control, applied to the hidden state space or hierarchical response units of the health response reference model. Selective reverse wake-up is performed on hidden state nodes, state transition paths, and constraint propagation links that match the reverse path template. During the wake-up process, based on the hierarchical encoding, branch encoding, and priority information in the path template, the candidate hidden state regions in the health response reference model most relevant to the current target eMMC offset features are triggered layer by layer. The awakened state responses are then screened and scored in conjunction with operating condition consistency constraints, topology consistency constraints, and coupling feature matching constraints. For hidden states with low matching degree to the reverse path template, significant constraint conflicts, or activated only under local noise, suppression, weight reduction, or path truncation can be performed, ultimately outputting candidate hidden response state data.

[0169] To address the distribution of candidate hidden layer response state data across different levels, path branches, and constraint subspaces, a state folding inversion and reconstruction process is performed to reconstruct the dispersed hidden layer state representations into candidate healthy responses that can be back-substituted into the observation feature space. This state folding inversion and reconstruction can include at least one or a combination of state path folding, cross-layer state alignment, branch state merging, and back projection reconstruction. Multiple hidden layer candidate states are folded and merged along the propagation direction indicated by the back path template, and then back-mapped and reconstructed under operating condition constraints and topological constraints to form a healthy response interpretation result that most closely approximates the current target eMMC sample in terms of structural trend. During the reconstruction process, local smoothing, weight adjustment, and boundary correction are performed on state conflict segments, low-confidence state fragments, or transitional discontinuities to avoid distortion of the back-substitution results caused by discontinuous hidden layer state switching, ultimately yielding candidate healthy response back-substitution data.

[0170] The candidate health response back-substitute data is reprojected into the constraint space corresponding to the inverse constraint probe data and the reverse path template constraint space to perform inverse constraint reprojection verification. This verifies whether the candidate health response back-substitute data can maintain structural consistency and path closure consistency under the inverse constraint system. During the verification process, the consistency of offset direction, offset magnitude matching, topological correspondence, and path propagation continuity between the candidate health response back-substitute data and the inverse constraint probe data must be reprojected for verification. Local segments that do not meet the verification conditions are marked, downweighted, or removed. If necessary, one or more rounds of local reprojection correction and consistency review are performed on the candidate health response back-substitute data to improve the interpretability and stability of the back-substitute results under inverse constraint conditions. Finally, the candidate health response back-substitute data after inverse constraint reprojection verification (i.e., candidate health response back-substitute data that meets the inverse constraint matching conditions) is output.

[0171] In this embodiment, by encoding the activation path of the inverse constraint probe data and forming inverse path template data, and then performing hidden state inverse wake-up analysis, state folding inversion and recombination, and inverse constraint reprojection verification on the health response reference model based on the template, the candidate interpretation process inside the health model can be upgraded from "static matching" to a closed-loop generative matching process of "path-driven hidden wake-up - back-substitution recombination - reprojection verification". This makes the candidate health response back-substitution data not only close to the target sample in surface features, but also has higher interpretability and credibility in terms of hidden state evolution path and inverse constraint consistency. Compared with the method of directly retrieving similar responses from the health sample library or performing one-time fitting back-substitution, this scheme is more conducive to suppressing local mismatches, improving the structural integrity and path consistency of candidate health response back-substitution data, and providing a more stable input basis for subsequent constraint conflict recombination and matching residual spectrum construction.

[0172] In one exemplary embodiment, an eMMC potential defect detection system based on temperature rise rate and current fingerprint is provided. The system includes a server and a terminal. The server is used to perform controlled read / write load perturbation stimulation on the target eMMC to obtain detection process data; the detection process data is acquired through the terminal and transmitted from the terminal to the server via the network. The server is used to perform thermal response deconvolution processing on the detection process data to obtain current time series data and temperature rise rate time series data. The server is used to perform thermoelectric coupling dynamic feature analysis on current time series data and temperature rise rate time series data to obtain current fingerprint features and phase map features. The server is used to perform deviation analysis on current fingerprint features and phase map features to obtain abnormal deviations. The server is used to determine defects based on abnormal deviations and obtain potential defect detection results for the target eMMC.

[0173] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. This computer device includes a processor, memory, input / output interfaces (I / O), and communication interfaces.

[0174] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0175] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0176] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0177] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps in the above method embodiments.

[0178] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0179] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0180] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0181] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for detecting potential defects in eMMC based on temperature rise rate and current fingerprint, characterized in that, The method includes: The target eMMC is subjected to controlled read / write load perturbation stimulation to obtain detection process data; The detection process data is subjected to thermal response deconvolution processing to obtain current time series data and temperature rise rate time series data; Thermoelectric coupling dynamic feature analysis is performed on the current time series data and the temperature rise rate time series data to obtain current fingerprint features and phase map features; Deviation analysis is performed on the current fingerprint features and the phase map features to obtain the abnormal deviation. The abnormal deviation is used to determine the defect, and the potential defect detection results of the target eMMC are obtained.

2. The method according to claim 1, characterized in that, The thermoelectric coupling dynamic feature analysis of the current time series data and the temperature rise rate time series data is performed to obtain current fingerprint features and phase map features, including: Based on the temperature rise rate time series data, a phase-locked reference is constructed for the dominant frequency band of thermal response to obtain thermal phase-locked reference data. Based on the thermal lock-in reference data, the current timing data is synchronously demodulated and analyzed to obtain thermoelectric coupling current response component data; The thermoelectric coupled current response component data is subjected to coupled fingerprint feature analysis to obtain the current fingerprint feature and the phase map feature.

3. The method according to claim 2, characterized in that, The step of synchronously demodulating and analyzing the current timing data based on the thermal lock-in reference data to obtain thermoelectric coupling current response component data includes: The thermally locked phase-locked reference data is frequency-shift encoded to obtain thermal reference frequency-shift template data; Based on the thermal reference frequency shift template data, the current time series data is subjected to heterogeneous frequency mapping analysis to obtain candidate difference frequency response data; The candidate difference frequency response data is subjected to coherent enhancement processing to obtain corrected difference frequency response data; The corrected difference frequency response data is inversely mapped and demodulated to obtain the thermoelectric coupling current response component data.

4. The method according to claim 3, characterized in that, The step of performing heterogeneous frequency mapping analysis on the current time series data based on the thermal reference frequency shift template data to obtain candidate difference frequency response data includes: Phase surface encoding is performed on the thermal reference frequency shift template data to obtain template phase surface data; Based on the template phase surface data, the current time series data is subjected to surface index mapping to obtain surface-mapped current data; The surface-mapped current data is subjected to difference frequency energy folding processing to obtain candidate difference frequency energy maps; Peak cluster connectivity analysis is performed on the candidate difference frequency energy map to obtain candidate difference frequency component data; Mapping stability analysis is performed on the candidate difference frequency component data to obtain the candidate difference frequency response data.

5. The method according to claim 3, characterized in that, The step of performing coherent enhancement processing on the candidate difference frequency response data to obtain corrected difference frequency response data includes: The phase stability of the candidate difference frequency response data is constructed to obtain phase stability spectrum data; Based on the phase stability spectrum data, coherent gated resampling is performed on the candidate difference frequency response data to obtain gated difference frequency response data; The gated difference frequency response data is subjected to inverse residual cancellation to obtain residual corrected difference frequency response data; The residual correction difference frequency response data is coherently superimposed to obtain the correction difference frequency response data.

6. The method according to claim 2, characterized in that, The process of performing coupled fingerprint feature analysis on the thermoelectric coupled current response component data to obtain the current fingerprint features and the phase map features includes: Phase trajectory skeleton extraction is performed on the thermoelectric coupling current response component data to obtain coupled phase skeleton data; Based on the coupled phase skeleton data, a topological hierarchical analysis is performed on the coupled response morphology to obtain the phase map features; Based on the phase diagram features and the thermoelectric coupling current response component data, a reverse mapping analysis is performed on the current response details to obtain candidate current fingerprint data. The candidate current fingerprint data is subjected to fingerprint aggregation and correction to obtain the current fingerprint features.

7. The method according to claim 1, characterized in that, The deviation analysis of the current fingerprint features and the phase map features to obtain the abnormal deviation includes: The current fingerprint features and the phase map features are coupled and expanded to obtain manifold offset representation data. Based on the manifold offset representation data, inverse constraint matching analysis is performed on the health response reference model corresponding to the target eMMC to obtain matching residual spectrum data; The abnormal deviation degree is obtained by performing inverse constraint backfeed verification on the matched residual spectrum data.

8. The method according to claim 7, characterized in that, The step of performing inverse constraint matching analysis on the health response reference model corresponding to the target eMMC based on the manifold offset representation data to obtain matching residual spectrum data includes: The manifold offset representation data is expanded by offset anchor points to obtain inverse constraint probe data; Based on the inverse constraint probe data, reverse activation matching analysis is performed on the health response reference model to obtain candidate health response back-substitution data; The candidate health response back-substitution data is reorganized to obtain constraint conflict residual data; The constraint conflict residual data is subjected to spectral decomposition to obtain the matching residual spectral data.

9. The method according to claim 8, characterized in that, The step of performing reverse activation matching analysis on the health response reference model based on the inverse constraint probe data to obtain candidate health response back-substitution data includes: The inverse constraint probe data is activated by path encoding to obtain inverse path template data; Based on the reverse path template data, the hidden layer state reverse wake-up analysis is performed on the health response reference model to obtain candidate hidden layer response state data. The candidate hidden layer response state data is folded and inverted to reconstruct the data, and the candidate healthy response back-substitution data is obtained. The candidate health response back-substitution data is then subjected to inverse constraint reprojection verification to obtain the candidate health response back-substitution data.

10. A latent defect detection system for eMMC based on temperature rise rate and current fingerprint, characterized in that, The system includes a server and a terminal; The server is used to perform controlled read / write load perturbation stimulation on the target eMMC to obtain detection process data; The detection process data is acquired through the terminal, and the terminal transmits the data to the server via the network. The server is used to perform thermal response deconvolution processing on the detection process data to obtain current time-series data and temperature rise rate time-series data. The server is used to perform thermoelectric coupling dynamic feature analysis on the current time series data and the temperature rise rate time series data to obtain current fingerprint features and phase map features. The server is used to perform deviation analysis on the current fingerprint features and the phase map features to obtain the abnormal deviation. The server is used to determine the defects based on the abnormal deviation and obtain the potential defect detection results of the target eMMC.

Images