AI spectrum monitoring and early warning system and method for power key device performance degradation

Abstract

This invention discloses an AI-based spectral monitoring and early warning system and method for performance degradation of key power supply components, relating to the field of power supply component performance monitoring technology. The system includes the following steps: Under high-temperature, high-carrier excitation conditions, a phase-continuous mapping field characterizing nonlinear interactions is generated by establishing a time-frequency coherent baseline of multi-source signals and injecting micro-amplitude phase perturbations; a cross-scale energy differential inversion chain is constructed based on this mapping field to extract energy residuals and identify transient energy aggregations; a dynamic pseudo-peak constraint chain is formed through envelope self-reflection calibration; further, time-frequency drift correction and dynamic weight refeedback are performed to generate an adaptive health matrix; finally, the energy spectrum is topologicalized, and continuous monitoring and intelligent early warning of degradation risks are achieved based on topological perturbations. This invention achieves proactive identification and intelligent early warning of power supply component degradation by dynamically fusing energy and phase characteristics through time-frequency drift correction and the topological health matrix, thereby improving the accuracy of performance degradation monitoring for key power supply components.

Description

Technical Field

[0001] This invention relates to the field of power device performance monitoring technology, specifically to an AI spectrum monitoring and early warning system and method for performance degradation of key power device components. Background Technology

[0002] Monitoring and early warning of performance degradation of key power supply components using AI spectroscopy is a power supply health management technology that integrates artificial intelligence and multidimensional signal analysis. Its core lies in using AI models to dynamically perceive and predict the degradation of key internal components of the power supply throughout its entire lifecycle, including power semiconductor devices, capacitors, inductors, and transformers. This method simultaneously collects multi-source signals such as voltage, current, temperature, magnetic field, and acoustic emission during device operation, converting them from the time, frequency, and harmonic domains into energy spectral fingerprints reflecting internal energy changes. This fingerprints then construct a multidimensional feature matrix describing the device's health status. Through deep learning and pattern recognition of this spectral data, the AI ​​model can accurately distinguish early degradation signs caused by factors such as material aging, solder joint fatigue, thermal stress accumulation, or parasitic parameter drift. Combined with time-series rate of change analysis and abnormal spectral line shift detection, it achieves full-process identification and graded early warning from micro-degradation to sub-failure and critical fault. When the system detects abnormal drift in spectral characteristics or abrupt changes in energy distribution, the AI ​​model will output the corresponding risk level and remaining lifetime assessment results, and drive the control layer to execute strategies such as derating operation, enhanced heat dissipation, or redundancy switching. Ultimately, it achieves closed-loop health management of the power system from state perception and degradation judgment to risk handling, effectively preventing system downtime or damage caused by sudden failure of key components.

[0003] The existing technology has the following shortcomings:

[0004] Under high-temperature, high-carrier excitation conditions, the multi-source signals within key power supply components exhibit significant nonlinear interaction characteristics, with complex coupling and redistribution effects in the spectral energy between channels. At this point, abnormal energy accumulation is highly likely to occur in the spectral characteristics. When high-frequency components from different signal channels form short-term pseudo-peak resonance aggregation at a specific carrier frequency, the energy density is abnormally amplified within a narrow frequency band, generating a false stable structure in the local spectral lines. During the learning and feature extraction process, AI models may misjudge this short-term energy concentration caused by harmonic coupling and carrier interaction as a steady-state response of the internal structure of the device, leading to statistical bias during the health matrix construction and evaluation phase. This misjudgment will cause the system to be identified as a low-risk operating state even when degradation evolution such as solder joint microcrack propagation or interface fatigue accumulation actually exists, thus masking potential structural degradation signals and failing to expose early failure signs in time, ultimately creating hidden dangers for sudden failures of key components.

[0005] For example, in a long-term high-temperature aging test of a high-power switching power supply, the power MOSFET and rectifier diode operated at a continuous high carrier frequency, causing continuous and drastic fluctuations in their internal junction temperature and conduction current. Due to the nonlinear coupling between the edge steepness of the current pulse and the junction temperature change, the current signal and the thermal stress signal superimposed and resonated at high frequencies, forming multiple spurious energy peaks. These spurious peaks appeared as short-term stable aggregations of frequency domain energy distributions in AI spectral analysis. Consequently, the AI ​​model misidentified this state as a characteristic pattern of structural health, ignoring the low-frequency energy decay and weak phase drift signals caused by solder joint fatigue and interface delamination. As operating time progressed, these unidentified microcracks continued to propagate, eventually leading to a sharp increase in device thermal resistance, a decrease in the breakdown threshold, and transient failure under sudden load changes.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide an AI spectrum monitoring and early warning system and method for performance degradation of key power supply components, so as to solve the problems in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] An AI-based spectral monitoring and early warning method for performance degradation of key power supply components includes the following steps:

[0010] Step 1: Under the preset high temperature and high carrier excitation conditions, synchronously acquire multi-source signals such as voltage, current, temperature, magnetic field and acoustic emission signals of key power supply components, establish time-frequency coherence baseline of multi-source signals, and inject micro-amplitude phase perturbation sequence into the key power supply components. Based on the dynamic convergence trajectory of the energy distribution of each signal, generate a phase continuous mapping field that characterizes nonlinear interaction features, which is used as a unified reference for subsequent energy anomaly aggregation identification.

[0011] Step 2: Based on the phase continuous mapping field, construct a cross-scale energy difference inversion chain according to the energy convergence trajectory. Extract the energy residual feature matrix using time window sliding and frequency band resampling. With the phase continuous mapping field as a constraint, capture the transient energy aggregation region caused by harmonic coupling, and provide feature input for subsequent pseudo-peak separation.

[0012] Step 3: Based on the energy residual feature matrix, perform envelope self-reflection calibration, compare the energy change curve of the transient energy aggregation region with the historical steady-state envelope curve, detect the local energy amplification trend, and feed the calibration results back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain, which is used to constrain the adaptive update of subsequent phase and energy features.

[0013] Step 4: Based on the dynamic pseudo-peak constraint chain, the time-frequency drift criterion correction mechanism is used to extract the phase stability index of the transient energy aggregation region, and the time-frequency drift characteristics and envelope difference results are jointly corrected. Dynamic weight backfeed is performed to optimize the feature learning path. Based on the corrected phase and energy parameters, an adaptively updatable health matrix is ​​generated to reflect the operational degradation state of key power supply components.

[0014] Step 5: Based on the monitoring output of the health matrix, the energy spectrum is constructed into a high-dimensional topological network. The curvature and density gradient of the energy flow between network nodes are calculated. When local spectral lines generate abnormal aggregation, the abnormal region is identified according to the topological disturbance response. The feature weight allocation of the AI ​​model is dynamically adjusted according to the correction results, thereby realizing continuous monitoring and intelligent early warning of the internal degradation risk of key power supply components.

[0015] Preferably, a phase-continuous mapping field characterizing the nonlinear interaction features is generated based on the dynamic convergence trajectory of each signal energy distribution. The specific steps are as follows:

[0016] Multi-channel synchronous signal acquisition is performed on the operation of key power supply components under high temperature and high carrier excitation conditions. Voltage sampling nodes, current sampling nodes, temperature measurement points, magnetic field induction points and acoustic emission detection points are arranged at the input end, power transmission path and thermal sensitive parts of key power supply components to obtain a multi-source signal matrix after anti-aliasing filtering and time base correction.

[0017] After completing the construction of the time-frequency coherent baseline, a micro-amplitude phase perturbation sequence is applied to the inside of the key power supply components. Under the condition of maintaining stable power output and ambient temperature, the multi-source signals after perturbation are collected, and the dynamic convergence trajectory of energy distribution is obtained by comparing the difference with the multi-source signal matrix before perturbation.

[0018] Using a time-frequency coherent baseline as a reference, the phase changes and energy distribution trajectories of each signal after the disturbance response are mapped in the time-frequency space. The energy center frequency point is mapped to the phase offset value point by point, generating a phase continuous mapping field that characterizes the nonlinear interaction features, which serves as a unified reference for subsequent identification of energy anomaly clusters.

[0019] Preferably, the energy residual feature matrix is ​​extracted, and the transient energy aggregation region caused by harmonic coupling is captured by using a phase continuous mapping field as a constraint. The specific steps are as follows:

[0020] Based on the establishment of the phase continuous mapping field, according to the energy convergence trajectory in the phase continuous mapping field, the multi-source signals are synchronously resampled in both time and frequency dimensions. Signal alignment is performed by time sliding and frequency interpolation to obtain continuous energy migration data along the time-frequency axis.

[0021] Based on the aligned energy migration data, a cross-scale energy difference inversion chain is constructed with reference to the energy convergence trajectory in the phase continuous mapping field. Energy difference between different time levels forms an energy change sequence spanning multiple time scales.

[0022] After the cross-scale energy differential inversion chain is formed, the energy residual feature matrix is ​​extracted by time window sliding and frequency band resampling, and the degree of deviation of energy change is characterized in the form of time-frequency spatial distribution.

[0023] After obtaining the energy residual feature matrix, the transient energy aggregation region caused by harmonic coupling is captured based on the density of the energy residual and the phase gradient change, constrained by the phase continuous mapping field, and used for subsequent pseudo-peak separation and energy anomaly identification.

[0024] Preferably, in the process of capturing the transient energy aggregation region caused by harmonic coupling, the energy residual feature matrix is ​​matched point by point with the adjacent energy convergence trajectory in the phase continuous mapping field. When the energy residual continues to increase within a preset time and the phase gradient shifts continuously, the local time-frequency position that meets the condition is determined as the real energy aggregation region generated by harmonic coupling, and is used as the transient energy aggregation region. The energy change characteristics of the local time-frequency position that meets the condition are used as the input basis for subsequent pseudo-peak separation.

[0025] Preferably, the steps for forming a dynamic pseudo-peak constraint chain include:

[0026] Based on the energy residual feature matrix, the energy density distribution of each signal channel is scanned point by point. Local areas where the energy residual density is greater than the density threshold range and continues to change are identified as transient energy aggregation areas, and energy change curves are extracted along the time axis.

[0027] After obtaining the energy change curve, the historical steady-state envelope curve of the key power supply components under non-disturbance conditions is established and aligned with the energy change curve on the time and frequency coordinates to serve as a reference baseline for energy change.

[0028] Based on the alignment of the energy change curve and the historical steady-state envelope curve, a differential comparison is performed to obtain the energy shift curve. The local energy amplification trend in the energy shift curve is detected, and transient pseudo-peak amplification and structural degradation amplification are distinguished.

[0029] After identifying the local energy amplification trend, envelope autoreflection calibration is performed to map the energy shift result back to the phase continuous mapping field to correct the energy response shift, mark the spurious peak candidate region and adjust the energy weight;

[0030] The envelope self-reflection calibration results are fed back to the phase continuous mapping field and stored as dynamic constraint parameters to form a dynamic pseudo-peak constraint chain, which is used to constrain the adaptive update of subsequent phase and energy characteristics.

[0031] Preferably, during the envelope self-reflection calibration process, when the deviation between the energy change curve and the historical steady-state envelope curve exceeds a preset deviation threshold range, the corresponding region is marked as a pseudo-peak candidate region, and the energy weight corresponding to the pseudo-peak candidate region is reduced in the phase continuous mapping field; when the energy change returns to the range of the historical steady-state envelope, the energy weight corresponding to the pseudo-peak candidate region is automatically restored, so as to achieve adaptive balance and phase consistency maintenance of the dynamic pseudo-peak constraint chain.

[0032] Preferably, the steps for generating an adaptively updatable health matrix include:

[0033] After the dynamic pseudo-peak constraint chain is established, its feedback result is used as the initial constraint condition to dynamically track the transient energy aggregation region in the phase continuous mapping field, record the drift trajectory of the transient energy aggregation region in the time and frequency dimensions, and identify the drift anomaly region.

[0034] After determining that there is a drift trend in the transient energy aggregation region, the phase change characteristics are extracted to obtain the phase stability index. The phase stability index of each transient energy aggregation region is then mapped back to the phase continuous mapping field to form a global phase stability distribution map.

[0035] Based on the extracted phase stability index, the time-frequency drift characteristics and envelope difference results are jointly corrected to synchronously adjust the phase response and energy intensity of the drift anomaly region and mark the potential degradation feature region.

[0036] After joint correction, dynamic weight backfeed is performed. The feature weights of each transient energy aggregation region in the phase continuous mapping field are dynamically adjusted according to the corrected drift information to achieve feature optimization based on historical correction and real-time feedback.

[0037] After the dynamic weighted feedback is completed, an adaptively updatable health matrix is ​​generated based on the corrected phase and energy parameters to reflect the operational degradation status of key power supply components and to be used for lifetime prediction and risk assessment.

[0038] Preferably, the step of dynamically adjusting the feature weight allocation of the AI ​​model based on the correction results includes:

[0039] After the health matrix is ​​generated, the energy and phase parameters are converted into nodes and connections of a high-dimensional topology network, and a multi-dimensional energy topology space is constructed based on the energy flow direction and rate of change.

[0040] After establishing a high-dimensional topological network, the curvature and density gradient of energy flow between network nodes are calculated to form a time-varying feature field describing the dynamic changes of energy transfer paths.

[0041] When a local spectral line anomalous convergence is detected in the time-varying characteristic field of energy flow curvature and density gradient, the anomalous region is calibrated based on the topological perturbation response, and the magnitude and duration of the topological perturbation response are recorded.

[0042] After the abnormal regions are identified, the topological perturbation response is fused with the health matrix correction data, and the feature weight allocation is dynamically adjusted according to the correction results. The weight is increased for persistent abnormal regions and decreased for transient abnormal regions.

[0043] The dynamically adjusted feature weight distribution is fed back to the health matrix, enabling the model to adaptively correct itself in subsequent updates, thereby completing the continuous monitoring and intelligent early warning of degradation risks of key power supply components.

[0044] Preferably, the process of dynamically adjusting the feature weight allocation includes: when the topological perturbation response exceeds the preset response threshold range, increasing the feature weight of the corresponding node in real time, and gradually decreasing the feature weight in subsequent cycles according to the recovery speed of energy flow curvature and density gradient; when the abnormal region persists and the topological perturbation response has not recovered to a stable range, marking the corresponding abnormal region as a high-risk node in the health matrix to enhance the model's ability to continuously respond to degradation trends.

[0045] An AI-based spectral monitoring and early warning system for performance degradation of key power supply components includes a phase mapping construction module, an energy differential inversion module, an envelope calibration constraint module, a drift correction and update module, and a topology weight control module.

[0046] The phase mapping construction module establishes a time-frequency coherent baseline for multi-source signals under high temperature and high carrier excitation conditions, injects a micro-amplitude phase perturbation sequence into the internal components of the power supply, and generates a phase continuous mapping field characterizing nonlinear interaction features based on the dynamic convergence trajectory of the energy distribution of each signal.

[0047] The energy differential inversion module, based on the phase continuous mapping field, constructs a cross-scale energy differential inversion chain according to the energy convergence trajectory, extracts the energy residual feature matrix by using time window sliding and frequency band resampling, and captures the transient energy aggregation region caused by harmonic coupling with the phase continuous mapping field as a constraint.

[0048] The envelope calibration constraint module performs envelope self-reflection calibration based on the energy residual feature matrix. It compares the energy change curve of the transient energy aggregation region with the historical steady-state envelope curve to detect the local energy amplification trend and feeds the calibration results back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain.

[0049] The drift correction and update module is based on a dynamic pseudo-peak constraint chain and a runtime frequency drift criterion correction mechanism. It extracts the phase stability index of the transient energy aggregation region and jointly corrects the time-frequency drift features with the envelope difference results. It performs dynamic weight backfeeding to optimize the feature learning path and generates an adaptively updatable health matrix based on the corrected phase and energy parameters.

[0050] The topology weight control module, based on the monitoring output of the health matrix, constructs the energy spectrum into a high-dimensional topology network, calculates the curvature and density gradient of energy flow between network nodes, and when local spectral lines generate abnormal aggregation, it identifies abnormal regions based on the topology perturbation response and dynamically adjusts the feature weight allocation of the AI ​​model according to the correction results.

[0051] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0052] This invention establishes a time-frequency coherent baseline for multi-source signals under high-temperature, high-carrier-load operation conditions and introduces a phase-continuous mapping field and a dynamic pseudo-peak constraint chain, enabling accurate identification of anomalous energy distribution clusters under a phase-consistent physical reference. Through cross-scale energy differential inversion and envelope self-reflection calibration, the trend of energy residual changes can be continuously tracked, effectively suppressing spurious energy peaks caused by harmonic coupling. This allows the AI ​​model to focus only on physically representative real degradation features during the learning phase. This approach significantly improves the interpretability of energy spectral data and the authenticity of feature extraction, enabling early exposure of early degradation signs in key power supply components and avoiding health assessment biases caused by misjudgments.

[0053] This invention achieves dynamic fusion and adaptive weight adjustment of energy and phase characteristics through time-frequency drift correction and the establishment of a topologically networked health matrix. The health matrix can dynamically correct the weight allocation based on changes in energy flow curvature and density gradient, enabling the model to develop self-learning and self-correction capabilities during continuous monitoring. This transforms the degradation process of key power supply components from passive diagnosis to active identification, achieving spatial localization of energy spectrum anomalies and adaptive risk level assessment. It significantly improves the stability and early warning accuracy of degradation monitoring, providing continuous intelligent support for the long-term reliable operation of power supply systems. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0055] Figure 1 This is a flowchart of the AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to the present invention.

[0056] Figure 2 This is a schematic diagram illustrating the principle of forming a dynamic pseudo-peak constraint chain in this invention.

[0057] Figure 3 A schematic diagram illustrating the principle of generating an adaptively updatable health matrix for this invention;

[0058] Figure 4 This is a schematic diagram of the AI ​​spectrum monitoring and early warning system for performance degradation of key power supply components according to the present invention. Detailed Implementation

[0059] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0060] like Figures 1 to 3 As shown, this invention provides an AI spectrum monitoring and early warning method for performance degradation of key power supply components, including the following steps:

[0061] Step 1: Under the preset high temperature and high carrier excitation conditions, synchronously acquire multi-source signals such as voltage, current, temperature, magnetic field and acoustic emission signals of key power supply components, establish time-frequency coherence baseline of multi-source signals, and inject micro-amplitude phase perturbation sequence into the key power supply components. Based on the dynamic convergence trajectory of the energy distribution of each signal, generate a phase continuous mapping field that characterizes nonlinear interaction features, which is used as a unified reference for subsequent energy anomaly aggregation identification.

[0062] A phase-continuous mapping field characterizing the nonlinear interaction features is generated based on the dynamic convergence trajectory of each signal energy distribution. The specific steps are as follows:

[0063] Multi-channel synchronous signal acquisition was performed on the operation of key power supply components under high-temperature, high-carrier excitation conditions. To ensure the acquired signals possess complete dynamic characteristics, multiple signal acquisition points were arranged at the input terminals, power transmission paths, and thermosensitive parts of the key power supply components, including voltage sampling nodes, current sampling nodes, temperature measurement points, magnetic field sensing points, and acoustic emission detection points. Voltage signals were directly acquired from the power drive end using a high-bandwidth differential sampling probe to capture voltage waveform distortion and transient spikes within the carrier modulation period. Current signals were synchronously acquired using a high-sensitivity current transformer to record transient current responses at high carrier frequencies and current transition characteristics during conduction and turn-off. Temperature signals were measured using an embedded thermocouple and an infrared temperature detector, allowing simultaneous acquisition of junction temperature changes and surface temperature distribution to ensure timing accuracy during thermal diffusion. Magnetic field signals were acquired by a Hall sensor surrounding the power device package to reflect changes in electromagnetic coupling strength and the dynamic distribution of the magnetic field under carrier excitation. Acoustic emission signals were acquired using a high-sensitivity piezoelectric transducer to record the acoustic energy release characteristics caused by microcrack initiation or material stress relaxation within the device. The acquisition of other signals will not be elaborated upon here, and will be based on existing technology. All signals are sampled using the same time reference, with the sampling frequency set to five to ten times the carrier frequency to ensure complete recording of the high-frequency response within each carrier cycle. The acquired multi-source signals are filtered for anti-aliasing and corrected for the time reference to form a multi-source signal matrix with perfect time alignment. Based on this, by performing sliding calculations on each signal within multiple time windows, the short-time frequency distribution of the signal within each window is obtained, and a time-frequency joint coordinate system is established with time as the horizontal axis and frequency as the vertical axis. By comparing the amplitude and phase information of different signal channels under the same time index, the phase difference distribution and energy coupling trend between each channel at that moment can be calculated, thereby obtaining a time-frequency coherence baseline reflecting the current operating state of key power supply components. This time-frequency coherence baseline, with time as the continuous axis and the energy distribution and phase difference of each signal as spatial variables, demonstrates the natural coupling relationship between multiple physical quantities within the key power supply components, providing a benchmark reference for subsequent phase disturbance analysis.

[0064] After establishing the time-frequency coherent baseline, a micro-amplitude phase perturbation sequence is applied to the power supply components while ensuring their normal operation. This phase perturbation is achieved by applying a small phase delay signal to the carrier drive signal, with the phase change range controlled within the order of one-hundredth to one-thousandth of the original carrier phase to avoid substantial impact on power output. During the perturbation process, the input power, load conditions, and ambient temperature of the power supply components are kept stable to ensure that the perturbation only affects the phase response of the signal without changing the overall energy input. The micro-amplitude phase perturbation is applied to the carrier signal repeatedly in the form of short-period pulses, causing phase fine-tuning between voltage, current, and magnetic field signals within each perturbation cycle. The multi-source signals after the perturbation are then re-acquired, and the transient response changes of each signal at the same time are recorded. To distinguish between the changes caused by the perturbation and intrinsic fluctuations, the original signal matrix acquired before the perturbation and the signal matrix after the perturbation are compared point-by-point to calculate the energy distribution difference curve, and the results are mapped to the time-frequency joint coordinate space. By continuously comparing the energy difference changes within multiple perturbation cycles, the trend of energy redistribution between different signal channels can be observed. For example, at the rising edge of the carrier modulation period, the high-frequency components of the voltage signal are transferred to the magnetic field signal channel, while at the falling edge, some of the high-frequency energy from the current channel is transferred to the acoustic emission channel. These changes reflect the transient characteristics of electromagnetic coupling and thermoacoustic response within key power supply components. When the energy response trajectory after multiple perturbations shows a gradually converging trend in the time-frequency joint coordinate system, it indicates that the nonlinear interaction characteristics have been successfully excited, and the coupling relationship between signals has changed from a nondeterministic random distribution to a predictable continuous energy flow pattern. At this point, the dynamic convergence trajectory of the energy distribution becomes important physical evidence describing the nonlinear interaction behavior within key power supply components.

[0065] It should be noted that:

[0066] Intrinsic fluctuations refer to the micro-fluctuations in the signal of key power supply components naturally generated by factors such as internal materials, structure, parasitic parameters, or thermal noise when there are no external disturbances. They are inherent response components reflecting the natural steady-state behavior of the device. In this invention, it serves as a background reference to distinguish between the nonlinear interaction effects caused by applied phase perturbations and the inherent fluctuations of the device itself, thereby ensuring the accuracy of energy convergence trajectory analysis and the physical authenticity of the phase mapping field.

[0067] After obtaining the dynamic convergence trajectory of the energy in the disturbance response, the previously established time-frequency coherence baseline is used as a reference to perform a full-range mapping of the phase changes and energy distribution trajectories of each signal in the time-frequency space, generating a phase-continuous mapping field characterizing the nonlinear interaction features. Specifically, the energy center frequency point of each signal in each time window after disturbance is mapped to the corresponding phase offset value point by point, and these points are connected in chronological order to form a continuous energy migration path. Each path reflects the energy flow trend of a signal channel. When the paths of multiple signal channels intersect, separate, or overlap in space, energy convergence regions and phase separation regions are formed in the phase-continuous mapping field. The energy convergence region indicates that different signal channels have a strong coupling relationship under specific frequency and time conditions, while the phase separation region represents that the responses of each signal channel are independent of each other. By comparing the phase change rate and energy density of each region in the phase-continuous mapping field, the nonlinear interaction features generated by the combined effects of high carrier excitation and temperature stress can be identified. These characteristics often manifest as a dynamic energy transfer process from high frequency to low frequency and from the main excitation channel to the passive response channel, with its trajectory extending in a spiral or radial pattern within the phase-continuous mapping field. To ensure the stability of the mapping field under multiple perturbation cycles, the mapping results for each cycle are normalized and superimposed, retaining the stable components of the energy change trajectory and removing fluctuations caused by short-term noise. The resulting phase-continuous mapping field exhibits a smooth and continuous phase distribution structure throughout the entire time-frequency space, with clearly discernible energy convergence trajectories and significantly prominent energy anomalous accumulation regions. This phase-continuous mapping field reflects both the true energy transfer patterns within key power supply components under high-temperature, high-carrier excitation and reveals the harmonic energy coupling behavior caused by nonlinear interactions. By using this phase-continuous mapping field as a unified phase and energy reference, it can be used as a coherent comparison benchmark in subsequent energy anomalous accumulation detection and health status identification processes, enabling effective removal of spurious peak features and early identification of degradation signs.

[0068] Through the above steps, the entire process from synchronous acquisition of multi-source signals, phase perturbation response excitation, to nonlinear interactive mapping is completed. The entire process maintains phase continuity and energy conservation at the physical level, enabling the visualization of multi-field coupling behavior of key power supply components under high-temperature, high-carrier excitation conditions. This provides a data foundation for subsequent energy anomaly accumulation identification, health matrix construction, and degradation early warning.

[0069] Step 2: Based on the phase continuous mapping field, construct a cross-scale energy difference inversion chain according to the energy convergence trajectory. Extract the energy residual feature matrix using time window sliding and frequency band resampling. With the phase continuous mapping field as a constraint, capture the transient energy aggregation region caused by harmonic coupling, and provide feature input for subsequent pseudo-peak separation.

[0070] The energy residual feature matrix is ​​extracted, and the transient energy aggregation region caused by harmonic coupling is captured by using a phase continuous mapping field as a constraint. The specific steps are as follows:

[0071] Based on the established phase-continuous mapping field, and according to the energy convergence trajectory displayed in the phase-continuous mapping field, multi-source signals are synchronously resampled in both time and frequency dimensions to obtain details of energy changes in each channel. Specifically, the dense region of energy trajectory in the phase-continuous mapping field is used as the initial analysis window. Transient energy values ​​and phase delay values ​​of different signal channels are extracted within each time slice and registered with the corresponding energy convergence trajectory in the phase-continuous mapping field. To ensure time accuracy, a fixed-length time slip is performed on each signal channel, causing the sampling window to move along the time axis with small steps, thereby capturing minute changes in energy flow within the carrier modulation period. When the signal crosses the frequency domain, the energy distribution of different channels at the same time point is interpolated to maintain uniform spacing between energy points on the frequency axis, ensuring consistent alignment of frequency band information between different channels. Through this continuous slip and alignment method in both time and frequency dimensions, the dynamic migration process of energy along the time and frequency axes can be accurately depicted under the constraints of the phase-continuous mapping field, providing fundamental data for constructing a cross-scale energy differential inversion chain.

[0072] Based on the time-frequency aligned energy data, a cross-scale energy differential inversion chain is constructed using the energy convergence trajectory in the phase-continuous mapping field as a reference. Specifically, the start and end points of the energy convergence trajectory are used as reference boundaries between different scales. Small-scale time slices reflect short-term energy fluctuations, while large-scale time slices reflect long-term energy trends. This scale division decomposes energy flow into several nested time-level structures. For each time level, the average energy value and phase offset within the corresponding frequency range are extracted, and the difference between adjacent time levels is used as the energy change gradient. This gradient reflects the rate at which energy transfers from high frequency to low frequency or from the main signal channel to the coupling channel at different time scales. When the cross-scale difference reflects a continuous convergence characteristic of energy distribution across multiple time levels, it indicates the existence of a relatively stable nonlinear interaction path within the key power supply components. By connecting the energy differential results at each scale along the time axis, a cross-scale energy differential inversion chain spanning multiple time levels can be formed. This inversion chain comprehensively describes the temporal changes in energy flow of key power supply components under high temperature and high carrier conditions, providing a three-dimensional representation of energy aggregation phenomena across multiple scales.

[0073] After the cross-scale energy differential inversion chain is formed, the energy residual feature matrix is ​​extracted using time window sliding and frequency band resampling. Time window sliding is used to capture the transient features of energy changes over time along the time axis. By setting a sliding window with a high overlap ratio, sufficient time crossover regions are ensured between adjacent windows, thus capturing the continuous transition of energy changes. As the sliding window traverses the entire time series sequentially, the average distribution and peak difference of energy within each window are recorded, and the energy difference sequence between adjacent windows is calculated. Frequency band resampling is used to enhance the fine-grained identification capability of high-frequency and low-frequency energy distributions along the frequency axis. By dividing the frequency range into several overlapping sub-bands, the energy distribution remains smoothly connected in each sub-band during the resampling process, thus avoiding abrupt changes at frequency boundaries. Combining the time sliding results and the frequency band resampling results yields an energy residual feature matrix covering the entire time-frequency space. Each element in this matrix corresponds to the degree of deviation of energy changes under specific time and frequency conditions, and its spatial distribution reflects the convergence and diffusion state of energy in the time-frequency space. By comparing with the phase-continuous mapping field, it can be observed that some energy residuals exhibit local anomalous accumulation around the energy convergence region of the mapping field. These regions are often the result of the combined effects of harmonic coupling and high-temperature stress.

[0074] After obtaining the energy residual feature matrix, the transient energy aggregation region caused by harmonic coupling is captured using the phase continuous mapping field as a constraint, and the energy change characteristics of this region are used for subsequent pseudo-peak separation. Specifically, the energy residual feature matrix is ​​mapped point-by-point to the energy convergence trajectory in the phase continuous mapping field, and the density and continuity of the energy residual are calculated between adjacent energy convergence trajectories in the phase continuous mapping field. When the energy residual continuously increases within the same frequency range over a preset time period and intersects with the energy trajectories of multiple signal channels in space, transient energy aggregation can be identified in this region. To distinguish between true energy aggregation caused by harmonic coupling and pseudo-aggregation caused by transient carrier jitter, the rate of change of the energy residual in the time dimension is compared with the phase gradient at the corresponding position in the phase continuous mapping field. If the energy aggregation is accompanied by a continuous shift in phase change rather than an abrupt change, the aggregation is confirmed as a result of harmonic coupling effects. In this way, multiple transient energy aggregation regions can be identified in the phase-continuous mapping field. Each region represents a specific nonlinear interaction mode, such as high-frequency harmonic fusion between current and magnetic field channels, or low-frequency coupled oscillations between acoustic emission and temperature signals. These transient energy aggregation regions will serve as feature inputs in subsequent energy anomaly analysis to distinguish between genuine structural degradation signals and short-term spurious peaks caused by spectral interactions.

[0075] Through the above steps, the entire process from constructing a phase-continuous mapping field to a cross-scale energy differential inversion chain, and then to extracting the energy residual feature matrix and identifying transient energy aggregation regions, is realized. This process achieves energy reconstruction at multiple time scales and frequency levels while maintaining phase continuity, providing a three-dimensional representation of the energy distribution evolution within key power supply components. By capturing transient energy aggregation regions formed by harmonic coupling under the constraint of a phase-continuous mapping field, highly reliable feature inputs are provided for subsequent pseudo-peak separation and degradation identification.

[0076] Step 3: Based on the energy residual feature matrix, perform envelope self-reflection calibration, compare the energy change curve of the transient energy aggregation region with the historical steady-state envelope curve, detect the local energy amplification trend, and feed the calibration results back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain, which is used to constrain the adaptive update of subsequent phase and energy features.

[0077] The specific steps to form a dynamic pseudo-peak constraint chain are as follows:

[0078] Based on the energy residual feature matrix, energy change curves are extracted from transient energy aggregation regions. By scanning the spatial distribution of energy density values ​​of each signal channel in the energy residual feature matrix point by point, local regions where the energy residual density exceeds the density threshold and continuously changes are identified as transient energy aggregation regions. Within each transient energy aggregation region, continuous energy value change curves are extracted along the time axis to record the growth, convergence, and decay trends of energy over a short period. To ensure the physical representativeness of the extracted curves, the energy change curves are limited to the energy convergence path range of the phase continuous mapping field, so that the energy change curves reflect both the dynamic process of harmonic coupling and remain consistent with the overall energy migration trend. The energy change curves obtained in this way can accurately reflect the energy evolution characteristics within the transient energy aggregation region and are the core input data for subsequent envelope alignment and self-reflection calibration.

[0079] After obtaining the energy change curve of the transient energy aggregation region, a historical steady-state envelope curve is established as a reference baseline for energy change. Through statistical analysis of long-term operating data of key power supply components under undisturbed conditions, the average range of energy value variation and typical steady-state fluctuation amplitude are extracted from each signal channel, and a historical steady-state envelope curve is constructed based on this. This historical steady-state envelope curve represents the upper and lower limits of energy fluctuation over time for each signal channel under normal and healthy conditions. Its shape is smooth and its changes are slow, reflecting the energy balance characteristics of the device under conditions of no degradation and no harmonic coupling interference. To ensure the comparability of the historical steady-state envelope curve and the transient energy change curve, the historical steady-state envelope curve is repositioned according to the energy time axis of the phase continuous mapping field, so that the two are completely corresponding in time and frequency coordinates. Thus, when the energy change curve of the transient energy aggregation region is superimposed on the historical steady-state envelope curve, the degree of deviation of the energy change from the steady-state range can be clearly observed.

[0080] Subsequently, a differential comparison process is performed based on the aligned energy change curve and the historical steady-state envelope curve to detect local energy amplification trends. Specifically, the energy value at each time point of the transient energy change curve is compared with the average energy value at the corresponding position in the steady-state envelope to obtain the energy offset curve. If the energy offset curve continuously deviates in the positive direction and the amplitude gradually increases over a certain period, it indicates that an abnormal energy amplification phenomenon has occurred in that region. At the same time, by observing the continuous directional changes of the energy offset curve, it can be determined whether the energy amplification is a short-term burst or a slow accumulation. If the energy offset curve shows a step-like or pulse-like increase, it indicates that the amplification trend is caused by transient harmonic interference or parasitic parameter coupling; if the energy offset curve is smooth and rises over a long period, it indicates that the energy aggregation may originate from structural degradation or material stress accumulation. Through this differential comparison method, it is possible to effectively distinguish between transient pseudo-peak energy amplification and energy enhancement caused by real degradation, thus providing a quantitative basis for subsequent calibration and constraint.

[0081] After identifying a localized amplification trend in energy, envelope autoreflection calibration is performed to correct the energy response shift in that region within the phase continuous mapping field. The core idea of ​​envelope autoreflection calibration is to use the comparison between the energy change curve and the historical steady-state envelope curve to map the amplification magnitude of the energy anomaly region back to the corresponding phase coordinates in the phase continuous mapping field, and then adjust the energy intensity representation of that region in the phase continuous mapping field through reflection. When the deviation between the energy change curve and the historical steady-state envelope curve exceeds a preset deviation threshold, that region is marked as a spurious peak candidate region, and its energy weight in the phase continuous mapping field is reduced to prevent it from misleading the overall energy convergence trend. If the energy change in that region recovers to within the steady-state envelope range in subsequent time windows, the weight of that region is restored to its normal value through autoreflection. By continuously performing this forward and reverse calibration, the energy distribution in the phase continuous mapping field can be kept in dynamic equilibrium. This process is equivalent to establishing a self-correction mechanism for the phase continuous mapping field, enabling it to respond in real time to the detection results of localized energy amplification, maintaining overall phase consistency and energy continuity.

[0082] Finally, the results of the envelope self-reflection calibration are fed back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain, which is used to constrain the adaptive updates of subsequent phase and energy features. Specifically, after completing a full envelope self-reflection calibration, the energy weight adjustment information, phase correction information, and pseudo-peak region identifiers are stored as dynamic constraint parameters. These parameters are then called in the energy analysis of the next cycle to constrain the energy update process in the phase continuous mapping field. When new energy data is input into the phase continuous mapping field, the dynamic pseudo-peak constraint chain can automatically determine whether the data is located in a region where pseudo-peaks have previously appeared, and adjust its energy contribution based on historical calibration information to prevent pseudo-peak features from being absorbed by the model as normal energy patterns again. At the same time, the dynamic pseudo-peak constraint chain can also automatically adjust the constraint strength according to the recurrence frequency of pseudo-peaks in the time series, increasing the constraint level for frequently occurring pseudo-peak regions and applying weaker constraints to occasional abnormal regions, so that the phase continuous mapping field maintains sensitivity while possessing anti-interference capabilities. Through this dynamic feedback and constraint mechanism, the subsequent phase and energy feature update process can be carried out under a stable benchmark, ensuring the authenticity and consistency of the energy aggregation analysis results.

[0083] Through the execution of the above steps, the phase continuous mapping field is no longer a static energy description structure, but becomes an energy distribution mapping space with self-correction and dynamic constraint capabilities. This envelope self-reflection calibration method can effectively eliminate transient energy spurious peaks caused by high carrier excitation, harmonic superposition, and environmental disturbances, ensuring a more accurate and reliable correspondence between energy aggregation characteristics and the actual degradation state of key power supply components.

[0084] Step 4: Based on the dynamic pseudo-peak constraint chain, the time-frequency drift criterion correction mechanism is used to extract the phase stability index of the transient energy aggregation region, and the time-frequency drift characteristics and envelope difference results are jointly corrected. Dynamic weight backfeed is performed to optimize the feature learning path. Based on the corrected phase and energy parameters, an adaptively updatable health matrix is ​​generated to reflect the operational degradation state of key power supply components.

[0085] The specific steps for generating an adaptively updatable health matrix are as follows:

[0086] After the dynamic pseudo-peak constraint chain is established, its feedback results are used as initial constraints to dynamically track and determine the drift of transient energy aggregation regions in the phase continuous mapping field. Since the dynamic pseudo-peak constraint chain has recorded the historical pseudo-peak occurrence frequency and energy weight correction information of each transient energy aggregation region, the calibration data in the constraint chain can be used as a reference to monitor the positional shift of the transient energy aggregation region in the time and frequency dimensions. When the key power supply components are in a high-temperature, high-carrier-load operation state, the transient energy aggregation region often drifts slowly along the frequency axis or time axis due to factors such as parasitic parameter drift, local thermal mismatch, and stress diffusion. Therefore, the phase center position of each transient energy aggregation region in the phase continuous mapping field is defined, and its positional change trajectory is recorded in a continuous time slice. By superimposing this trajectory with the energy weight distribution recorded in the dynamic pseudo-peak constraint chain, it can be determined which regions' drift is within the controllable range and which drifts have an abnormal diffusion trend. If a transient energy aggregation region continuously shifts towards higher frequencies in multiple sampling periods, while the energy density does not show a significant attenuation, it indicates that the phase coupling state of this region has become unstable and needs to enter the drift correction process.

[0087] After identifying a drift trend in transient energy aggregation regions, the phase change characteristics of these regions are refined to obtain phase stability indices. Specifically, using the transient energy aggregation regions identified by the dynamic pseudo-peak constraint chain as boundaries, piecewise integration is performed on the phase continuity curve within each transient energy aggregation region, recording the total phase change and direction within adjacent time windows. By comparing the smoothness and continuity of phase changes within adjacent time windows, the stability of the phase response can be reflected. When the phase change rate of a transient energy aggregation region fluctuates significantly within a short period, or when the phase direction exhibits periodic reversals, it indicates the presence of phase drift instability induced by nonlinear interaction reinforcement or degenerative evolution. To ensure global consistency of this phase stability index, the phase stability indices of each transient energy aggregation region are mapped back to the global coordinates of the phase continuity mapping field, and the spatial distribution of the phase stability indices is labeled using color gradients or numerical weights, thus forming a mapping map reflecting the global phase stability state. This mapping map not only reveals the differences in phase stability among different transient energy aggregation regions but also provides a visual basis for subsequent joint correction.

[0088] After extraction, based on the extracted phase stability index, the time-frequency drift features and envelope difference results are jointly corrected to improve the accuracy of drift calibration. This joint correction process uses the time-series energy change trend in the energy residual feature matrix as a reference, and maps the phase-instability regions in the time-frequency drift features to the energy amplification regions in the envelope difference point by point. If a transient energy aggregation region simultaneously meets the conditions of frequent phase drift and continuous deviation of the energy envelope, it indicates that there is a strong nonlinear interactive imbalance in the region, and its weight should be corrected. Specifically, by calling the previously calibrated energy weight parameters in the dynamic pseudo-peak constraint chain, the phase response and energy intensity of the drift region are adjusted synchronously to make the phase change return to the steady-state range. If the drift region fails to recover to the steady-state envelope range within multiple cycles, it indicates that the energy aggregation trend in the region continues to strengthen, and it should be marked as a potential degradation feature region, and its attention weight in the subsequent feature learning process should be increased. The significance of this joint correction lies in the fact that by integrating phase stability and energy envelope changes, it achieves a dual constraint on drift characteristics, enabling drift correction to not only rely on phase information but also consider the real trend of energy evolution, thereby improving the physical rationality and stability of the correction.

[0089] After the joint correction of drift features is completed, dynamic weight backfeeding is performed to optimize the learning path of phase and energy features. The core of this step lies in dynamically adjusting the feature weights of each transient energy aggregation region in the phase continuous mapping field based on the jointly corrected drift information. This allows the phase continuous mapping field to automatically suppress spurious peak enhancement and strengthen the representation of true degradation signals during subsequent updates. Specifically, the corrected phase stability index is used as the main reference value to weight and update the phase response intensity of each transient energy aggregation region in the phase continuous mapping field. When phase stability is high and the energy envelope remains stable, the feature weight of this region is appropriately reduced to avoid the model over-reliance on stable signals; conversely, when phase stability decreases and the energy envelope deviates over a long period, the feature weight of this region is automatically increased, making the model more sensitive to potential degradation features. This weight adjustment process is not a one-time correction but is executed cyclically in the form of dynamic backfeeding. Each time the phase continuous mapping field completes a round of updates, the new weight adjustment results are fed back to the dynamic spurious peak constraint chain, causing subsequent drift criteria to be recalculated under the new weight conditions, thus realizing a closed-loop feature optimization mechanism based on historical correction and real-time feedback. This dynamic weight backfeeding not only enhances the model's adaptability but also enables the phase continuous mapping field to have the ability to learn and continuously correct itself during long-term operation.

[0090] After dynamic weight refeedback is completed, an adaptively updatable health matrix is ​​generated based on the corrected phase and energy parameters to reflect the operational degradation state of key power supply components. This health matrix is ​​composed of phase stability, energy density distribution, and drift correction weights in transient energy aggregation regions. By integrating these multi-dimensional characteristic parameters over time, a health state expression space with dynamic evolutionary properties is formed. In this space, the health level of key power supply components is jointly determined by energy distribution balance, phase stability continuity, and drift convergence. When a component is in a healthy state, the phase distribution in the health matrix remains stable, and the energy density distribution exhibits a symmetrical shape. When early degradation occurs, local energy nodes in the health matrix show density anomalies or intensified phase drift, reflecting a trend of nonlinear interactive imbalance. When degradation further progresses to a critical stage, the energy convergence path in the health matrix breaks or shifts, manifesting as an abnormal amplification of local energy aggregation regions. By continuously updating the health matrix, the degradation process of key power supply components can be monitored in real time, and the remaining service life and potential risk level can be predicted based on the matrix evolution trend. Since the generation of the health matrix depends on the corrected phase and energy parameters, its output can accurately reflect the real physical degradation state, avoiding misjudgment problems caused by harmonic coupling and spurious peak response.

[0091] Through the above steps, from drift detection in transient energy aggregation regions, phase stability extraction, joint correction, dynamic weight refeeding to health matrix generation, adaptive fusion of energy and phase features is achieved. The entire process maintains energy consistency and phase continuity under the constraint of a dynamic pseudo-peak constraint chain, enabling the real-time capture and accurate characterization of the degradation state of key power supply components under high-temperature, high-carrier conditions, thereby significantly improving the reliability of degradation identification and the accuracy of health assessment.

[0092] Step 5: Based on the monitoring output of the health matrix, the energy spectrum is constructed into a high-dimensional topological network. The curvature and density gradient of the energy flow between network nodes are calculated. When local spectral lines generate abnormal aggregation, the abnormal area is marked according to the topological disturbance response. The feature weight allocation of the AI ​​model is dynamically adjusted according to the correction results, thereby realizing continuous monitoring and intelligent early warning of the internal degradation risk of key power supply components.

[0093] The AI ​​model's feature weights are dynamically adjusted based on the calibration results to continuously monitor and intelligently warn of internal degradation risks in key power supply components. The specific steps are as follows:

[0094] After the health matrix is ​​generated, its energy and phase parameters are converted into nodes and connections in a high-dimensional topological network to construct the spatial topology of the energy spectrum. Each transient energy aggregation region in the health matrix corresponds to a time-frequency energy node within a key power supply device. The node's energy intensity, phase stability, and weight value jointly determine its geometric position and connection strength in the topological network. Specifically, the energy flow direction between adjacent transient energy aggregation regions in the health matrix is ​​defined as the connection edge between nodes, and the energy change rate is defined as the edge weight. These are superimposed across multiple time series to form a continuous high-dimensional topological network. This network not only includes the two basic coordinate axes of time and frequency but also introduces energy density, phase gradient, and drift correction values ​​as additional dimensions, thus forming a multi-dimensional energy topological space capable of dynamically expressing the energy transmission state. In this topological space, the energy transmission relationship of each signal channel within the key power supply device is represented by the distribution and connection pattern of nodes. The energy aggregation trend is manifested through the shortening of the distance between nodes, the enhancement of connection weights, and the concentration of local energy density. Through this topology processing, the energy and phase information that were originally scattered in the health matrix are organized into a structure with spatial geometric features, so that the health status of key power supply components can be identified and analyzed in a more intuitive and continuous topological semantics.

[0095] After establishing a high-dimensional topological network, the energy flow characteristics between network nodes are analyzed, and the curvature and density gradient of the energy flow between nodes are calculated to reveal the dynamic changes in the energy transfer path. Specifically, the direction of energy transfer between nodes is taken as the path vector, and the rate of change of energy flow intensity along the path direction is measured. The rate of change of energy intensity with path length is defined as the curvature of the energy flow. Curvature reflects the degree of bending of the energy flow path in the topological space: when the curvature of the energy flow is small, it indicates that the energy flow is smooth and continuous; when the curvature increases, it indicates that the energy flow is deflected due to nonlinear interactive perturbations. At the same time, a local energy density region is defined around each node. By calculating the spatial gradient of the energy distribution in the neighborhood of the node, the energy density gradient can be obtained. This gradient reflects the diffusion or accumulation trend of energy in the topological space. When the energy density gradient is large, it indicates that there is obvious energy accumulation or energy outflow in the region; while when the energy density gradient approaches zero, it indicates that the energy distribution tends to be stable or in equilibrium. By jointly analyzing the energy flow curvature and density gradient, a time-varying characteristic field can be formed to describe the topological energy evolution state. This feature field can intuitively reveal the dynamic flow relationship of energy between different regions inside key power supply components, providing a quantitative basis for subsequent anomaly aggregation identification.

[0096] When anomalous aggregation of local spectral lines is detected in the time-varying characteristic field of energy flow curvature and density gradient, the anomalous region is calibrated based on the topological perturbation response. Anomalous aggregation of local spectral lines typically manifests as a rapid shortening of distances between nodes, a significant increase in connection weights, or a sharp concentration of energy density within a short period. To accurately calibrate these anomalous regions, the topological perturbation response of the network structure is first monitored in continuous time slices, i.e., the magnitude of geometric changes in the connections between network nodes. If the topological perturbation response exceeds a preset threshold range within a certain time period, and this region is simultaneously accompanied by a sharp increase in energy flow curvature and abrupt changes in density gradient, then anomalous aggregation is identified in this region. Such aggregation may be caused by physical phenomena such as abnormal junction temperature of power semiconductor devices, microcrack propagation at solder joint interfaces, nonlinear increase in parasitic capacitance, or magnetic flux coupling imbalance. For the calibrated anomalous regions, the magnitude and duration of their topological perturbation response are further recorded to characterize the intensity and persistence of the anomalous changes. A larger topological perturbation response indicates that the energy transport path in this region is severely distorted, and the system's health is significantly affected. By continuously monitoring the temporal changes in the topological perturbation response, the evolutionary trend of the degradation process can be identified, namely the transitional stage from slight coupling drift to energy-locked aggregation, thereby enabling dynamic tracking of early degradation.

[0097] After the abnormal regions are identified, the topological perturbation response and the corrected data from the health matrix are fused. Based on the correction results, the feature weight allocation is dynamically adjusted to achieve adaptive optimization and degradation risk warning for the model. Specifically, the node weights corresponding to the identified abnormal regions are combined with the phase stability index of the same region in the health matrix. The drift direction and energy aggregation trend are compared to determine the priority of this region in the feature learning process. When an abnormal region persists or spreads over multiple time periods, it indicates an increased degradation risk in that region, requiring an increase in its proportion in the feature weights to make the model more sensitive to its changes. For occasional perturbations or transient energy anomalies, their feature weights are reduced based on the recovery speed of the topological perturbation response to avoid the model being interfered with by short-term noise. Dynamic weight adjustment is not only reflected in a single time slice but also iteratively updated in subsequent periods. After each weight adjustment, the new weight distribution is fed back to the health matrix, allowing it to automatically correct the balance between phase and energy distribution in the next energy spectrum calculation, thereby achieving adaptive evolution. Over time, the topology network structure will gradually stabilize under the drive of dynamic weight adjustments, and the changes in energy flow curvature and density gradient will tend to be gradual, indicating that the operating state of key power supply components has transitioned from an unstable stage to a relatively stable stage. Through this process, closed-loop management from anomaly detection to risk warning can be achieved without external intervention.

[0098] It should be noted that:

[0099] In one specific implementation, after constructing a high-dimensional topology network based on the health matrix and completing the abnormal region calibration, the feature weight allocation process is further dynamically adjusted to achieve continuous monitoring and intelligent early warning of degradation risks of key power supply components. Specifically, when continuous monitoring of the topology network reveals that the topology disturbance response of a certain node exceeds a preset response threshold range, the node is determined to be in an abnormally enhanced energy transfer state, and its feature weight in the health matrix is ​​immediately increased, giving it higher priority in subsequent energy and phase analysis. Simultaneously, the changes in the energy flow curvature and density gradient corresponding to the node are continuously tracked. When the energy flow curvature and density gradient are observed to gradually decrease over time and show a recovery trend, the feature weight of the node is gradually reduced based on its recovery rate to avoid long-term impacts of transient disturbances on the overall health assessment results.

[0100] Furthermore, when an abnormal region identified by the topology disturbance response persists across multiple consecutive monitoring cycles, and the corresponding topology disturbance response fails to recover to a stable range, it is determined that the key power supply components corresponding to this abnormal region face a continuous degradation risk, and the abnormal region is marked as a high-risk node in the health matrix. By marking high-risk nodes, the health matrix continuously enhances its monitoring of energy changes, phase stability, and degradation trends in this region during subsequent updates, thereby enabling early identification and risk warning of the internal degradation state of key power supply components.

[0101] Through the above steps, not only is a structured expression of the internal energy evolution of key power supply components achieved, but a quantifiable index system for degradation risk is also established through topological perturbation response. The resulting topological network possesses self-adaptive and self-correcting capabilities, continuously identifying degradation signs and dynamically adjusting energy distribution weights during operation, enabling long-term monitoring and intelligent early warning of the health status of key power supply components. This implementation method, while maintaining physical consistency and temporal continuity, elevates energy spectral analysis from the numerical level to the topological semantic level, significantly improving the accuracy of degradation diagnosis and early prediction capabilities.

[0102] This invention establishes a time-frequency coherent baseline for multi-source signals under high-temperature, high-carrier-load operation conditions and introduces a phase-continuous mapping field and a dynamic pseudo-peak constraint chain, enabling accurate identification of anomalous energy distribution clusters under a phase-consistent physical reference. Through cross-scale energy differential inversion and envelope self-reflection calibration, the trend of energy residual changes can be continuously tracked, effectively suppressing spurious energy peaks caused by harmonic coupling. This allows the AI ​​model to focus only on physically representative real degradation features during the learning phase. This approach significantly improves the interpretability of energy spectral data and the authenticity of feature extraction, enabling early exposure of early degradation signs in key power supply components and avoiding health assessment biases caused by misjudgments.

[0103] This invention achieves dynamic fusion and adaptive weight adjustment of energy and phase characteristics through time-frequency drift correction and the establishment of a topologically networked health matrix. The health matrix can dynamically adjust the weight allocation based on changes in energy flow curvature and density gradient, enabling the model to develop self-learning and self-correction capabilities during continuous monitoring. This method transforms the degradation process of key power supply components from passive diagnosis to active identification, realizing spatial localization of energy spectrum anomalies and adaptive risk level assessment. It significantly improves the stability and early warning accuracy of degradation monitoring, providing continuous intelligent support for the long-term reliable operation of power supply systems.

[0104] This invention provides, for example Figure 4The AI ​​spectrum monitoring and early warning system for performance degradation of key power supply components shown includes a phase mapping construction module, an energy differential inversion module, an envelope calibration constraint module, a drift correction and update module, and a topology weight control module.

[0105] The phase mapping construction module establishes a time-frequency coherent baseline for multi-source signals under high temperature and high carrier excitation conditions, injects a micro-amplitude phase perturbation sequence into the internal components of the power supply, and generates a phase continuous mapping field characterizing nonlinear interaction features based on the dynamic convergence trajectory of the energy distribution of each signal.

[0106] The energy differential inversion module, based on the phase continuous mapping field, constructs a cross-scale energy differential inversion chain according to the energy convergence trajectory, extracts the energy residual feature matrix by using time window sliding and frequency band resampling, and captures the transient energy aggregation region caused by harmonic coupling with the phase continuous mapping field as a constraint.

[0107] The envelope calibration constraint module performs envelope self-reflection calibration based on the energy residual feature matrix. It compares the energy change curve of the transient energy aggregation region with the historical steady-state envelope curve to detect the local energy amplification trend and feeds the calibration results back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain.

[0108] The drift correction and update module is based on a dynamic pseudo-peak constraint chain and a runtime frequency drift criterion correction mechanism. It extracts the phase stability index of the transient energy aggregation region and jointly corrects the time-frequency drift features with the envelope difference results. It performs dynamic weight backfeeding to optimize the feature learning path and generates an adaptively updatable health matrix based on the corrected phase and energy parameters.

[0109] The topology weight control module, based on the monitoring output of the health matrix, constructs the energy spectrum into a high-dimensional topology network, calculates the curvature and density gradient of energy flow between network nodes, and when local spectral lines generate abnormal aggregation, it identifies abnormal regions based on the topology perturbation response and dynamically adjusts the feature weight allocation of the AI ​​model according to the correction results.

[0110] The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components provided in this embodiment of the invention is implemented through the aforementioned AI spectrum monitoring and early warning system for performance degradation of key power supply components. For details of the specific methods and processes of the AI ​​spectrum monitoring and early warning system for performance degradation of key power supply components, please refer to the embodiments of the aforementioned AI spectrum monitoring and early warning method for performance degradation of key power supply components, which will not be repeated here.

[0111] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for AI spectrum monitoring and early warning of performance degradation of key power supply components, characterized in that, Includes the following steps: Step 1: Under preset excitation conditions, establish a time-frequency coherent baseline for multi-source signals, inject a phase perturbation sequence into the internal components of the power supply, and generate a phase continuous mapping field characterizing the nonlinear interaction features based on the dynamic convergence trajectory of the energy distribution of each signal. Step 2: Based on the phase continuous mapping field, construct a cross-scale energy difference inversion chain according to the energy convergence trajectory, extract the energy residual feature matrix using time window sliding and frequency band resampling, and capture the transient energy aggregation region caused by harmonic coupling with the phase continuous mapping field as a constraint. Step 3: Based on the energy residual feature matrix, perform envelope self-reflection calibration, compare the energy change curve of the transient energy aggregation region with the historical steady-state envelope curve, detect the local energy amplification trend, and feed the calibration results back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain. Step 4: Based on the dynamic pseudo-peak constraint chain, run the time-frequency drift criterion correction mechanism, extract the phase stability index of the transient energy aggregation region, and jointly correct the time-frequency drift features and envelope difference results. Perform dynamic weight backfeed to optimize the feature learning path, and generate an adaptively updated health matrix based on the corrected phase and energy parameters. Step 5: Based on the monitoring output of the health matrix, construct the energy spectrum into a topological network, calculate the curvature and density gradient of the energy flow between network nodes, and when local spectral lines generate abnormal aggregation, mark the abnormal region according to the topological perturbation response, and dynamically adjust the feature weight allocation of the AI ​​model according to the correction results.

2. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 1, characterized in that, A phase-continuous mapping field characterizing the nonlinear interaction features is generated based on the dynamic convergence trajectory of each signal energy distribution. The specific steps are as follows: Multi-channel synchronous signal acquisition is performed on the operation process of key power supply components under preset excitation conditions to obtain a multi-source signal matrix after anti-aliasing filtering and time base correction; After completing the construction of the time-frequency coherent baseline, a phase perturbation sequence is applied to the inside of the key power supply components. Under the condition of maintaining stable power output and ambient temperature, the multi-source signals after perturbation are collected, and the dynamic convergence trajectory of energy distribution is obtained by comparing the difference with the multi-source signal matrix before perturbation. Using a time-frequency coherent baseline as a reference, the phase changes and energy distribution trajectories of each signal after the disturbance response are mapped in the time-frequency space. The energy center frequency point is mapped to the phase offset value point by point to generate a phase continuous mapping field that characterizes the nonlinear interaction features.

3. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 2, characterized in that, The energy residual feature matrix is ​​extracted, and the transient energy aggregation region caused by harmonic coupling is captured by using a phase continuous mapping field as a constraint. The specific steps are as follows: Based on the establishment of the phase continuous mapping field, according to the energy convergence trajectory in the phase continuous mapping field, the multi-source signals are synchronously resampled in both time and frequency dimensions. Signal alignment is performed by time sliding and frequency interpolation to obtain continuous energy migration data along the time-frequency axis. Based on the aligned energy migration data, a cross-scale energy difference inversion chain is constructed with reference to the energy convergence trajectory in the phase continuous mapping field. Energy difference between different time levels forms an energy change sequence spanning multiple time scales. After the cross-scale energy differential inversion chain is formed, the energy residual feature matrix is ​​extracted by time window sliding and frequency band resampling, and the degree of deviation of energy change is characterized in the form of time-frequency spatial distribution. After obtaining the energy residual characteristic matrix, the transient energy aggregation region caused by harmonic coupling is captured based on the density of the energy residual and the phase gradient change, constrained by the phase continuous mapping field.

4. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 3, characterized in that, In the process of capturing the transient energy aggregation region caused by harmonic coupling, the energy residual feature matrix is ​​mapped point by point to the adjacent energy convergence trajectory in the phase continuous mapping field. When the energy residual continues to increase within a preset time and the phase gradient shifts continuously, the local time-frequency position that meets the condition is determined as the real energy aggregation region generated by harmonic coupling, and thus the transient energy aggregation region.

5. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 3, characterized in that, The steps to form a dynamic pseudo-peak constraint chain include: Based on the energy residual feature matrix, the energy density distribution of each signal channel is scanned point by point. Local areas where the energy residual density is greater than the density threshold range and continues to change are identified as transient energy aggregation areas, and energy change curves are extracted along the time axis. After obtaining the energy change curve, the historical steady-state envelope curve of the key power supply components under non-disturbance conditions is established, and it is aligned with the energy change curve on the time and frequency coordinates as a reference baseline for energy change. Based on the alignment of the energy change curve and the historical steady-state envelope curve, a differential comparison is performed to obtain the energy shift curve. The local energy amplification trend in the energy shift curve is detected, and transient pseudo-peak amplification and structural degradation amplification are distinguished. After identifying the local energy amplification trend, envelope autoreflection calibration is performed to map the energy shift result back to the phase continuous mapping field to correct the energy response shift, mark the spurious peak candidate region and adjust the energy weight; The envelope self-reflection calibration results are fed back to the phase continuous mapping field and stored as dynamic constraint parameters to form a dynamic pseudo-peak constraint chain.

6. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 5, characterized in that, During the envelope self-reflection calibration process, when the deviation between the energy change curve and the historical steady-state envelope curve exceeds the preset deviation threshold range, the corresponding region is marked as a pseudo-peak candidate region, and the energy weight corresponding to the pseudo-peak candidate region is reduced in the phase continuous mapping field. When the energy change returns to the historical steady-state envelope range, the energy weights corresponding to the spurious peak candidate regions are automatically restored.

7. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 1, characterized in that, The steps to generate an adaptively updated health matrix include: After the dynamic pseudo-peak constraint chain is established, its feedback result is used as the initial constraint condition to dynamically track the transient energy aggregation region in the phase continuous mapping field, record the drift trajectory of the transient energy aggregation region in the time and frequency dimensions, and identify the drift anomaly region. After determining that there is a drift trend in the transient energy aggregation region, the phase change characteristics are extracted to obtain the phase stability index. The phase stability index of each transient energy aggregation region is then mapped back to the phase continuous mapping field to form a global phase stability distribution map. Based on the extracted phase stability index, the time-frequency drift characteristics and envelope difference results are jointly corrected to synchronously adjust the phase response and energy intensity of the drift anomaly region and mark the potential degradation feature region. After joint correction, dynamic weight backfeed is performed, and the characteristic weights of each transient energy aggregation region in the phase continuous mapping field are dynamically adjusted according to the corrected drift information. After the dynamic weight refeeding is completed, an adaptively updated health matrix is ​​generated based on the corrected phase and energy parameters.

8. The AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components according to claim 7, characterized in that, The steps for dynamically adjusting the feature weight allocation of the AI ​​model based on the calibration results include: After the health matrix is ​​generated, the energy and phase parameters are converted into nodes and connections of the topology network, and a multidimensional energy topology space is constructed based on the energy flow direction and rate of change. After establishing the topology network, the curvature and density gradient of energy flow between network nodes are calculated to form a time-varying feature field describing the dynamic changes of energy transfer paths. When a local spectral line anomalous convergence is detected in the time-varying characteristic field of energy flow curvature and density gradient, the anomalous region is calibrated based on the topological perturbation response, and the magnitude and duration of the topological perturbation response are recorded. After the abnormal regions are identified, the topological perturbation response is fused with the health matrix correction data, and the feature weight allocation is dynamically adjusted according to the correction results. The weight is increased for persistent abnormal regions and decreased for transient abnormal regions. The dynamically adjusted feature weight distribution is fed back to the health matrix to complete the continuous monitoring and intelligent early warning of degradation risks of key power supply components.

9. The method for AI spectrum monitoring and early warning of performance degradation of key power supply components according to claim 8, characterized in that, The process of dynamically adjusting feature weight allocation includes: when the topological perturbation response exceeds the preset response threshold range, the feature weight of the corresponding node is increased in real time, and the feature weight is gradually reduced according to the recovery speed of energy flow curvature and density gradient; when the abnormal region persists and the topological perturbation response has not recovered to a stable range, the corresponding abnormal region is marked as a high-risk node in the health matrix.

10. An AI spectrum monitoring and early warning system for performance degradation of key power supply components, used to implement the AI ​​spectrum monitoring and early warning method for performance degradation of key power supply components as described in any one of claims 1-9, characterized in that, It includes a phase mapping construction module, an energy difference inversion module, an envelope calibration constraint module, a drift correction and update module, and a topology weight adjustment module; The phase mapping construction module establishes a time-frequency coherent baseline for multi-source signals under preset excitation conditions, injects a phase perturbation sequence into the power supply key components, and generates a continuous phase mapping field characterizing nonlinear interaction features based on the dynamic convergence trajectory of the energy distribution of each signal. The energy differential inversion module, based on the phase continuous mapping field, constructs a cross-scale energy differential inversion chain according to the energy convergence trajectory, extracts the energy residual feature matrix by using time window sliding and frequency band resampling, and captures the transient energy aggregation region caused by harmonic coupling with the phase continuous mapping field as a constraint. The envelope calibration constraint module performs envelope self-reflection calibration based on the energy residual feature matrix. It compares the energy change curve of the transient energy aggregation region with the historical steady-state envelope curve to detect the local energy amplification trend and feeds the calibration results back to the phase continuous mapping field to form a dynamic pseudo-peak constraint chain. The drift correction and update module is based on a dynamic pseudo-peak constraint chain and a runtime frequency drift criterion correction mechanism. It extracts the phase stability index of the transient energy aggregation region and jointly corrects the time-frequency drift features with the envelope difference results. It performs dynamic weight backfeeding to optimize the feature learning path and generates an adaptively updated health matrix based on the corrected phase and energy parameters. The topology weight control module, based on the monitoring output of the health matrix, constructs the energy spectrum into a topology network, calculates the curvature and density gradient of energy flow between network nodes, and when local spectral lines generate abnormal aggregation, it identifies abnormal regions based on the topology perturbation response and dynamically adjusts the feature weight allocation of the AI ​​model according to the correction results.

Images