A method and system for predicting thermal failure of generator rectifier bridge based on digital twins

By constructing a thermal-electric coupling state fingerprint vector and an aging state parameter cluster template library, millisecond-level online adaptive calibration of generator rectifier bridge thermal failure was achieved. This solves the problems of low prediction accuracy and high computational resource consumption in existing technologies, improves prediction accuracy and response sensitivity, and reduces operation and maintenance costs.

CN123115935APending Publication Date: 2026-11-17AOBOHUA ELECTRONIC APPLIANCE TAIAN CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611237834.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to achieve millisecond-level online adaptive calibration for predicting thermal failure of generator rectifier bridges. Model parameter drift cannot be captured quickly in real time, resulting in low prediction accuracy, high computational resource consumption, and a lack of physical interpretability in parameter adjustments. Consequently, these technologies are ill-suited for high availability requirements and environments with limited edge computing power.

Method used

By collecting the transient junction temperature distribution and diode forward voltage drop waveforms of key nodes in the generator rectifier bridge, a thermal-electric coupling state fingerprint vector is constructed. The fingerprint similarity is retrieved using an aging parameter cluster template library, and a calibrated digital twin simulation model is generated. The fingerprint drift is monitored in real time and a thermal failure risk warning is output. Parameter adjustment is performed by combining edge-embedded Bayesian inference.

Benefits of technology

It improves the accuracy and robustness of generator rectifier bridge thermal failure prediction, reduces computational burden and system outage risk, enhances the response sensitivity and engineering practicality of model updates, and reduces operation and maintenance costs and communication dependence.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115935A_ABST
    Figure CN123115935A_ABST
Patent Text Reader

Abstract

This invention relates to a method and system for predicting thermal failure of a generator rectifier bridge based on digital twins. By deploying miniature thermocouple arrays and high-speed voltage probes at key nodes of the rectifier bridge, synchronous acquisition of transient junction temperature and on-state voltage drop sequences under multiple operating conditions is achieved. A thermo-electric coupling state fingerprint vector is constructed through time-frequency coding and multi-source data fusion. Combining laboratory aging test results, an aging state parameter cluster template library is established, and real-time similarity retrieval is used to complete aging state determination and parameter replacement of the digital twin model. Subsequently, the digital twin model extrapolates thermal failure trends based on the target parameter clusters, and adaptive parameter fine-tuning is achieved using a Bayesian inference engine, ultimately outputting a thermal failure risk warning signal. This technology improves the accuracy of rectifier bridge aging state identification and thermal failure risk prediction capabilities, achieving high real-time performance and adaptability of the model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of thermal-electric coupling fault prediction and adaptive calibration technology of digital twin models for power electronic devices, and in particular to a method and system for predicting thermal failure of generator rectifier bridges based on digital twins. Background Technology

[0002] Current technologies for predicting thermal failure of generator rectifier bridges and for model self-calibration mainly employ three approaches. The first approach focuses on static parameter calibration using empirical models. This involves obtaining model parameters offline and manually updating them periodically during operation. Relying on a fixed calibration frequency and manual intervention, it cannot handle gradual parameter drift or abrupt changes. The second approach is dominated by residual feedback or data-driven online calibration methods. These utilize techniques such as LSTM, residual neural networks, and particle filtering assimilation to use the difference between the model output and online observations as error terms to drive parameter adjustments. However, this approach relies on continuous historical data and extensive online training, resulting in high computational resource consumption, high latency, and a lack of physical interpretability for parameter adjustments. The third approach incorporates fusion algorithms such as augmented generative models, multi-objective decision engines, and integrated Bayesian inference. However, this approach has high computational complexity, making it difficult to balance parameter update frequency with operating condition switching response speed, and is challenging to implement under conditions requiring uninterrupted equipment operation.

[0003] Existing technologies are primarily suitable for scenarios with small parameter variations and tolerance for offline downtime calibration. However, they struggle to achieve millisecond-level online adaptive calibration for rectifier bridges operating in real-world environments. The inability to capture model parameter drift in real time leads to deviations between the digital twin model's simulation output and the actual state, impacting the accuracy of thermal failure prediction. Residual-driven or optimization iteration schemes are highly dependent on computational resources and historical data, and parameter adjustment results lack traceability based on physical mechanisms. Furthermore, existing technologies require frequent manual calibration, equipment downtime, or external database intervention, making them unsuitable for high availability requirements and environments with limited edge computing power. Summary of the Invention

[0004] This application provides a method and system for predicting the thermal failure of a generator rectifier bridge based on digital twins, aiming to solve one of the problems or issues of the prior art mentioned in the background.

[0005] This application provides a method and system for predicting the thermal failure of a generator rectifier bridge based on digital twins, specifically including:

[0006] S1: Collect the transient junction temperature distribution sequence and diode forward voltage drop waveform of key nodes of generator rectifier bridge under typical operating conditions, and generate the original thermal-electric coupling sampling dataset by correlation packaging;

[0007] S2: Perform time-frequency joint coding on the original thermal-electric coupling sampling dataset, extract multi-dimensional time-frequency domain feature parameters, and construct a thermal-electric coupling state fingerprint vector;

[0008] S3: Based on the failure paths of solder joint cold solder joints, contact oxidation, and thermal stress cracking covered by laboratory accelerated aging tests, the thermal-electric coupling state fingerprint vector is used as an index to pre-set an aging state parameter cluster template library.

[0009] S4: Based on the thermal-electric coupling state fingerprint vector, perform fingerprint similarity retrieval and template index positioning operations, and output the current rectifier bridge aging state attribution result and the corresponding target parameter cluster template index identifier;

[0010] S5: Retrieve the corresponding target parameter cluster from the aging parameter cluster template library according to the target parameter cluster template index identifier, inject it into the preset digital twin model to replace the physical parameters and generate a calibrated digital twin simulation model;

[0011] S6: Enable the calibrated digital twin simulation model to perform thermal failure trend deduction for the next operating cycle, synchronously calculate the fingerprint drift distance value within several consecutive cycles, and generate a real-time fingerprint drift monitoring sequence.

[0012] S7: If the fingerprint drift distance value exceeds the preset threshold and no new template matching is triggered, the retrieved target parameter cluster is adjusted, and the final thermal failure risk warning signal is output.

[0013] This application provides a method and system for predicting the thermal failure of a generator rectifier bridge based on digital twins, which has the following advantages:

[0014] This scheme mines the inherent thermo-electric coupling response fingerprint characteristics during the physical degradation of devices. It utilizes a miniature thermocouple array and a high-speed voltage probe to simultaneously acquire transient junction temperature distribution and on-state voltage drop waveforms under multiple operating conditions. Combined with time-frequency joint coding, it extracts a three-dimensional joint characterization of wavelet packet energy spectrum, phase consistency matrix, and thermal diffusion hysteresis time window, forming a state fingerprint with strong aging stage identification capabilities. This improves the accuracy and robustness of degradation state identification and reduces the dependence of traditional methods on complex training and residual feedback. By establishing a parameter cluster template library covering the entire lifecycle aging range and designing a template indexing mechanism, only fingerprint similarity detection is performed during online runtime. The system can quickly locate the best-matching aging interval template and complete parameter injection, improving the response sensitivity and engineering practicality of model updates, and reducing the computational burden and system interruption risk during parameter updates. By introducing a closed-loop adaptive maintenance mechanism, when fingerprint drift exceeds the threshold and no new template matching is triggered, a low-frequency auxiliary calibration process is automatically started. The edge-embedded Bayesian inference engine is called to locally fine-tune the weights of the loaded parameter clusters, improving the system's adaptability to extrapolation and generalization performance, and reducing the risk caused by blind spots in the static template library coverage. The overall architecture relies on local perception-matching-injection closed-loop operation, reducing operation and maintenance costs and communication dependencies. Attached Figure Description

[0015] Figure 1 This is the main flowchart of a generator rectifier bridge thermal failure prediction method and system based on digital twins;

[0016] Figure 2 This is a sub-flowchart of a generator rectifier bridge thermal failure prediction method and system based on digital twins;

[0017] Figure 3 This is another sub-flowchart of a generator rectifier bridge thermal failure prediction method and system based on digital twins. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0020] like Figure 1 As shown, this application provides a method and system for predicting the thermal failure of a generator rectifier bridge based on digital twins, specifically including:

[0021] S1: Collect the transient junction temperature distribution sequence and diode forward voltage drop waveform of key nodes of generator rectifier bridge under typical operating conditions, and generate the original thermal-electric coupling sampling dataset by correlation packaging;

[0022] S2: Perform time-frequency joint coding on the original thermal-electric coupling sampling dataset, extract multi-dimensional time-frequency domain feature parameters, and construct a thermal-electric coupling state fingerprint vector;

[0023] S3: Based on the failure paths of solder joint cold solder joints, contact oxidation, and thermal stress cracking covered by laboratory accelerated aging tests, the thermal-electric coupling state fingerprint vector is used as an index to pre-set an aging state parameter cluster template library.

[0024] S4: Based on the thermal-electric coupling state fingerprint vector, perform fingerprint similarity retrieval and template index positioning operations, and output the current rectifier bridge aging state attribution result and the corresponding target parameter cluster template index identifier;

[0025] S5: Retrieve the corresponding target parameter cluster from the aging parameter cluster template library according to the target parameter cluster template index identifier, inject it into the preset digital twin model to replace the physical parameters and generate a calibrated digital twin simulation model;

[0026] S6: Enable the calibrated digital twin simulation model to perform thermal failure trend deduction for the next operating cycle, synchronously calculate the fingerprint drift distance value within several consecutive cycles, and generate a real-time fingerprint drift monitoring sequence.

[0027] S7: If the fingerprint drift distance value exceeds the preset threshold and no new template matching is triggered, the retrieved target parameter cluster is adjusted, and the final thermal failure risk warning signal is output.

[0028] Step S1: Acquire the transient junction temperature distribution sequence and diode forward voltage drop waveform of key nodes in the generator rectifier bridge under typical operating conditions, and generate the original thermo-electric coupling sampling dataset through correlation packaging. The transient junction temperature distribution sequence and diode forward voltage drop waveform are acquired by a miniature thermocouple array and high-speed voltage probe deployed at the key nodes of the generator rectifier bridge. The typical operating conditions include at least one of no-load, half-load, and full-load. Specifically, they include:

[0029] S1.1: Obtain the generator rectifier bridge operation status monitoring command, and based on the preset threshold values ​​of three typical operating conditions (no load, half load, and full load), perform real-time range determination on the rectifier bridge load current signal and generate an operating condition trigger signal containing the current operating condition type identifier.

[0030] Receive the generator rectifier bridge operation status monitoring command, parse the sampling frequency, operating condition judgment threshold and time window parameters contained in the command, and initialize the buffer pointer and interrupt trigger logic of the load current signal acquisition module.

[0031] The instantaneous value sequence of the DC-side load current of the rectifier bridge output by the Hall sensor is read in real time. The moving average filter is used to remove high-frequency electromagnetic interference noise and extract a smooth current characteristic curve that reflects the actual load level of the rectifier bridge.

[0032] The smoothed load current characteristic curve is compared with the preset threshold ranges for three typical operating conditions: no-load, half-load, and full-load. The no-load threshold is set to 0% to 5% of the rated current, the half-load threshold is set to 45% to 55% of the rated current, and the full-load threshold is set to 95% to 100% of the rated current.

[0033] When the load current is detected to remain within the current operating condition threshold range for more than the preset minimum dwell time window, the current rectifier bridge is determined to be in a stable operating state, and a status confirmation signal containing the current operating condition type identifier is generated.

[0034] Based on the status confirmation signal, a status trigger signal data packet containing the status type code, start timestamp, and effective data length information is constructed and sent to the multi-sensor synchronous acquisition and control module via the internal bus.

[0035] S1.2: Receive the operating condition trigger signal, use a high-precision synchronous clock source to perform microsecond-level time alignment calibration on the miniature thermocouple array and high-speed voltage probe deployed at the key nodes of the rectifier bridge, and generate a multi-sensor synchronous acquisition control command with a unified time reference.

[0036] Receive the working condition trigger signal generated in S1.1, extract the working condition type identifier and timestamp reference information contained therein, and generate the input condition parameter set for the multi-sensor synchronous acquisition control command.

[0037] The high-precision synchronous clock source module is invoked to read the current system global time base, calculate the hardware transmission delay difference between the micro thermocouple array and the high-speed voltage probe, and generate microsecond-level time alignment compensation parameters.

[0038] Based on the time alignment compensation parameters, phase advance correction is performed on the sampling start time of the micro thermocouple array, and phase lag correction is performed on the trigger threshold window of the high-speed voltage probe to generate corrected sampling time parameters after eliminating timing deviations.

[0039] A unified timestamp mapping matrix is ​​constructed to strictly align the corrected thermocouple sampling points with the voltage probe sampling points in the time domain, ensuring that thermoelectric data at the same physical moment has a unique corresponding time index, and thus generating the timestamp mapping matrix.

[0040] Based on the timestamp mapping matrix, a multi-sensor synchronous acquisition control command with a unified time reference is generated. The command includes a time-calibrated sampling frequency, trigger edge polarity, and data packaging format, and is output to step S1.3.

[0041] S1.3: Based on the multi-sensor synchronous acquisition control command, drive the micro thermocouple array to capture the transient change trajectory of the rectifier bridge diode junction temperature within a single cycle, and simultaneously drive the high-speed voltage probe to record the diode forward voltage drop waveform at the corresponding moment, generating the original transient junction temperature distribution sequence and the original forward voltage drop waveform sequence containing timestamps.

[0042] It receives control commands for synchronous acquisition from multiple sensors, parses the parameters including sampling frequency, trigger delay, and channel mapping relationship, and initializes the underlying drive registers of the micro thermocouple array and high-speed voltage probe.

[0043] Based on the microsecond-level time reference in the instruction, a synchronous start pulse is sent to the thermocouple signal conditioning circuit to activate the high-precision analog-to-digital converters of each channel, and capture the transient temperature changes of the rectifier bridge diode junction region at a preset high sampling rate.

[0044] The analog voltage signal output by the thermocouple array is read in real time, and linearization compensation and cold junction temperature correction are performed to convert the voltage value into an absolute temperature value, generating a raw transient junction temperature distribution sequence with high-precision timestamps.

[0045] A synchronous trigger signal is sent to the high-speed voltage probe in parallel. The differential amplifier circuit is used to isolate common-mode interference, and the instantaneous potential difference across the diode is collected to record the details of voltage drop fluctuations during conduction.

[0046] The acquired voltage waveforms are denoised and baseline corrected. The peak and average values ​​of the on-state voltage drop within each switching cycle are extracted to generate the original on-state voltage drop waveform sequence that is strictly aligned with the temperature sequence time.

[0047] By using a hardware timer interrupt mechanism, the precise correspondence between temperature data points and voltage data points on the microsecond-level time axis is ensured, eliminating phase errors caused by differences in sensor response time.

[0048] The generated original transient junction temperature distribution sequence and the original on-state voltage drop waveform sequence are packaged into a structured data stream, and operating condition type identifier and cycle counter metadata are added.

[0049] S1.4: Perform multi-source data fusion processing on the original transient junction temperature distribution sequence and the original on-state voltage drop waveform sequence, and perform structured correlation mapping between the temperature spatial distribution data and voltage time series data at the same timestamp to generate a preliminary thermo-electric coupling data frame with multi-condition spatiotemporal characteristic correlation.

[0050] The system receives the original transient junction temperature distribution sequence and the original on-state voltage drop waveform sequence generated by S1.3, which contain timestamps, as input objects for multi-source data fusion.

[0051] The original transient junction temperature distribution sequence and the original on-state voltage drop waveform sequence are subjected to a data alignment operation based on a unified time reference. A microsecond-level synchronous clock source is used to eliminate the sensor acquisition delay difference and construct a spatiotemporally aligned thermoelectric data index table.

[0052] Based on the aforementioned thermo-electric data index table, the spatial distribution matrix of diode junction temperature and the scalar value of on-state voltage drop at the same timestamp are extracted to establish a point-to-point mapping relationship between temperature field and electric field intensity.

[0053] Linear interpolation is used to resample non-uniform sampling points to ensure strict consistency between temperature spatial grid nodes and voltage time-series sampling points in the time dimension, thereby generating standardized thermal-electric coupling data pairs.

[0054] The standardized temperature spatial distribution data and voltage time series data are structured and encapsulated, and operating condition type identifiers and cycle sequence number metadata are added to form a preliminary thermo-electric coupling data frame with multi-dimensional attribute labels.

[0055] S1.5: Integrate the preliminary thermal-electric coupling data frames generated within multiple consecutive power frequency cycles, classify, aggregate, and standardize them according to the operating condition type identifier, and finally generate the original thermal-electric coupling sampling dataset.

[0056] Step S2: Perform time-frequency joint coding on the original thermal-electric coupling sampling dataset to extract multi-dimensional time-frequency domain feature parameters and construct a thermal-electric coupling state fingerprint vector. The multi-dimensional time-frequency domain feature parameters include: wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters. Specifically, this includes:

[0057] S2.1: Obtain the original thermal-electric coupling sampling dataset, use the wavelet packet transform algorithm to perform multi-scale decomposition on the transient junction temperature distribution sequence and diode on-state voltage drop waveform, calculate the energy proportion of each frequency band node, and generate wavelet packet energy spectrum components containing frequency domain energy distribution characteristics.

[0058] The original thermo-electric coupling sampling dataset is acquired, which contains a time-aligned and calibrated transient junction temperature distribution sequence and diode on-state voltage drop waveform. Discrete wavelet packet transform (DPT) is performed on the transient junction temperature distribution sequence and on-state voltage drop waveform, using the Daubechies-db4 wavelet basis function as the decomposition kernel and setting the decomposition level to 3 levels, dividing the signal frequency band into 8 equal-width sub-band nodes. The sum of squares of the wavelet packet decomposition coefficients for each sub-band node is calculated to obtain the energy value of each node, reflecting the degree of energy concentration of the signal within a specific frequency band. The energy values ​​of all sub-band nodes are normalized, and the ratio of the energy of each node to the total energy is calculated to construct a wavelet packet energy spectral component vector characterizing the frequency domain energy distribution of the signal.

[0059] S2.2: Based on the time-frequency coefficients corresponding to the wavelet packet energy spectrum components, perform Hilbert-Huang transform processing to extract instantaneous phase information, calculate the statistical correlation matrix of the signal phase angle under different operating conditions, and generate phase consistency matrix elements characterizing the synchronization characteristics of the thermoelectric signal.

[0060] Obtain the time-frequency coefficient matrix corresponding to the wavelet packet energy spectrum components generated in step S2.1, and use it as the input signal source for the Hilbert-Huang transform. Perform empirical mode decomposition on the time-frequency coefficient sequence. By selecting extreme points and the mean of the envelope, the non-stationary thermoelectric signal is decomposed into several intrinsic mode function components, eliminating trend term interference and retaining local oscillation characteristics. Perform Hilbert transform on the intrinsic mode function components to construct an analytic signal to extract the instantaneous phase angle. Calculate the rate of change of phase with time at each moment to generate an instantaneous phase time sequence reflecting the synchronicity of the thermoelectric coupling dynamic response. For the instantaneous phase time sequences under different operating conditions, calculate the cross-correlation coefficients of the phase angles under no-load, half-load, and full-load conditions, and construct a statistical correlation matrix characterizing the phase lag relationship between multiple operating conditions. Perform eigenvalue decomposition on the statistical correlation matrix, extract the principal component vectors as elements of the phase consistency matrix, and quantify the coupling tightness between the internal thermal diffusion path and the carrier transport process of the rectifier bridge.

[0061] S2.3: Using the thermal response start point identified by the phase consistency matrix elements, combined with the rise slope change of the transient junction temperature distribution sequence, the cross-correlation analysis method is used to calculate the time delay of heat transfer from the junction region to the heat dissipation interface, and generate thermal diffusion hysteresis time window parameters that reflect the degree of heat dissipation interface degradation.

[0062] Obtain the elements of the phase consistency matrix generated in S2.2, extract the time indices corresponding to the peak values ​​of cross-correlation coefficients near the diagonal of the matrix, and map them to the thermal response start time points. This time point characterizes the physical trigger moment when the Joule heat generated at the instant of current conduction begins to transfer to the semiconductor junction region and packaging structure.

[0063] The raw transient junction temperature distribution sequence acquired by S1.3 is read, and the temperature sampling point aligned with the thermal response start time is located. The rising slope of the temperature curve within a preset time window after this point is calculated. The maximum temperature rise rate and its corresponding time interval are determined by a linear regression fitting algorithm, identifying the key turning point where heat accumulates in the junction region and overcomes the interfacial thermal resistance.

[0064] A cross-correlation analysis function is constructed, using the diode forward voltage drop waveform as the reference signal and the transient junction temperature distribution sequence as the target signal. The sliding cross-correlation coefficient sequence between the two in the time domain is calculated. The cross-correlation calculation formula is as follows:

[0065]

[0066] Where R(τ) is the cross-correlation value under time delay τ, V(t) is the on-state voltage drop sequence, T(t) is the junction temperature sequence, and N is the number of sampling points.

[0067] The location of the global maximum value in the cross-correlation coefficient sequence is searched to determine the optimal time delay parameter that maximizes the correlation of the thermoelectric signals. This optimal time delay parameter is the heat transfer hysteresis time between the change in on-state voltage drop and the junction temperature response, which characterizes the time delay required for heat to be conducted from the junction region through the packaging structure to the heat dissipation interface, and is defined as the heat diffusion hysteresis time.

[0068] The optimal time delay parameter is weighted and corrected by incorporating the rate of change of the temperature rise slope to eliminate unsteady-state heat transfer errors caused by load fluctuations, thus generating the final thermal diffusion hysteresis time window parameter. This parameter quantifies the increase in contact thermal resistance caused by the aging of the heat dissipation interface material.

[0069] S2.4: Receive the wavelet packet energy spectrum components, phase consistency matrix elements and thermal diffusion hysteresis time window parameters, perform a three-dimensional vector space mapping operation, and orthogonally concatenate the feature parameters of the above three dimensions according to preset weights to generate a thermal-electric coupling state fingerprint vector.

[0070] The wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters generated by S2.1 to S2.3 are received as the original feature inputs for constructing the three-dimensional state fingerprint.

[0071] The wavelet packet energy spectrum components are subjected to Z-Score normalization to eliminate the differences in energy dimensions between different frequency bands and generate a normalized frequency domain energy feature vector.

[0072] Eigenvalue decomposition is performed on the elements of the phase consistency matrix, and principal component eigenvalues ​​are extracted as scalar indices characterizing thermal-electric coupling synchronization, generating normalized phase synchronization eigenvalues.

[0073] The thermal diffusion hysteresis time window parameter is mapped to the [0,1] interval using the Mi-Max normalization method to eliminate the influence of time scale fluctuations and generate a normalized thermal hysteresis characteristic scalar.

[0074] Based on preset weighting coefficients, the normalized frequency domain energy feature vector, normalized phase synchronization feature value, and normalized thermal hysteresis feature scalar are orthogonally spliced ​​together to construct a high-dimensional feature space mapping matrix.

[0075] The Euclidean norm of the thermal-electric coupling state fingerprint vector is calculated using the following formula, which is used for benchmark calibration of subsequent similarity matching:

[0076]

[0077] Where F is the fingerprint vector magnitude, w i Let x be the preset weight of the i-th dimension feature. i Let be the normalized eigenvalue of the i-th dimension, and N be the total dimension of the features.

[0078] The concatenated high-dimensional feature matrix is ​​associated and encapsulated with the fingerprint vector magnitude to generate a uniquely identified thermal-electric coupling state fingerprint vector.

[0079] like Figure 2 As shown, step S3: Based on the solder joint failure paths (poor soldering, contact oxidation, and thermal stress cracking) covered by the accelerated aging test in the laboratory, a pre-set aging state parameter cluster template library is established using the thermal-electric coupling state fingerprint vector as an index. The aging state parameter cluster template library includes thermal resistance network topology, carrier recombination coefficient correction terms, and thermal capacity attenuation factors. Specifically, it includes:

[0080] S3.1: Obtain the original degradation sample set covering the failure paths of solder joint malfunction, contact oxidation, and thermal stress cracking in the accelerated aging test in the laboratory. Perform multi-physics coupling mechanism analysis on the original degradation sample set, extract the thermal resistance evolution law, carrier recombination coefficient correction term change law, and thermal capacity decay factor change curve under different aging stages, and generate aging stage feature dataset.

[0081] Obtain the original degradation sample set covering the failure paths of solder joint malfunction, contact oxidation, and thermal stress cracking in laboratory accelerated aging tests. The original degradation sample set contains transient thermal response data and electrical characteristic parameters of the rectifier bridge under different aging times, which serve as the input data source for multi-physics coupling mechanism analysis.

[0082] Multiphysics coupling mechanism analysis was performed on the original degradation sample set. A three-dimensional transient thermal model was constructed based on Fourier's heat conduction law. The heat flux density distribution from the junction to the heat dissipation substrate was calculated. The equivalent thermal resistance evolution curve characterizing the interface contact degradation was extracted, and a dataset of thermal resistance evolution laws under different aging stages was generated.

[0083] By combining the semiconductor carrier transport equation, the nonlinear drift characteristics of the forward voltage drop of a diode with temperature are analyzed, the variation law of the carrier recombination coefficient correction term during the aging process is derived, and a dataset of the variation law of the carrier recombination coefficient correction term under different aging stages is generated.

[0084] Differential scanning calorimetry was used to measure the change in specific heat capacity of the rectifier bridge module before and after aging. Combined with mass loss rate data, the heat capacity decay effect caused by microcrack propagation in the packaging material was quantified, and a dataset of heat capacity decay factor change curves under different aging stages was generated.

[0085] The datasets of thermal resistance evolution, carrier recombination coefficient correction term variation, and thermal capacity decay factor variation are spatiotemporally aligned and fused according to the aging time axis to generate an aging stage feature dataset containing physical degradation characteristics, and then output to step S3.2.

[0086] S3.2: Based on the aging stage feature dataset, the parameters of each aging interval are solved in reverse using finite element simulation inversion processing. The values ​​of the thermal resistance network parameters, the carrier recombination coefficient correction term, and the thermal capacity attenuation factor coefficient corresponding to the thermal resistance network topology evolution law are calculated, generating a candidate parameter cluster set. This finite element simulation inversion processing is completed offline in the laboratory, and the generated candidate parameter cluster set is stored as a pre-set content in a template library, without occupying computing resources during online operation.

[0087] The system receives the characteristic dataset of the aging stage and extracts the transient thermal response boundary conditions and electrical loss distribution curves corresponding to different aging intervals, which are used as the initial input constraints for finite element simulation inversion processing.

[0088] A multi-physics coupled finite element model of the rectifier bridge is constructed, and the topology parameters of the thermal resistance network are mapped to the thermal conductivity tensor of the material. The carrier recombination coefficient correction term is mapped to the recombination lifetime parameter of the semiconductor region. The thermal capacity decay factor is mapped to the specific heat capacity temperature dependence function.

[0089] The objective function is set as minimizing the root mean square error between the simulated junction temperature waveform and the measured transient junction temperature distribution sequence. A regularization term is introduced to constrain the physical rationality of the parameters and prevent parameter divergence caused by overfitting. The specific formula is as follows:

[0090]

[0091] Where J is the total loss function, T sim To simulate junction temperature, T meas θ represents the measured junction temperature, N represents the number of sampling points, λ represents the regularization coefficient, and θ represents the vector of parameters to be optimized.

[0092] The adjoint variable method is used to calculate the sensitivity matrix of the objective function to thermal resistance, composite coefficient and heat capacity parameters. The conjugate gradient descent algorithm is used to iteratively update the parameter values ​​until the loss function converges to below the preset threshold.

[0093] The optimal parameter combination after convergence is extracted, including the equivalent thermal resistance between nodes, the value of the carrier recombination coefficient correction term, and the value of the effective thermal capacity attenuation factor of the encapsulation material, forming a set of candidate parameter clusters.

[0094] By using finite element simulation inversion processing, the characteristic data of the aging stage are transformed into quantitative physical parameters, generating a set of candidate parameter clusters with high fidelity, and realizing accurate matching between the parameters of the digital twin model and the aging state of the physical entity.

[0095] S3.3: Perform association mapping processing on the thermal-electric coupling state fingerprint vector and the candidate parameter cluster set to establish the correspondence between the fingerprint feature space and the parameter physical space, and generate a fingerprint parameter mapping index table.

[0096] Obtain the thermal-electric coupling state fingerprint vector generated in the previous step. This vector contains wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters, which serve as the input feature set for the mapping process.

[0097] Read the candidate parameter set generated in step S3.2. This set contains physical parameter groups such as thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor corresponding to different aging stages.

[0098] A nonlinear mapping function from the high-dimensional feature space to the physical parameter space is constructed. The radial basis function kernel method is used to calculate the Euclidean distance between the fingerprint vector and the center point of the candidate parameter cluster. The feature similarity is quantified, and an initial matching degree weight sequence between each candidate parameter cluster and the three-dimensional thermo-electric coupling state fingerprint vector is generated.

[0099] The matching weight between the fingerprint vector and the k-th candidate parameter cluster is calculated using the following formula:

[0100]

[0101] Among them, F vec For real-time 3D state fingerprint vectors, C k Let be the feature center vector of the k-th candidate parameter cluster, and ε be a small constant to prevent division by zero.

[0102] The initial matching degree weight sequence is normalized to ensure that the sum of the matching degree weights of all candidate clusters is 1, thereby generating a sequence of association strength indexes in the form of a probability distribution.

[0103] The candidate parameter cluster index corresponding to the maximum matching weight in the association strength index sequence is selected as the main mapping target, and the second largest weight value is recorded to generate the confidence evaluation result of the mapping relationship.

[0104] A bidirectional index link is established between the fingerprint feature identifier and the physical address of the parameter cluster. The main mapping target, the confidence evaluation result, and the current timestamp are associated and encapsulated to generate a structured mapping record containing the fingerprint hash value, parameter cluster ID, confidence score, and timestamp.

[0105] The structured mapping record is written into the fingerprint parameter mapping index table in memory to generate the fingerprint parameter mapping index table, and then output to step S3.4.

[0106] S3.4: Based on the fingerprint parameter mapping index table, the candidate parameter cluster set is standardized and encapsulated, and the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor are integrated into atomic data units to generate an aging state parameter cluster template.

[0107] Obtain the fingerprint parameter mapping index table and the corresponding candidate parameter cluster set generated in S3.3, and parse the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor associated with each aging state interval in the index table as input data for standardized encapsulation processing.

[0108] The thermal resistance network topology is serialized and encoded by converting the node thermal resistance value and the inter-node coupling thermal capacity value into a fixed-length binary byte stream to generate a serialized data block of the thermal resistance network topology, ensuring the consistency of the topology description under different aging stages.

[0109] The carrier recombination coefficient correction term is subjected to floating-point normalization and memory alignment according to the IEEE 754 double-precision standard to generate a normalized data block of the carrier recombination coefficient correction term, thereby eliminating calculation errors caused by differences in data precision.

[0110] The heat capacity decay factor is normalized and compressed to a dimensionless range of 0 to 1, and timestamp metadata is added to identify the experimental batch information generated by the parameter cluster, thus generating a normalized data block of the heat capacity decay factor.

[0111] The serialized data block of the thermal resistance network topology, the normalized data block of the carrier recombination coefficient correction term, and the normalized data block of the thermal capacity attenuation factor are concatenated to construct a data packet structure containing a header identifier, a load length field, and a checksum, thereby generating a preliminary encapsulated data packet.

[0112] The SHA-256 hash algorithm is used to calculate the digest of the contents of the initially encapsulated data packet, generating a unique hexadecimal string, which serves as the globally unique index identifier of the aging parameter cluster template.

[0113] The globally unique index identifier is written into the header of the initially encapsulated data packet to generate an atomic data unit with a unique index identifier that cannot be tampered with, which serves as an aging state parameter cluster template.

[0114] S3.5: Receive all generated aging parameter cluster templates, write them into a non-volatile storage medium for structured storage, and build an aging parameter cluster template library containing full life cycle aging characteristics to support real-time fingerprint similarity retrieval and template indexing and positioning operations.

[0115] Step S4: Based on the thermal-electric coupling state fingerprint vector, perform fingerprint similarity retrieval and template index localization operations, and output the current rectifier bridge aging state attribution result and the corresponding target parameter cluster template index identifier. Specifically, this includes:

[0116] S4.1: Obtain the real-time generated thermal-electric coupling state fingerprint vector as input data, and based on the predefined graph node mapping rules, map the wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion lag time window parameters in the thermal-electric coupling state fingerprint vector to query graph node attributes respectively, and generate a query graph structure.

[0117] The real-time thermal-electric coupling state fingerprint vector output from step S2 is received. This vector contains wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters, which serve as input data for constructing the query graph structure.

[0118] Based on predefined graph node mapping rules, the three-dimensional features of the thermal-electric coupling state fingerprint vector are instantiated as node attributes in the query graph structure, generating an initialized query graph node set. Specifically, the first type of node maps to the high-frequency and low-frequency energy proportions of the wavelet packet energy spectrum, assigning it a weight attribute representing the distribution of local hotspots; the second type of node maps to the diagonal elements and off-diagonal correlation coefficients of the phase consistency matrix, assigning it a connection weight representing the synchronicity of the thermal-electric response; and the third type of node maps to the start time and duration of the thermal diffusion lag time window, assigning it a temporal weight representing the change in thermal resistance at the heat dissipation interface.

[0119] Establish directed edge connections between nodes in the query graph node set, set edge weights according to physical causality, point energy distribution nodes to phase synchronization nodes, and then point phase synchronization nodes to thermal hysteresis nodes to generate an initial query graph topology with physical meaning.

[0120] Normalization is performed on the initial query graph topology to eliminate the influence of dimensional differences on subsequent graph isomorphic matching, ensuring that the attribute values ​​of each node are in the same numerical range, generating the normalized query graph structure, and outputting it to step S4.2.

[0121] S4.2: Perform topological alignment processing using the query graph structure and the pre-stored reference graph template set in the aging parameter cluster template library. Use a subgraph isomorphic matching algorithm based on adjacency matrix sparsification to calculate the structural similarity metric between the query graph and each reference graph template, and generate a similarity score sequence.

[0122] Obtain the query graph structure generated by S4.1 and perform topological alignment of its node attributes with the pre-stored reference graph template set in the aging parameter cluster template library. Based on the principle of graph theory isomorphism, map the wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters in the query graph to weighted feature vectors of the graph nodes, and establish the node correspondence between the query graph and the reference graph.

[0123] An adjacency matrix is ​​constructed for the query graph and the reference graph. To address the sparse connectivity characteristics of the rectifier bridge thermal-electric coupling network, a threshold truncation method is used to sparsify the adjacency matrix, removing weak connections with weights below a preset threshold. This generates a sparse adjacency matrix to reduce computational complexity and highlight key topological structures.

[0124] Based on the sparsified adjacency matrix, a subgraph isomorphic matching algorithm is applied to calculate the structural similarity measure between the query graph and each reference graph template. By iteratively searching the maximum common subgraph, the overlap of the current real-time state fingerprint and the historical aging stage template in terms of topological structure is quantified, generating a preliminary structural matching score.

[0125] A node attribute distance penalty term is introduced to calculate the Euclidean distance between the feature vectors of nodes in the query graph and the corresponding feature vectors of nodes in the reference graph, which serves as an attribute similarity index. The structure matching score and the attribute similarity index are weighted and fused to obtain a comprehensive similarity score, reflecting the overall matching degree between the current state and each aging stage.

[0126] Traverse all reference graph templates in the aging state parameter cluster template library, and repeatedly perform the above-mentioned topology alignment, sparsification, structure matching, and attribute fusion operations to generate a similarity score sequence containing the matching degree of all aging stages.

[0127] S4.3: Perform maximum value filtering and confidence threshold verification operations on the similarity score sequence, determine the optimal matching item based on the reference graph template index corresponding to the maximum similarity score, and generate the current rectifier bridge aging state attribution result under the condition of satisfying the preset confidence interval.

[0128] Receive the similarity score sequence generated by S4.2, which contains the structural similarity measure between the query graph and each reference graph template in the template library.

[0129] The similarity score sequence is sorted in descending order, and the score item with the highest structural similarity metric value is extracted as the candidate optimal matching score. The unique index identifier of the corresponding reference graph template is recorded.

[0130] Based on a preset confidence threshold range, the candidate optimal matching score is validated to determine whether the score falls within the high confidence matching domain, thereby eliminating the risk of mismatch due to noise interference.

[0131] If the candidate optimal matching score meets the pre-set confidence interval conditions, then the reference graph template index corresponding to the score is locked, and it is determined as the optimal category of the current rectifier bridge aging state.

[0132] If the candidate optimal matching score does not meet the preset confidence interval conditions, the current matching result is marked as a low confidence state, triggering the subsequent low-frequency auxiliary calibration process instead of direct parameter replacement.

[0133] After confirming a high-confidence match, structured output data is generated, which includes the current aging state attribution result of the rectifier bridge and the corresponding target parameter cluster template index identifier.

[0134] S4.4: Based on the current rectifier bridge aging state attribution determination result, retrieve the metadata index table in the aging state parameter cluster template library and extract the target parameter cluster template index identifier that is uniquely bound to the optimal matching item.

[0135] Receive the current aging state attribution result of the rectifier bridge output by S4.3. This result includes the unique identifier of the optimal matching reference template and the confidence score.

[0136] The unique identifier is parsed, and the aging stage category code and physical degradation path index encoded therein are extracted as the primary key for retrieving the aging state parameter cluster template library.

[0137] Based on the primary key, a hash lookup operation is performed in the preset metadata index table to locate the logical pointer pointing to the storage address of the target parameter cluster.

[0138] The header metadata of the target parameter cluster is read through the logical pointer to verify the data integrity flag and version consistency, ensuring that the retrieved parameters are compatible with the current digital twin model architecture.

[0139] Extract the thermal resistance network topology matrix, carrier recombination coefficient correction term value, and thermal capacity attenuation factor coefficient from the verified target parameter cluster, and encapsulate them into a standardized parameter calling object.

[0140] The parameter call object is bound to the target parameter cluster template index identifier to generate a target parameter cluster template index identifier output package containing complete physical parameter mapping relationships.

[0141] like Figure 3 As shown, step S5: Based on the target parameter cluster template index identifier, retrieve the corresponding target parameter cluster from the aging parameter cluster template library, inject it into the preset digital twin model to replace physical parameters, and generate a calibrated digital twin simulation model. Specifically, this includes:

[0142] S5.1: Based on the target parameter cluster template index identifier, perform an address mapping retrieval operation in the aging parameter cluster template library to locate and read the original data block of the target parameter cluster containing the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor.

[0143] Based on the target parameter cluster template index identifier, a hash mapping retrieval process is performed to quickly locate the physical address of the corresponding data block in the non-volatile storage medium, avoiding the latency overhead caused by a full table scan. The original data stream at the target address is read, and a binary data structure containing the thermal resistance network topology matrix, carrier recombination coefficient correction term, and thermal capacity attenuation factor is parsed. A CRC cyclic redundancy check is performed on the read data block to verify the integrity of the data transmission process, generating a verified target parameter cluster original data block, which is then output to step S5.2.

[0144] S5.2: Perform integrity verification and format standardization on the original data block of the target parameter cluster. Use memory atomic operation instructions to encapsulate the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity attenuation factor into an indivisible atomic parameter update package to ensure data consistency and transaction security during the parameter replacement process.

[0145] The system receives the original data block of the target parameter cluster retrieved in step S5.1. This data block contains the thermal resistance network topology matrix, the carrier recombination coefficient correction term vector, and the thermal capacity attenuation factor scalar. A CRC-32 cyclic redundancy check is performed on the original data block. The checksum of the data block is calculated and compared with a pre-stored standard checksum to verify the integrity of the data transmission process. If the check fails, an abnormal interruption is triggered, and the data is re-requested. The data blocks that pass the check are parsed using format standardization, extracting the data type identifier and dimensional information of each physical parameter. Non-standard floating-point numbers are converted to IEEE 754 double-precision format to ensure that the numerical accuracy meets the requirements of the simulation solver. A sequence of atomic memory operation instructions is constructed, and a mutex mechanism is defined to lock the memory address space of the parameters to be updated in the digital twin model, preventing data contention caused by concurrent access by multiple threads. Zero-copy technology is used to encapsulate the standardized thermal resistance network topology, carrier recombination coefficient correction term, and thermal capacity attenuation factor into an indivisible atomic parameter update package. This update package is represented as a continuous and aligned data structure in memory. By utilizing hardware-supported atomic write instructions, atomic parameter update packages are written to the target memory region in one go, ensuring that all parameters are either updated successfully or remain unchanged, thus achieving transaction-level consistency guarantees.

[0146] S5.3: Obtain the current running handles of the boundary condition module and the conductivity temperature correction module of the heat conduction equation in the digital twin model, perform zero-copy memory mapping injection operation based on the atomized parameter update package, directly overlay the thermal resistance network topology onto the spatial discretization matrix of the boundary condition module of the heat conduction equation, and write the carrier recombination coefficient correction term and the thermal capacity decay factor into the nonlinear iterative coefficient register of the conductivity temperature correction module in real time.

[0147] Receive the atomization parameter update packet output by S5.2, obtain the current running handles of the boundary condition module and the conductivity temperature correction module of the heat conduction equation in the digital twin model, and establish a direct mapping channel with the underlying solver memory space.

[0148] Zero-copy memory mapping injection is performed based on the atomic parameter update package, locking the target memory address segment to prevent concurrent write conflicts.

[0149] The thermal resistance network topology data in the atomized parameter update package is parsed and converted into a sparse matrix format to fit the finite element discretization structure.

[0150] The transformed thermal resistance matrix is ​​directly overlaid onto the spatial discretization coefficient matrix of the boundary condition module of the heat conduction equation, thereby achieving instantaneous updating of the thermal boundary conditions.

[0151] Extract the carrier recombination coefficient correction term from the atomization parameter update package and calculate its nonlinear deviation relative to the standard temperature.

[0152] The carrier recombination coefficient correction term and the thermal capacity decay factor are written into the nonlinear iterative coefficient register of the conductivity temperature correction module in real time to update the electro-thermal coupling constitutive relation.

[0153] S5.4: Monitor the parameter loading status flags of the boundary condition module and the conductivity temperature correction module of the heat conduction equation. When it is confirmed that all atomized parameter update packages have been successfully written and locked without conflict, trigger the state reconstruction signal of the digital twin model, and force refresh the initial condition cache of the solver inside the digital twin model to activate the newly injected thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity attenuation factor.

[0154] Read the memory status registers of the boundary condition module and the conductivity temperature correction module of the heat conduction equation to obtain the status flag data of the current parameter loading process.

[0155] A polling monitoring operation is performed on the status flag data to compare the write completion flag of the atomic parameter update packet with the mutex release signal in real time.

[0156] When it is detected that all data blocks of the target parameter clusters have been successfully written to the specified memory address and there are no concurrent access conflicts, a parameter ready confirmation signal is generated.

[0157] Based on the parameter readiness confirmation signal, a state reconstruction interruption request is sent to the core solver of the digital twin model to pause the iterative calculation of the current simulation step.

[0158] Clear the initial condition buffer inside the solver that stores the temperature field distribution and electric potential energy of the previous moment, and release the occupied computing resources.

[0159] The newly injected thermal resistance network topology matrix is ​​mapped to the finite element mesh nodes, and the spatial discretization coefficients of the heat conduction equation are reinitialized.

[0160] The carrier recombination coefficient correction term and the thermal capacity decay factor are loaded into the nonlinear iterative register of the conductivity temperature correction module to update the material property mapping table.

[0161] The solver's cold start reset command is triggered, forcibly refreshing the internal variable stack so that the new parameters take effect immediately and are ready to enter the next simulation cycle.

[0162] S5.5: Perform a single-step static equilibrium calculation based on the digital twin model with newly injected parameters activated. Verify whether the output responses of the heat conduction equation boundary condition module and the conductivity temperature correction module conform to the physical constraints. If the verification passes, generate a calibrated digital twin simulation model, which will serve as the execution subject for the next cycle's thermal failure trend prediction. This calibrated digital twin simulation model is generated based on the previously constructed basic digital twin model by injecting a target parameter set that matches the current aging state and verifying it through physical constraints. It will serve as the execution subject for the next cycle's thermal failure trend prediction.

[0163] Obtain the digital twin model that has completed state reconstruction in S5.4, and extract the temperature field distribution data and current density distribution data of each node of the rectifier bridge at the current moment as the initial boundary conditions for static equilibrium calculation.

[0164] Perform a single-step static equilibrium calculation to solve the discretized heat conduction equations and obtain the theoretical junction temperature prediction values ​​of each key node under the steady-state assumption.

[0165] A physical constraint verification function is constructed to compare and verify the theoretical junction temperature prediction with the material melting point threshold, the maximum allowable temperature difference gradient, and the residual of the heat flow continuity equation.

[0166] If all verification indicators are within the preset safety confidence interval, the verification is deemed successful, and a calibrated digital twin simulation model is generated.

[0167] Step S6: Activate the calibrated digital twin simulation model to perform thermal failure trend prediction for the next operating cycle, simultaneously calculate fingerprint drift distance values ​​over several consecutive cycles, and generate a real-time fingerprint drift monitoring sequence. Specifically, this includes:

[0168] S6.1: Obtain the current parameter configuration of the boundary condition module of the heat conduction equation and the conductivity temperature correction module in the calibrated digital twin simulation model, and drive the digital twin simulation model to perform transient thermal field simulation calculation for the next running cycle based on the current parameter configuration, and generate thermal failure trend inference results including junction temperature distribution curves of key nodes within the future time window.

[0169] The spatial discretization matrix parameters of the boundary condition module of the heat conduction equation in the calibrated digital twin simulation model, as well as the nonlinear iterative coefficient register data of the conductivity temperature correction module, are obtained as the initial physical constraints for transient thermal field simulation.

[0170] Based on the initial physical constraints, the finite volume method solver is invoked to spatially discretize the three-dimensional geometric mesh of the rectifier bridge, and a global thermal conductivity matrix is ​​constructed, which includes nodal thermal capacity, branch thermal resistance and boundary convective heat transfer coefficient.

[0171] The load current sequence acquired in real time is introduced as a Joule heat source term. The heat generation rate per unit volume of each diode chip and the connecting busbar is calculated according to Joule's law and mapped to the corresponding heat source node of the finite element mesh.

[0172] The implicit Euler method is used to advance the transient heat conduction partial differential equations by time step, and the nodal temperature vectors under each time slice in the next running cycle are solved.

[0173] The nodal temperature evolution is calculated using the following formula:

[0174]

[0175] Where C is the nodal thermal capacity matrix, K is the global thermal conductivity matrix, and T is the total thermal conductivity matrix. t Let T be the temperature vector of the node at the current time. t+1 Let Δt be the node temperature vector at the next moment, Δt be the simulation time step, and Q be the Joule heat source vector.

[0176] The trajectory data of the junction temperature of key power devices over time is extracted from the simulation results. Combined with the preset safety threshold envelope, the thermal failure trend prediction results containing the junction temperature distribution curve of key nodes within the future time window are generated.

[0177] S6.2: Extract the thermal-electric coupling state fingerprint vector generated in the current running cycle as the reference fingerprint vector, and read the thermal-electric coupling state fingerprint vectors corresponding to several consecutive previous cycles from the historical data cache as the reference fingerprint vector set. Construct a multi-dimensional time-series fingerprint comparison matrix based on the reference fingerprint vector and the reference fingerprint vector set.

[0178] Obtain the thermal-electric coupling state fingerprint vector generated in step S2 of the current operating cycle, and mark it as the baseline fingerprint vector. This baseline fingerprint vector contains feature data in three dimensions: wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters.

[0179] From the historical data cache of the edge computing node, read the thermal-electric coupling state fingerprint vectors of several consecutive previous cycles in reverse timestamp order to form a reference fingerprint vector set F. ref ={Ft-1 , F t-2 , ..., F t-N Ensure the integrity of the read data; if missing periods exist, use linear interpolation to fill in the gaps.

[0180] Initialize a multidimensional temporal fingerprint comparison matrix with dimensions (N+1)×3. Fill the first row of the matrix with the three eigenvalues ​​of the baseline fingerprint vector as the reference baseline for the current state, where N is the total number of preceding periods used for comparison.

[0181] Traverse each vector in the reference fingerprint vector set, extract its corresponding wavelet packet energy spectrum, phase consistency and thermal diffusion hysteresis parameters, and fill them into the subsequent rows of the multidimensional time-series fingerprint comparison matrix to form a fingerprint feature sequence that evolves over time.

[0182] Normalization is performed on the data in each column of the multidimensional time-series fingerprint comparison matrix to eliminate numerical differences between different physical dimensions and ensure the weight balance of subsequent distance metric calculations.

[0183] S6.3: Perform Euclidean distance metric operation on the base fingerprint vector and each reference fingerprint vector in the multidimensional time-series fingerprint comparison matrix, calculate the difference value of wavelet packet energy spectrum components, the difference value of phase consistency matrix elements and the difference value of thermal diffusion hysteresis time window parameters respectively, and synthesize the fingerprint drift distance value corresponding to a single period.

[0184] The baseline fingerprint vector for the current period and the reference fingerprint vectors for several consecutive previous periods are extracted from the multidimensional temporal fingerprint comparison matrix and used as the input data source for the Euclidean distance metric.

[0185] The wavelet packet energy spectrum component in the baseline fingerprint vector is normalized to eliminate the influence of dimensional differences on distance calculation and ensure the equivalence of features in each dimension in numerical space.

[0186] The deviation between the baseline vector and each reference vector in the wavelet packet energy spectrum dimension is calculated using the Euclidean distance metric, and the energy spectrum difference value is calculated using the following formula:

[0187]

[0188] Among them, D E E represents the energy spectrum difference value. i The energy of the i-th frequency band of the reference vector. The reference vector represents the energy of the corresponding frequency band, and N is the total number of frequency bands.

[0189] Based on the phase consistency matrix elements, the Frobenius norm distance between the reference matrix and the reference matrix is ​​calculated to quantify the degree of deviation of the thermoelectric signal synchronization characteristics and generate phase consistency difference values.

[0190] Extract the thermal diffusion hysteresis time window parameters, calculate the absolute difference between the baseline parameters and the reference parameters, reflect the linear drift of the thermal resistance change at the heat dissipation interface, and generate the thermal diffusion hysteresis difference value.

[0191] By introducing preset weighting coefficients, the energy spectrum difference value, phase consistency difference value, and thermal diffusion hysteresis difference value are weighted and summed to synthesize the comprehensive fingerprint drift distance value corresponding to a single cycle.

[0192] S6.4: The fingerprint drift distance values ​​calculated over several consecutive periods are serialized and encapsulated in the order of timestamps to generate a real-time fingerprint drift monitoring sequence that characterizes the degree of deviation of the digital twin model parameters over time.

[0193] Step S7: If the fingerprint drift distance value exceeds a preset threshold and no new template matching is triggered, the retrieved target parameter cluster is adjusted, and a final thermal failure risk warning signal is output. Specifically, this includes:

[0194] S7.1: Obtain the comparison result between the fingerprint drift distance value in the real-time fingerprint drift monitoring sequence and the preset threshold. Based on the comparison result, determine that the current digital twin model is in a state of gradual parameter mismatch and generate a weight fine-tuning start command to trigger the edge-embedded Bayesian inference engine.

[0195] Receive the real-time fingerprint drift monitoring sequence output in step S6, extract the fingerprint drift distance value within the current running cycle, and generate the fingerprint drift distance value to be compared.

[0196] Read the preset threshold parameter, perform a numerical comparison operation between the fingerprint drift distance value to be compared and the preset threshold, and generate a Boolean comparison result flag.

[0197] Synchronously retrieve the template matching status identifier for the current period, confirm whether a new aging state parameter cluster switch has been triggered in the current period, and generate the template matching status identifier for the current period.

[0198] When the comparison result flag indicates that the real-time fingerprint drift distance value exceeds the preset threshold, and the template matching status indicator shows that no new template matching has been triggered, the current digital twin model is determined to be in a parameter gradual mismatch state, and a mismatch state determination result is generated.

[0199] Based on the mismatch state determination result, the current timestamp, the real-time fingerprint drift distance value, and the mismatch state code are associated and encapsulated to generate a weight fine-tuning start command data packet, which is then output to step S7.2.

[0200] S7.2: Receive the weight fine-tuning start command, call the edge embedded Bayesian inference engine, use the thermal-electric coupling state fingerprint vector of the current period as observation evidence, perform posterior probability distribution update calculation on the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity attenuation factor in the target parameter cluster, and generate parameter weight correction amount containing confidence interval.

[0201] The system receives the weight fine-tuning start command output by S7.1, parses the current period thermal-electric coupling state fingerprint vector contained therein, and generates the observation evidence set for Bayesian inference.

[0202] The edge-embedded Bayesian inference engine is invoked to load the prior probability distribution models of the thermal resistance network topology, carrier recombination coefficient correction term, and thermal capacity attenuation factor in the target parameter cluster, thereby generating the prior probability distribution model of the target parameter cluster. The edge-embedded Bayesian inference engine is trained offline based on laboratory aging test data and is stored in the edge storage medium as lightweight parameters.

[0203] Using the observation evidence set of the Bayesian inference and the prior probability distribution model of the target parameter cluster as input, a likelihood function is constructed to calculate the probability density of each physical parameter taking a specific value under the current observed fingerprint vector conditions, thereby generating a likelihood function model representing the mapping relationship between the parameters and fingerprint features.

[0204] The posterior probability distribution is calculated using the following formula:

[0205]

[0206] Where θ is the physical parameter vector to be calibrated, D is the three-dimensional state fingerprint observation data of the current period, P(θ) is the prior probability, P(D|θ) is the likelihood function, and P(D) is the normalization constant.

[0207] Monte Carlo sampling is performed on the posterior probability distribution to extract the mean and variance of the parameter distribution, determine the optimal estimate and uncertainty range of each physical parameter, and generate the mean and variance estimation results of the parameter distribution.

[0208] Based on the mean and variance estimation results of the parameter distribution, the deviation between the optimal estimate and the prior mean of each physical parameter is calculated as the correction amount for each physical parameter. The correction amount and its corresponding confidence interval are associated and encapsulated to generate a parameter correction amount containing the confidence interval, and output to step S7.3.

[0209] S7.3: Based on the parameter weight correction amount, the target parameter cluster retrieved from the aging state parameter cluster template library is atomically weighted and iteratively updated using weighted fusion processing to eliminate model boundary condition deviations caused by contact resistance drift or heat dissipation interface degradation, and generate a calibrated dynamic parameter cluster.

[0210] The system receives the parameter weight correction amount generated in step S7.2, which includes the thermal resistance network topology adjustment coefficient, carrier recombination coefficient offset, and thermal capacity attenuation factor compensation term. It then parses the original data of the target parameter cluster retrieved from the aging parameter cluster template library, extracting the baseline thermal resistance matrix, baseline recombination coefficient vector, and baseline thermal capacity scalar as the objects to be updated.

[0211] A weighted fusion processing model is constructed, with forgetting factors and confidence weights set. The baseline parameters are then updated using linear interpolation based on corrections calculated from the Bayesian posterior probability distribution. For the thermal resistance network topology, the updated thermal resistance value is calculated using the following formula:

[0212]

[0213] Among them, R n,t Let R be the thermal resistance update value of the nth node at time t, α be the forgetting factor, and R be the value of the node at time t. n,t-1 R is the thermal resistance value at the previous moment. p,t The predicted thermal resistance value is derived from Bayesian inference.

[0214] For the carrier recombination coefficient, vector-level weighted fusion is performed, superimposing the offset value in the correction amount onto the reference vector according to a preset ratio to eliminate the conductivity calculation deviation caused by contact resistance drift. For the thermal capacity attenuation factor, exponential smoothing filtering is used to process the correction amount, suppressing high-frequency noise interference and generating a stable thermal capacity compensation coefficient.

[0215] The updated thermal resistance, carrier recombination coefficient vector, and thermal capacity compensation coefficient are encapsulated into an atomic data structure, a version identifier and checksum are added, and a calibrated dynamic parameter cluster is generated.

[0216] S7.4: Inject the calibrated dynamic parameter cluster into the boundary condition module of the heat conduction equation and the conductivity temperature correction module of the digital twin simulation model, drive the calibrated digital twin simulation model to perform thermal failure trend deduction for the next running cycle, and generate thermal failure trend deduction results containing the junction temperature prediction curve for the future time window.

[0217] The system receives the calibrated dynamic parameter cluster generated by S7.3. This data cluster includes the thermal resistance network topology corrected by Bayesian inference, the carrier recombination coefficient correction term, and the thermal capacity attenuation factor, which serve as the input data source for parameter injection.

[0218] The atomized data structure in the calibrated dynamic parameter cluster is analyzed, and the topology parameters of the thermal resistance network are extracted to reconstruct the boundary condition discretization matrix of the heat conduction equation in the digital twin model, thereby generating the reconstructed boundary condition discretization matrix of the heat conduction equation.

[0219] The carrier recombination coefficient correction term and the thermal capacity decay factor are mapped to the nonlinear iterative coefficient register of the conductivity temperature correction module, overwriting the original static parameter values, and generating the updated conductivity temperature correction module parameter configuration.

[0220] Execute the memory atomic operation instructions to complete the lock-free writing of the reconstructed heat conduction equation boundary condition discretization matrix and the updated conductivity temperature correction module parameters within a microsecond time window, generating a digital twin simulation model instance with injected parameters.

[0221] The state reconstruction signal of the digital twin simulation engine is triggered, which forcibly refreshes the initial condition cache of the solver inside the digital twin simulation model instance with injected parameters, activates the newly injected thermal-electric coupling parameter configuration, and generates a calibrated digital twin simulation model instance with the new parameters activated.

[0222] Based on the calibrated digital twin simulation model instance with the newly activated parameters, the load current prediction sequence and ambient temperature boundary conditions for the next running cycle are loaded to generate a simulation solver instance loaded with the current physical boundary conditions.

[0223] The implicit finite difference method is used to numerically solve the transient heat conduction partial differential equation in the simulation solver instance with the current physical boundary conditions, and the junction temperature evolution trajectory of the key nodes of the rectifier bridge within a future time window is calculated:

[0224]

[0225] Where T(t) is the junction temperature at the current moment, C(T) is the heat capacity as a function of temperature, P(I,V) is the instantaneous power loss calculated based on the corrected conductivity, R(T) is the updated equivalent thermal resistance, and T amb The ambient temperature.

[0226] Iteratively solve the above equations until the preset time step endpoint is reached, generate a sequence of junction temperature prediction curves containing several future sampling points, which serve as the result of thermal failure trend inference, and output them to step S7.5.

[0227] S7.5: If the weight fine-tuning start command is triggered, the thermal failure trend projection results generated in S7.4 are obtained and analyzed. If the weight fine-tuning start command is not triggered, the projection results temporarily stored during the execution of S6 are obtained and analyzed. The overlap between the junction temperature prediction curve and the safety threshold envelope in the thermal failure trend projection results is analyzed, the thermal failure risk probability index is calculated, and combined with the current rectifier bridge aging state attribution result, a final thermal failure risk warning signal containing the failure prediction result and confidence level is generated.

[0228] Obtain the junction temperature prediction curve data and safety threshold envelope parameters from the thermal failure trend projection results as the input basis for risk quantification analysis. Extract the discrete sampling point sequence of the junction temperature prediction curve within a preset future time window, and simultaneously read the upper and lower boundary values ​​of the safety threshold at the corresponding time points to construct a time-series temperature-threshold comparison dataset.

[0229] The cumulative deviation of the junction temperature prediction curve from the safety threshold envelope is calculated using the integral overlap area method, and the thermal failure risk probability index is calculated using the following formula:

[0230]

[0231] Where P is the probability index of thermal failure risk, and T is the length of the prediction time window. max T is the maximum allowable junction temperature of the rectifier bridge, and T(t) is the predicted junction temperature at time t. limit This is a dynamic safety threshold. The formula quantifies the combined impact of overheating amplitude and duration, transforming physical temperature deviations into standardized risk probability values.

[0232] Based on the current aging status determination results of the rectifier bridge, a pre-set risk confidence mapping table is consulted to obtain the model prediction confidence coefficient for the corresponding aging stage. Using a Bayesian correction algorithm, the thermal failure risk probability index and the model prediction confidence coefficient are weighted and fused to calculate the final thermal failure risk level.

[0233] Based on the calculated thermal failure risk level, a pre-defined early warning strategy library is matched to generate a final thermal failure risk early warning signal that includes the failure prediction result, the risk level value, and the confidence level. If the risk level exceeds the emergency threshold, it is marked as a Level 1 warning and a shutdown protection recommendation is triggered; if it is within the attention range, it is marked as a Level 2 warning and a maintenance window is recommended.

[0234] In one embodiment, this application also provides a generator rectifier bridge thermal failure prediction system based on digital twins, including: a raw thermal-electric coupling sampling dataset acquisition module: used to collect the transient junction temperature distribution sequence and diode forward voltage drop waveform of key nodes of generator rectifier bridge under typical operating conditions, and generate raw thermal-electric coupling sampling dataset through associated packaging;

[0235] Thermo-electric coupling state fingerprint vector construction module: used to perform time-frequency joint coding processing on the original thermo-electric coupling sampling dataset, extract multi-dimensional time-frequency domain feature parameters, and construct thermo-electric coupling state fingerprint vector;

[0236] Aging state parameter cluster template library pre-set module: used to pre-set an aging state parameter cluster template library based on the failure paths of solder joint cold solder joints, contact oxidation and thermal stress cracking covered by laboratory accelerated aging tests, using the thermal-electric coupling state fingerprint vector as the index basis.

[0237] The target parameter cluster template index identifier acquisition module is used to perform fingerprint similarity retrieval and template index positioning operations based on the thermal-electric coupling state fingerprint vector, and output the current rectifier bridge aging state attribution result and the corresponding target parameter cluster template index identifier.

[0238] The calibration-post digital twin simulation model generation module is used to retrieve the corresponding target parameter cluster from the aging state parameter cluster template library according to the target parameter cluster template index identifier, inject it into the preset digital twin model to replace the physical parameters and generate the calibration-post digital twin simulation model.

[0239] Real-time fingerprint drift monitoring sequence generation module: used to enable the calibrated digital twin simulation model to perform thermal failure trend inference for the next running cycle, synchronously calculate the fingerprint drift distance value within several consecutive cycles, and generate a real-time fingerprint drift monitoring sequence;

[0240] Final thermal failure risk warning signal acquisition module: If the fingerprint drift distance value exceeds a preset threshold and no new template matching is triggered, the module adjusts the retrieved target parameter cluster and outputs the final thermal failure risk warning signal.

[0241] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0242] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0243] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting thermal failure of a generator rectifier bridge based on digital twins, characterized in that, include: S1: Collect the transient junction temperature distribution sequence and diode forward voltage drop waveform of key nodes of generator rectifier bridge under typical operating conditions, and generate the original thermal-electric coupling sampling dataset by correlation packaging; S2: Perform time-frequency joint coding on the original thermal-electric coupling sampling dataset, extract multi-dimensional time-frequency domain feature parameters, and construct a thermal-electric coupling state fingerprint vector; S3: Based on the failure paths of solder joint cold solder joints, contact oxidation, and thermal stress cracking covered by laboratory accelerated aging tests, the thermal-electric coupling state fingerprint vector is used as an index to pre-set an aging state parameter cluster template library. S4: Based on the thermal-electric coupling state fingerprint vector, perform fingerprint similarity retrieval and template index positioning operations, and output the current rectifier bridge aging state attribution result and the corresponding target parameter cluster template index identifier; S5: Retrieve the corresponding target parameter cluster from the aging parameter cluster template library according to the target parameter cluster template index identifier, inject it into the preset digital twin model to replace the physical parameters and generate a calibrated digital twin simulation model; S6: Enable the calibrated digital twin simulation model to perform thermal failure trend deduction for the next operating cycle, synchronously calculate the fingerprint drift distance value within several consecutive cycles, and generate a real-time fingerprint drift monitoring sequence. S7: If the fingerprint drift distance value exceeds the preset threshold and no new template matching is triggered, the retrieved target parameter cluster is adjusted, and the final thermal failure risk warning signal is output.

2. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 1, characterized in that, The aging parameter cluster template library includes thermal resistance network topology, carrier recombination coefficient correction term, and thermal capacity decay factor.

3. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 2, characterized in that, Step S3 specifically includes: Obtain the original degradation sample set covering the failure paths of solder joint cold welding, contact oxidation and thermal stress cracking in the laboratory accelerated aging test, perform multi-physics field coupling mechanism analysis on the original degradation sample set, extract the thermal resistance evolution law, carrier recombination coefficient correction term change law and thermal capacity decay factor change curve under different aging stages, and generate aging stage characteristic dataset. Based on the aging stage feature dataset, the parameters of each aging interval are solved in reverse, and the thermal resistance network parameter values, carrier recombination coefficient correction terms and thermal capacity decay factor coefficients corresponding to the thermal resistance network topology evolution law are calculated to generate a set of candidate parameter clusters. The thermal-electric coupling state fingerprint vector is associated with the candidate parameter cluster set to establish a correspondence between the fingerprint feature space and the parameter physical space, and a fingerprint parameter mapping index table is generated. Based on the fingerprint parameter mapping index table, the candidate parameter cluster set is standardized and encapsulated, and the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor are integrated into atomic data units to generate an aging state parameter cluster template. Receive all generated aging state parameter cluster templates, write them into a non-volatile storage medium for structured storage, and construct the aging state parameter cluster template library.

4. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 1, characterized in that, The multi-dimensional time-frequency domain feature parameters include: wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters.

5. The generator rectifier bridge thermal failure prediction method based on digital twin according to claim 4, characterized in that, Step S4 specifically includes: The real-time generated thermal-electric coupling state fingerprint vector is used as input data. Based on the predefined graph node mapping rules, the wavelet packet energy spectrum component, phase consistency matrix element and thermal diffusion lag time window parameter in the thermal-electric coupling state fingerprint vector are mapped to query graph node attributes respectively to generate a query graph structure. The query graph structure is used to perform topological alignment with the set of reference graph templates pre-stored in the aging parameter cluster template library. The structural similarity metric between the query graph and each reference graph template is calculated to generate a similarity score sequence. The maximum value filtering and confidence threshold verification operations are performed on the similarity score sequence. The optimal matching item is determined based on the reference graph template index corresponding to the maximum similarity score, and the current rectifier bridge aging state attribution result is generated under the condition of satisfying the preset confidence interval. Based on the current rectifier bridge aging state attribution determination result, the metadata index table in the aging state parameter cluster template library is retrieved, and the target parameter cluster template index identifier uniquely bound to the optimal matching item is extracted.

6. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 5, characterized in that, The real-time generated thermo-electric coupling state fingerprint vector is used as input data. Based on predefined graph node mapping rules, the wavelet packet energy spectrum components, phase consistency matrix elements, and thermal diffusion hysteresis time window parameters in the thermo-electric coupling state fingerprint vector are mapped to query graph node attributes, generating a query graph structure, including: Based on predefined graph node mapping rules, the three-dimensional features of the thermal-electric coupling state fingerprint vector are instantiated as node attributes in the query graph structure, generating an initialized query graph node set. Establish directed edge connections between nodes in the query graph node set, set edge weights according to the physical causality law, point energy distribution nodes to phase synchronization nodes, and then point phase synchronization nodes to thermal hysteresis nodes to generate the initial query graph topology. The initial query graph topology is normalized to generate the normalized query graph structure.

7. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 6, characterized in that, The first type of nodes in the query graph node set are mapped to the high-frequency and low-frequency energy ratios of the wavelet packet energy spectrum, and are given weight attributes to represent the distribution of local hotspots; the second type of nodes are mapped to the main diagonal elements and off-diagonal correlation coefficients of the phase consistency matrix, and are given connection weights to represent the synchronicity of the thermal-electric response; the third type of nodes are mapped to the start time and duration of the thermal diffusion lag time window, and are given temporal weights to represent the change in thermal resistance of the heat dissipation interface.

8. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 2, characterized in that, Step S5 specifically includes: Based on the target parameter cluster template index identifier, an address mapping retrieval operation is performed in the aging parameter cluster template library to locate and read the original data block of the target parameter cluster containing the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor. The original data blocks of the target parameter cluster are subjected to integrity verification and format standardization processing, and the thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor are encapsulated into an indivisible atomized parameter update package. Obtain the current running handles of the boundary condition module and the conductivity temperature correction module of the heat conduction equation in the digital twin model, perform a zero-copy memory mapping injection operation based on the atomized parameter update package, directly overlay the thermal resistance network topology onto the spatial discretization matrix of the boundary condition module of the heat conduction equation, and write the carrier recombination coefficient correction term and the thermal capacity decay factor into the nonlinear iterative coefficient register of the conductivity temperature correction module in real time. Monitor the parameter loading status flags of the boundary condition module and the conductivity temperature correction module of the heat conduction equation. When it is confirmed that all atomized parameter update packages have been successfully written and locked without conflict, trigger the state reconstruction signal of the digital twin model, and force refresh the initial condition cache of the solver inside the digital twin model to activate the newly injected thermal resistance network topology, carrier recombination coefficient correction term and thermal capacity decay factor. A single-step static equilibrium calculation is performed based on the digital twin model with the newly injected parameters activated to verify whether the output responses of the heat conduction equation boundary condition module and the conductivity temperature correction module meet the physical constraints. If the verification is successful, the calibrated digital twin simulation model is generated.

9. The method for predicting thermal failure of a generator rectifier bridge based on digital twins according to claim 1, characterized in that, The typical operating conditions include at least one of no-load, half-load, and full-load.

10. A generator rectifier bridge thermal failure prediction system based on digital twins, characterized in that, include: Original thermal-electric coupling sampling dataset acquisition module: used to collect the transient junction temperature distribution sequence and diode forward voltage drop waveform of key nodes of generator rectifier bridge under typical operating conditions, and generate the original thermal-electric coupling sampling dataset through correlation packaging; Thermo-electric coupling state fingerprint vector construction module: used to perform time-frequency joint coding processing on the original thermo-electric coupling sampling dataset, extract multi-dimensional time-frequency domain feature parameters, and construct thermo-electric coupling state fingerprint vector; Pre-set module for aging state parameter cluster template library: Based on the failure paths of solder joint cold solder joints, contact oxidation and thermal stress cracking covered by laboratory accelerated aging tests, the module uses the thermal-electric coupling state fingerprint vector as an index to pre-set the aging state parameter cluster template library. The target parameter cluster template index identifier acquisition module is used to perform fingerprint similarity retrieval and template index positioning operations based on the thermal-electric coupling state fingerprint vector, and output the current rectifier bridge aging state attribution result and the corresponding target parameter cluster template index identifier. The calibration-post digital twin simulation model generation module is used to retrieve the corresponding target parameter cluster from the aging state parameter cluster template library according to the target parameter cluster template index identifier, inject it into the preset digital twin model to replace the physical parameters and generate the calibration-post digital twin simulation model. Real-time fingerprint drift monitoring sequence generation module: used to enable the calibrated digital twin simulation model to perform thermal failure trend inference for the next running cycle, synchronously calculate the fingerprint drift distance value within several consecutive cycles, and generate a real-time fingerprint drift monitoring sequence; Final thermal failure risk warning signal acquisition module: If the fingerprint drift distance value exceeds a preset threshold and no new template matching is triggered, the module adjusts the retrieved target parameter cluster and outputs the final thermal failure risk warning signal.