A new energy vehicle electric control system fault prediction system
By using multi-dimensional feature analysis and health quantification modeling, the problem of early and accurate warning in the fault diagnosis of PFC inductor in on-board chargers has been solved, realizing the transformation from post-fault alarm to pre-fault prediction, improving the accuracy and reliability of prediction, and enhancing the safety and operation and maintenance efficiency of the charging system.
Patent Information
- Application Number
- CN202611220825.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-12
- Publication Date
- 2026-11-10
AI Technical Summary
Existing on-board charger PFC inductor fault diagnosis methods lack multi-dimensional feature collaborative analysis and cross-validation when facing complex operating conditions and multivariate coupling. This makes it difficult to achieve early and accurate fault isolation and early warning, and may lead to misjudgment as other system problems, resulting in maintenance delays and potential thermal runaway risks.
By acquiring multi-source real-time monitoring datasets, performing time-frequency domain analysis and feature extraction, generating a multi-dimensional feature spectrum set for PFC inductors, conducting spatiotemporal correlation analysis and fault mode matching among multi-dimensional features, calculating the confidence level characterizing the performance degradation of PFC inductors, generating quantitative health modeling and degradation trend prediction, and outputting a preventive maintenance report.
It enables precise detection of early performance degradation of PFC inductors, improves the accuracy and reliability of prediction, avoids false alarms, and enhances the safety of charging systems and the efficiency of vehicle operation and maintenance throughout the entire life cycle.
Smart Images

Figure CN123066444A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of electric control technology for new energy vehicles, and in particular to a fault prediction system for the electric control system of new energy vehicles. Background Technology
[0002] In the field of intelligent charging and safety maintenance of electric vehicles, the on-board charger is a core component in the electronic control system that ensures efficient and reliable energy replenishment of the power battery. The working state of its power factor correction circuit directly affects charging efficiency, grid quality and high-voltage safety of the whole vehicle, and is a key link to improve the user charging experience and extend the service life of the vehicle.
[0003] However, existing fault diagnosis methods for PFC inductors in on-board chargers mostly rely on threshold judgment of single characteristic parameters. When faced with complex operating conditions, multivariate coupling, and early progressive performance degradation, they lack a mechanism for collaborative analysis and cross-validation of multi-dimensional characteristics. This not only makes it difficult to achieve early and accurate fault isolation and early warning, but may also lead to misjudgment of other system problems during the fault incubation period, resulting in maintenance delays and potential thermal runaway risks. Summary of the Invention
[0004] This application provides a fault prediction system for the electronic control system of new energy vehicles to solve the above-mentioned technical problems. The system includes: A multi-source real-time monitoring dataset is acquired. Based on this dataset, time-frequency domain analysis and feature extraction are performed to separate and quantify the current waveform distortion, efficiency-temperature correlation anomalies, and anomaly synchronization features dominated by PFC inductor performance degradation, generating a multi-dimensional feature spectrum set for PFC inductors. Based on this multi-dimensional feature spectrum set, spatiotemporal correlation analysis and fault mode matching are performed between the multi-dimensional features. Through cross-validation and fault mode matching, the confidence level characterizing PFC inductor performance degradation is calculated, generating a PFC inductor fault isolation and confidence information set. Based on this PFC inductor fault isolation and confidence information set, quantitative health modeling and degradation trend prediction are performed, and a vehicle charger inductor health status prediction report is generated and output to guide preventive maintenance.
[0005] The above technical solutions can accurately capture and quantify the early performance degradation of PFC inductors, realizing the transformation from "fault alarm" to "fault prediction". Through multi-dimensional cross-validation and quantitative confidence assessment, the accuracy and reliability of prediction are significantly improved, and false alarms are effectively avoided. The final prediction report can directly guide preventive maintenance, thereby improving the safety, reliability and operation and maintenance efficiency of the charging system and the entire life cycle of the vehicle.
[0006] Optionally, the generation process of the PFC inductor multi-dimensional feature spectrum set includes: the multi-source real-time monitoring dataset includes the AC input current waveform under charging conditions, the temperature time-series data of the PFC inductor, and real-time charging efficiency data; high-frequency sampling and spectrum analysis are performed on the AC input current waveform to extract its total harmonic distortion rate and the proportion of specific harmonic amplitudes as current waveform distortion features; the dynamic correspondence between the temperature change of the PFC inductor and the real-time charging efficiency is analyzed, and the abnormal decay rate of efficiency with increasing temperature is analyzed as efficiency-temperature correlation anomaly features; the current waveform distortion features and efficiency-temperature correlation anomaly features are calibrated to strictly appear within the time window from charging start to end as abnormal synchronization features; the current waveform distortion features, the efficiency-temperature correlation anomaly features, and the abnormal synchronization features are integrated to generate the PFC inductor multi-dimensional feature spectrum set.
[0007] Optionally, the analysis process of the current waveform distortion characteristics includes: performing a fast Fourier transform on the AC input current waveform to obtain the current spectrum of the AC input current waveform; analyzing the real-time total harmonic distortion (THD) rate of the AC input current waveform by comparing the current spectrum with a standard sinusoidal current spectrum; simultaneously, extracting the amplitude sequence of specific harmonic components related to the switching frequency of the PFC circuit in the current spectrum, and using a linear regression method to analyze the growth slope and correlation coefficient of this amplitude over time to quantify its regularity enhancement trend; if the THD rate continues to increase and the regularity of the specific harmonic components enhances, then the THD rate and the specific harmonic components are used as the current waveform distortion characteristics characterizing the performance degradation of the PFC inductor core.
[0008] Optionally, the analysis process of the efficiency-temperature correlation anomaly includes: establishing a historical baseline correspondence map between the temperature of the PFC inductor and the real-time charging efficiency within the same or similar charging power range using a Gaussian process regression method; analyzing the deviation of the current charging efficiency from the expected efficiency in the historical baseline correspondence map under the same temperature rise conditions during the real-time acquisition of the multi-source real-time monitoring dataset; simultaneously analyzing the ratio of efficiency decrease to temperature increase within a unit time window, defining it as the efficiency-temperature sensitivity coefficient; if the deviation continues to increase and the absolute value of the efficiency-temperature sensitivity coefficient increases abnormally, then the deviation and the efficiency-temperature sensitivity coefficient are taken as the efficiency-temperature correlation anomaly.
[0009] Optionally, the analysis process of the abnormal synchronicity feature includes: attaching precise timestamps to the occurrence process of the current waveform distortion feature and the efficiency-temperature correlation anomaly feature; analyzing the entire timeline of the two from their appearance, change to disappearance, and strictly comparing it with the start-stop control signal of the charger; determining that the current waveform distortion feature and the efficiency-temperature correlation anomaly feature have strict abnormal synchronicity if and only if the active timelines of the current waveform distortion feature and the efficiency-temperature correlation anomaly feature are completely wrapped within the effective working time interval of the charger, and the vehicle is in a stationary condition with the charging engine running.
[0010] Optionally, the process of generating the PFC inductor fault isolation and confidence information set includes: constructing a three-dimensional correlation framework of time, space, and logic based on the multi-dimensional feature spectrum set of the PFC inductor; cross-validating the current waveform distortion feature, the efficiency-temperature correlation anomaly feature, and the anomaly synchronization feature based on the logic in the three-dimensional correlation framework to generate a multi-dimensional feature correlation strength index characterizing the causal linkage consistency between features; calculating the comprehensive confidence score of the motor characterizing the performance degradation of the PFC inductor in different periods based on the multi-dimensional feature correlation strength index obtained at different times; and integrating all the multi-dimensional feature correlation strength indices to generate the PFC inductor fault isolation and confidence information set.
[0011] Optionally, the construction process of the three-dimensional correlation framework of time, space, and logic includes: using the start-stop signal of the charger as a reference, performing time alignment on all feature data in the PFC inductor multi-dimensional feature spectrum set to construct a time-related dataset; identifying and locating a set of monitoring points at different physical locations within the on-board charger, mapping and grouping the feature data involving temperature and electrical response in the PFC inductor multi-dimensional feature spectrum set according to their source with the set of monitoring points to form a spatial correlation dataset for comparing local and global state differences; and defining the theoretical causal relationship link between the current waveform distortion feature, the efficiency-temperature correlation anomaly feature, and the temperature feature based on the basic principles of electricity and thermodynamics, thereby serving as the three-dimensional correlation framework for verifying whether the data linkage relationship in the multi-dimensional feature spectrum set conforms to physical laws.
[0012] Optionally, the logic in the three-dimensional correlation framework of time, space, and logic includes: Time correlation verification: verifying that the aggravation of the current waveform distortion characteristics begins in the early stage of the charging process and precedes the appearance of the efficiency-temperature correlation anomaly characteristics. At the same time, analyzing whether the intensity of the current waveform distortion characteristics and the temperature rise rate of the PFC inductor monitoring point show a synchronous aggravation trend throughout the charging process; Spatial correlation verification: comparing and analyzing the temperature rise rate of the PFC inductor with the temperature rise rate of other major heat sources in the on-board charger. If only the PFC inductor shows an independent abnormal temperature rise, while the temperature rise of other heat sources is normal, then it is confirmed that the thermal anomaly has spatial locality; Logical correlation verification: based on a preset fault causal chain model, analyzing whether the data linkage relationship of each feature in the multi-dimensional feature spectrum set of the PFC inductor is consistent with the theoretical derivation of this model. The theoretical transmission logic of the fault causal chain model is that the inductor performance degradation leads to an increase in losses and harmonics, causing local temperature rise anomalies, and ultimately resulting in a decrease in system efficiency.
[0013] Optionally, the cross-validation and motor comprehensive confidence score calculation includes: calculating the dynamic correlation between the intensity of the current waveform distortion feature and the temperature rise rate of the PFC inductor monitoring point, and the statistical correlation coefficient between the total harmonic distortion rate of the current and the temperature rise rate, based on the results of the time correlation validation and the logical correlation validation; using the weighted sum of the dynamic correlation and the statistical correlation coefficient as the multidimensional feature correlation strength index; inputting the multidimensional feature correlation strength index into a preset performance degradation scoring model, which, based on the thermal anomaly locality level confirmed by the spatial correlation validation and the multidimensional feature correlation strength index, calculates the motor comprehensive confidence score characterizing the degree of performance degradation of the PFC inductor at different times using a piecewise linear mapping algorithm.
[0014] Optionally, the generation process of the on-board charger inductor health status prediction report includes: constructing a comprehensive health index model based on historical and current PFC inductor fault isolation and confidence index sets and corresponding PFC inductor multi-dimensional feature spectrum set data; fitting the change trajectory of the comprehensive health index model using time series analysis methods to predict the future decline trend of the comprehensive health index; estimating the remaining reliable service life or the time window for the next recommended maintenance of the on-board charger PFC inductor based on the predicted decline trend curve of the comprehensive health index and a preset safety threshold line; and integrating the current health status, decline trend prediction, and maintenance time window to generate the on-board charger inductor health status prediction report in a structured manner. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application; Figure 2 A flowchart of a fault prediction system for the electronic control system of a new energy vehicle provided in an embodiment of this application. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0018] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.
[0019] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.
[0020] Existing fault diagnosis methods for PFC inductors in on-board chargers mostly rely on threshold judgment of single characteristic parameters. When faced with complex operating conditions, multivariate coupling, and early progressive performance degradation, they lack a mechanism for collaborative analysis and cross-validation of multi-dimensional characteristics. This not only makes it difficult to achieve early and accurate fault isolation and early warning, but may also lead to misjudgment of other system problems during the fault incubation period, resulting in maintenance delays and potential thermal runaway risks.
[0021] Based on this, this application provides a fault prediction system for the electronic control system of new energy vehicles. First, time-frequency analysis is performed on the current waveform to extract harmonic distortion features. Simultaneously, the relationship between inductor temperature rise and abnormal efficiency degradation is analyzed, confirming that the two occur strictly synchronously only during charging, thus generating a multi-dimensional feature spectrum set characterizing PFC inductor degradation. Second, spatiotemporal correlation analysis and fault mode matching are performed on these features to verify their causal logic consistency and calculate a quantified fault confidence level. Finally, a health degradation model is constructed based on the confidence level sequence to predict future degradation trends, automatically generating a prediction report containing remaining service life or maintenance recommendations, and outputting it to the user. This solution can accurately capture and quantify the early performance degradation of PFC inductors, realizing a shift from "fault alarm" to "fault prediction." Through multi-dimensional cross-validation and quantified confidence level assessment, the accuracy and reliability of prediction are significantly improved, effectively avoiding false alarms. The final prediction report can directly guide preventative maintenance, thereby improving the safety, reliability, and overall vehicle lifecycle maintenance efficiency of the charging system.
[0022] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. In the process of predicting faults in automotive electronic control systems, the system provided in this application can improve the safety, reliability, and overall maintenance efficiency of the charging system throughout the vehicle's lifecycle.
[0023] Specifically, the system of this application is applied to any server that communicates with the vehicle-mounted sensors. It acquires multi-source real-time monitoring datasets provided by the vehicle-mounted sensors through the server. First, it performs time-frequency analysis on the current waveform to extract harmonic distortion features. Simultaneously, it analyzes the correlation between inductor temperature rise and abnormal efficiency degradation, confirming that the two only occur strictly synchronously during charging, thereby generating a multi-dimensional feature spectrum set characterizing PFC inductor degradation. Second, it performs spatiotemporal correlation analysis and fault mode matching on these features to verify their causal logic consistency and calculates a quantified fault confidence level. Finally, it constructs a health degradation model based on the confidence level sequence to predict future degradation trends, automatically generates a prediction report containing remaining service life or maintenance recommendations, and outputs it to the user.
[0024] For specific implementation details, please refer to the following examples.
[0025] Figure 2 This is a flowchart illustrating a fault prediction system for the electronic control system of a new energy vehicle, provided as an embodiment of this application. The system in this embodiment can be applied to servers in the above scenarios. For example... Figure 2 As shown, the system includes: S201. Obtain a multi-source real-time monitoring dataset. Based on the multi-source real-time monitoring dataset, perform time-frequency domain analysis and feature extraction to separate and quantify the current waveform distortion, efficiency-temperature correlation anomaly, and anomaly synchronization characteristics dominated by the performance degradation of PFC inductors, and generate a multi-dimensional feature spectrum set of PFC inductors.
[0026] The multi-source real-time monitoring dataset can be a collection of data collected in real time by on-board sensors during vehicle charging, specifically including but not limited to AC input current waveform, temperature of the PFC inductor body, output power and input voltage of the on-board charger, and current. Current waveform distortion can be a quantitative characterization of abnormal current harmonic components caused by PFC inductor core saturation or increased losses. Efficiency-temperature correlation anomaly can be a quantitative characterization of the correlation between abnormal system efficiency decline and accelerated local temperature rise caused by increased PFC inductor losses. Abnormal synchronicity characteristics can refer to the pattern characteristic that the above-mentioned current distortion and efficiency-temperature anomaly occur strictly synchronously only during the effective operation period of the on-board charger. The PFC inductor multi-dimensional feature spectrum set can be a feature data set that integrates the three types of quantitative features—current waveform distortion, efficiency-temperature correlation anomaly, and abnormal synchronicity characteristics—to comprehensively characterize the potential fault modes of PFC inductors.
[0027] Specifically, in the power factor correction (PFC) circuit of on-board chargers for new energy vehicles, inductors are key passive components, and their performance degradation (such as core aging and increased winding losses) is a significant potential factor leading to system failure. However, traditional fault diagnosis or prediction in electronic control systems typically focuses on active switching devices (such as MOSFETs and IGBTs) or overall system efficiency, and is significantly inadequate for early and accurate prediction of the performance degradation of passive components like PFC inductors. Existing methods either rely on single-parameter threshold alarms, resulting in low sensitivity and accuracy, and are prone to false alarms or missed alarms; or they require disassembly for offline testing, making it impossible to achieve on-board real-time prediction. This paper addresses this core issue by proposing to actively separate and quantify the unique fault characteristics "dominated" by PFC inductor performance degradation from multi-source real-time monitoring data, namely, current waveform distortion, abnormal efficiency-temperature correlation, and their abnormal temporal synchronicity. The necessity lies in the fact that PFC inductor degradation is a gradual multi-physical process (electromagnetic, thermal, and efficiency), which cannot be reliably characterized by a single feature; feature extraction and fusion from multiple physical dimensions (electrical, thermal, and temporal) are essential. Secondly, it is essential to "separate" the characteristics of inductor degradation from the complex noise and cross-effects of the system. For example, it is necessary to distinguish whether the current distortion is caused by inductor failure or switching device failure. This is a prerequisite for accurate "fault isolation" prediction. Finally, generating a structured "multi-dimensional feature spectrum set" is the only and necessary data foundation for subsequent in-depth correlation analysis and quantitative modeling.
[0028] S202. Based on the multi-dimensional feature spectrum set of PFC inductor, perform spatiotemporal correlation analysis and fault mode matching among multi-dimensional features. Through cross-validation and fault mode matching, calculate the confidence level characterizing the performance degradation of PFC inductor and generate a set of PFC inductor fault isolation and confidence index.
[0029] Spatiotemporal correlation analysis can be a comprehensive analysis of the temporal evolution sequence and coordinated change relationship of different features in the multi-dimensional feature spectrum set, as well as the distribution correlation of feature signal sources in physical space (such as the positions of different components within an on-board charger). Fault mode matching can be a comparison between the feature correlation patterns obtained from the current analysis and a preset physical model or historical fault mode library that characterizes typical degradation paths of PFC inductors. The PFC inductor fault isolation and confidence index set can be a comprehensive judgment information set that includes specific suspected faulty components (PFC inductors), degradation modes, and quantitative confidence assessment results.
[0030] Specifically, simply extracting multidimensional features is insufficient for reliable prediction, as the features themselves may be affected by accidental interference or other related faults. The core necessity of this step lies in achieving "fault isolation" and "credible quantification." First, through "spatiotemporal correlation analysis," it can be verified whether the inherent consistency between features conforms to the physical principles of PFC inductor degradation. For example, does current harmonic distortion precede or occur synchronously with the abnormal temperature rise of the inductor body? Is efficiency decline significantly correlated with the enhancement of specific harmonic components? This temporal causal / co-occurrence relationship and whether the spatial anomaly is limited to the inductor are key logical criteria for distinguishing inductor faults from faults in other parts of the system. Second, "fault mode matching" compares the feature patterns obtained from real-time analysis with typical degradation patterns in the knowledge base, such as whether it is core saturation-dominated or winding loss-dominated, providing a qualitative classification for prediction. Finally, "confidence calculation" is a crucial step in this scheme from qualitative analysis to quantitative decision-making. It uses algorithms (such as based on correlation strength, pattern matching consistency, etc.) to comprehensively calculate a probability value or score, quantifying the uncertainty of the current judgment. This is far more engineering-instructive than a simple "normal / abnormal" binary judgment, allowing maintenance strategies to adopt different responses based on confidence levels. Generating a "fault isolation and confidence information set" means that the system not only provides "what might be wrong" (PFC inductor), but also "how likely it is" (confidence level), providing accurate and measurable input for subsequent prediction and decision-making.
[0031] S203. Based on the PFC inductor fault isolation and confidence index set, perform quantitative modeling of health measurement and prediction of degradation trend, and generate and output a health status prediction report of the on-board charger inductor to guide preventive maintenance.
[0032] Health quantification modeling can refer to constructing one or more quantitative index models that comprehensively reflect the overall health status of PFC inductors using historical and current confidence information and characteristic data. Degradation trend prediction can refer to extrapolating and estimating the future trajectory of a health index based on its time series data using predictive algorithms. An on-board charger inductor health status prediction report can be a structured document or data message, and its content should at least include key information such as current health status assessment, degradation trend prediction, remaining useful life (RUL) estimate, or recommended maintenance time window.
[0033] Specifically, the ultimate value of fault diagnosis lies in predicting the future and guiding actions. This step is necessary to elevate "diagnosis of the current state" to "prediction of future risks," thereby achieving true "preventive maintenance." If only PFC inductor fault isolation and a confidence index set are available, the system can only inform that "the inductor may be degrading, with a certainty of X%," but cannot answer core maintenance questions such as "how long can it still be used" and "when is maintenance required." Through "quantitative health modeling," discrete confidence levels and multidimensional features are fused into one or more easily trackable trend indicators (such as a "comprehensive health index"), providing standardized input for prediction. "Degradation trend prediction" uses time series analysis, degradation models, and other methods to estimate the future trend of this index, thereby determining the rate of performance degradation. The final "prediction report" is the concentrated embodiment of the value of this solution, transforming complex analysis results into guiding information that vehicle users, maintenance technicians, or remote monitoring centers can directly understand. For example, the report may indicate "Moderate degradation of PFC inductor health, expected to require inspection after 150 charging cycles," allowing users or service providers to plan maintenance in advance before a failure occurs, thus avoiding safety hazards and inconvenience caused by sudden vehicle malfunctions during charging.
[0034] The method provided in this embodiment first performs time-frequency analysis on the current waveform to extract harmonic distortion features. Simultaneously, it correlates the abnormal decay relationship between inductor temperature rise and efficiency, confirming that both occur strictly synchronously only during charging, thus generating a multi-dimensional feature spectrum set characterizing PFC inductor degradation. Second, it performs spatiotemporal correlation analysis and fault mode matching on these features to verify their causal logic consistency and calculates a quantified fault confidence level. Finally, based on the confidence level sequence, it constructs a health degradation model to predict future degradation trends, automatically generating a prediction report containing remaining service life or maintenance recommendations, and outputting it to the user. This solution can accurately capture and quantify the early performance degradation of PFC inductors, achieving a shift from "fault alarm" to "fault prediction." Through multi-dimensional cross-validation and quantified confidence level assessment, it significantly improves the accuracy and reliability of predictions, effectively avoiding false alarms. The final prediction report can directly guide preventative maintenance, thereby improving the safety, reliability, and overall vehicle lifecycle maintenance efficiency of the charging system.
[0035] In some embodiments, the multi-source real-time monitoring dataset includes the AC input current waveform under charging conditions, the temperature time-series data of the PFC inductor, and real-time charging efficiency data. The AC input current waveform is subjected to high-frequency sampling and spectrum analysis to extract its total harmonic distortion rate and the proportion of specific harmonic amplitudes as current waveform distortion features. A correlation analysis is performed between the dynamic relationship between the temperature change of the PFC inductor and the real-time charging efficiency, analyzing the abnormal decay rate of efficiency with increasing temperature as an efficiency-temperature correlation anomaly feature. The current waveform distortion feature and the efficiency-temperature correlation anomaly feature are calibrated to strictly appear within the time window from charging start to end as an abnormal synchronization feature. The current waveform distortion feature, the efficiency-temperature correlation anomaly feature, and the abnormal synchronization feature are integrated to generate a multi-dimensional feature spectrum set for the PFC inductor.
[0036] The AC input current waveform can be the current signal waveform flowing through the input terminal (usually the AC side) of the on-board charger when it draws power from the grid, acquired in real time by a high-precision on-board current sensor. The temperature time-series data of the PFC inductor can be a series of temperature values changing over time, collected by temperature sensors (such as thermocouples or NTC thermistors) installed on the PFC inductor body or its heat dissipation structure. Real-time charging efficiency data can be the energy conversion efficiency value calculated and output by the on-board charger in real time during the charging process, typically the ratio of output DC power to input AC power, calculated and provided by the charger's control unit (MCU) based on voltage and current sensor data. Current waveform distortion characteristics can be a quantitative indicator representing the degree of deviation from a standard sine wave, obtained through frequency domain analysis of the AC input current waveform. This mainly includes the total harmonic distortion (THD) and the proportion of specific harmonic amplitudes related to the switching frequency. Efficiency-temperature correlation anomaly can be a quantitative indicator describing an abnormal pattern in the dynamic relationship between local temperature changes in the PFC inductor and changes in the overall system charging efficiency during charging. The core characteristic is the "abnormal rate of efficiency decay with increasing temperature." Anomaly synchronicity can be a logical and temporal identifier used to confirm that the current waveform distortion characteristic and the efficiency-temperature correlation anomaly are only synchronously active within the complete time interval of the on-board charger's effective operation, and disappear under other operating conditions.
[0037] Specifically, traditional on-board diagnostic systems for passive components like PFC inductors often rely on single-parameter threshold alarms (such as alarms for temperatures exceeding 85°C), which is a serious flaw. The performance degradation of a PFC inductor is a gradual, multi-physical process with very subtle initial symptoms. For example, slight core aging may only cause the harmonic current amplitude at the switching frequency to slowly increase from the standard value of 1.5% to 2.8%, while the inductor's temperature rise rate may only be 10% faster than the historical baseline, and the overall system efficiency decreases by less than 0.5%. These subtle changes, individually within the traditional threshold range, are considered normal fluctuations and ignored, but when correlated, they are clear signals of early faults. To address these issues, this step uses a Hall current sensor to acquire the AC input current waveform at a sampling frequency of at least 100kHz; collects temperature timing data using a thermocouple mounted on the surface of the PFC inductor core; and obtains real-time charging efficiency data through the on-board charger controller's internal computing unit. Subsequently, a Fast Fourier Transform (FFT) is performed on the AC input current waveform to extract its total harmonic distortion (THD). The percentage of the amplitude of a specific harmonic (e.g., the second harmonic of the switching frequency) associated with the PFC circuit switching frequency relative to the fundamental frequency amplitude is calculated, serving as the current waveform distortion characteristic. Simultaneously, time-series alignment and correlation analysis are performed on the temperature time-series data and the real-time charging efficiency data. A linear regression method is used to calculate the rate of decrease in charging efficiency within a unit temperature rise interval (e.g., for every 1 degree Celsius increase in temperature). This rate is compared with a preset health benchmark model, and abnormal decay rates exceeding a threshold range are quantified as efficiency-temperature correlated abnormal characteristics. Furthermore, the system adds precise timestamps to the current waveform distortion characteristics and efficiency-temperature correlated abnormal characteristics, and performs a rigorous logical comparison between their active time windows and the duration of the charging enable signal issued by the on-board charger controller. An abnormal synchronization characteristic identifier is generated only when the active periods of both types of characteristics are completely within the effective level duration of the charging enable signal. Finally, the current waveform distortion characteristics, efficiency-temperature correlation anomaly characteristics, and abnormal synchronization characteristics are structurally integrated to output a multi-dimensional feature spectrum set of PFC inductors.
[0038] The method provided in this embodiment, based on the joint extraction of multi-dimensional features and synchronization verification, separates weak fault features dominated only by PFC inductor degradation from complex system noise, achieving true early warning and overcoming the fundamental defects of traditional methods such as high false negative rate and inability to warn of progressive faults.
[0039] In some embodiments, a fast Fourier transform is performed on the AC input current waveform to obtain the current spectrum of the AC input current waveform. By comparing the current spectrum with the standard sinusoidal current spectrum, the real-time total harmonic distortion (THD) of the AC input current waveform is analyzed. Simultaneously, the amplitude sequence of specific harmonic components related to the switching frequency of the PFC circuit is extracted from the current spectrum, and a linear regression method is used to analyze the growth slope and correlation coefficient of this amplitude over time to quantify its regularity enhancement trend. If the THD continuously increases and the regularity of the specific harmonic components enhances, then the THD and the specific harmonic components are used as current waveform distortion characteristics characterizing the performance degradation of the PFC inductor core.
[0040] The Fast Fourier Transform (FFT) is an algorithm that converts a time-domain signal into a frequency-domain signal. It is used to decompose the frequencies and amplitudes of each harmonic from a sampled time-domain current waveform, thus obtaining the current spectrum. The current spectrum can be a set of frequency-domain data representing the magnitude and phase relationship of each frequency component (fundamental wave, harmonics) after the AC input current waveform has been transformed by FFT. A standard sinusoidal current spectrum refers to the frequency-domain representation of a pure sinusoidal AC current under ideal conditions, where its energy is completely concentrated at the fundamental frequency, and the amplitudes of other harmonic components are theoretically zero, serving as a benchmark for comparative analysis. The real-time total harmonic distortion (THD) of the current can be a quantitative indicator measuring the degree to which the current waveform deviates from a sine wave, defined as the percentage of the root sum of the squares of the effective values of all harmonic components other than the fundamental wave to the effective value of the fundamental wave. A specific harmonic component amplitude sequence refers to an array formed by arranging the amplitudes of one or more harmonics directly related to the switching frequency of the PFC circuit (such as integer multiples of the switching frequency) extracted from the current spectrum over multiple consecutive sampling analysis periods in chronological order. Linear regression is a statistical method used to analyze the linear relationship between two variables. In this process, it is used to fit the trend of the amplitude sequence over time, calculating its growth slope and correlation coefficient. The growth slope refers to the slope of the straight line obtained through linear regression, used to quantify the average rate of change of a specific harmonic amplitude over time. The correlation coefficient is a statistic that measures the strength of the linear relationship between the amplitude of a specific harmonic and time; the closer its absolute value is to 1, the stronger the regularity. An increasing regularity trend means that the amplitude of a specific harmonic component not only shows a stable increase (positive growth slope) over time, but this increase process also has a high degree of linear correlation, rather than random fluctuations.
[0041] Specifically, traditional harmonic analysis often stops at calculating the instantaneous or average value of the total harmonic distortion (THD) and comparing it with a fixed threshold. This method has significant drawbacks: on the one hand, the THD value is easily affected by external interference such as background harmonics in the power grid and sudden load changes, leading to false alarms; on the other hand, it cannot capture the early, subtle characteristics of the specific fault mode of PFC inductor core performance degradation. The degradation of the PFC inductor core (such as a decrease in magnetic saturation point and an increase in hysteresis loss) will first and primarily affect the specific harmonics most relevant to the circuit switching action (such as the harmonic at 100kHz), causing its amplitude to exhibit a regular, continuous, and slow increase. For example, within a complete charging cycle, the amplitude of a specific harmonic of a healthy inductor may fluctuate randomly between 0.5A and 0.8A; while for an inductor in the early stages of degradation, the amplitude of this harmonic may increase regularly from an initial 0.6A to 0.9A at a stable rate of approximately 0.05A per charging cycle, while the THD value also slowly climbs from 3.0% to 4.5%. Traditional methods may consider a THD value of 4.5% as still within an acceptable range (e.g., setting the threshold to 5.0%) and ignore it, and cannot distinguish whether this increase is a regular fault trend or random interference. To address these issues, this step performs a Fast Fourier Transform on the acquired high-frequency AC input current waveform to obtain its current spectrum. First, the obtained spectrum is compared with the spectrum of an ideal sine wave to calculate its real-time total harmonic distortion rate. Secondly, the algorithm precisely identifies specific subharmonic components (e.g., second harmonic 100kHz) associated with the PFC circuit switching frequency (e.g., 50kHz) from the spectrum and extracts their amplitudes over multiple consecutive time windows (e.g., every 5 minutes), forming an amplitude sequence. Subsequently, a linear regression method is used to analyze this sequence: fitting the data with time as the independent variable and amplitude as the dependent variable. The calculated growth slope (e.g., 0.05A / h) quantifies the average growth rate of the amplitude, while the calculated correlation coefficient (e.g., 0.92) quantifies the regularity and stability of this growth trend. Finally, the system makes a joint judgment: when the real-time total harmonic distortion rate of the current shows a continuous upward trend (e.g., from 3.0% to 4.5%), and the growth slope of the specific subharmonic amplitude sequence is positive and the absolute value of the correlation coefficient is greater than a preset threshold (e.g., greater than 0.85), indicating a regular increasing trend, both are jointly determined as current waveform distortion characteristics characterizing the performance degradation of the PFC inductor core.
[0042] By focusing on specific subharmonics strongly correlated with PFC switch operation and analyzing the regular growth trend of their amplitude over time, the method provided in this embodiment can keenly capture early and specific electrical signs of PFC inductor core performance degradation, thereby significantly improving the accuracy and foresight of fault warning and overcoming the shortcomings of traditional methods that rely solely on THD thresholds for judgment, which have high false alarm rates and miss early faults.
[0043] In some embodiments, within the same or similarly preset charging power range, a historical baseline correlation map between the temperature of the PFC inductor and the real-time charging efficiency is established using a Gaussian process regression method. During the real-time acquisition of multi-source real-time monitoring datasets, the deviation of the current charging efficiency from the expected efficiency in the historical baseline correlation map is analyzed under the same temperature rise conditions. Simultaneously, the ratio of efficiency decrease to temperature increase within a unit time window is analyzed and defined as the efficiency-temperature sensitivity coefficient. If the deviation continues to expand and the absolute value of the efficiency-temperature sensitivity coefficient increases abnormally, the deviation and the efficiency-temperature sensitivity coefficient are taken as abnormal characteristics of efficiency-temperature correlation.
[0044] Gaussian process regression is a nonparametric machine learning method based on Bayesian statistics, suitable for establishing complex nonlinear probabilistic models between input variables (such as temperature) and output variables (such as efficiency). Its output includes not only the predicted mean but also variance representing uncertainty. A historical benchmark mapping can be a model generated by training Gaussian process regression on historical healthy operating data (i.e., multiple charging cycles of data collected when the PFC inductor is in good working order), mapping the probability distribution (mean and confidence interval) of the system efficiency to the current temperature input. The degree of deviation can be a quantitative indicator used to measure the difference between the current measured charging efficiency value and the predicted mean efficiency (i.e., expected efficiency) from the historical benchmark mapping under the same or similar charging power range and temperature conditions. It is usually expressed as an absolute difference or a relative percentage. The efficiency-temperature sensitivity coefficient can be a dynamic derivative index, defined as the ratio of the decrease in charging efficiency (in percentage) to the increase in temperature at the PFC inductor monitoring point (in degrees Celsius) within a set unit time analysis window (e.g., 10 minutes), quantifying the instantaneous response rate of efficiency to temperature changes.
[0045] Specifically, traditional condition monitoring methods for analyzing heat-related faults in PFC inductors typically monitor independently whether the temperature exceeds a threshold or the efficiency falls below a threshold. This method has fundamental limitations. The essence of inductor performance degradation is increased losses, which are simultaneously and dynamically reflected in both "abnormal heat generation" and "efficiency decline." Under constant charging power, the temperature-efficiency relationship of a healthy inductor is stable and predictable; however, for a degraded inductor, due to increased core or winding losses, the efficiency sacrifice for the same temperature rise will be greater. For example, at a 20kW charging power, when the PFC inductor temperature rises from 40℃ to 70℃, the efficiency of a healthy system may only decrease from 95.0% to 94.2% (a decrease of 0.8%); while for an inductor in the early stages of degradation, within the same temperature rise range, the efficiency may drop sharply from 94.8% to 93.0% (a decrease of 1.8%). To address the above issues, this step first collects and filters historical health data under various stable charging power ranges (e.g., 3kW, 7kW, 11kW) during long-term vehicle operation. For each power range, Gaussian process regression is used to train a massive dataset of PFC inductor temperature and corresponding system charging efficiency data, generating a historical benchmark mapping for that power range. This mapping can predict the expected efficiency value and its reasonable fluctuation range (e.g., 95% confidence interval) at any temperature. In real-time monitoring, the system first determines the preset range to which the current charging power belongs and calls the corresponding benchmark mapping. The algorithm continuously compares whether the deviation of the actual charging efficiency from the expected efficiency predicted by the mapping at the current measured PFC inductor temperature continues to widen and exceeds the confidence interval. Simultaneously, the algorithm calculates the efficiency-temperature sensitivity coefficient in real time within a sliding time window (e.g., every 10 minutes). (For example, if a 15°C increase in temperature is accompanied by a 1.3% decrease in efficiency, the sensitivity coefficient is approximately -0.087% / °C.) Finally, the system makes a comprehensive judgment: when the negative deviation of the real-time efficiency value from the historical benchmark shows a continuously increasing trend (for example, the deviation value gradually increases from -0.2% to -0.8%), and the absolute value of the calculated efficiency-temperature sensitivity coefficient is significantly and continuously higher than the healthy historical benchmark value under this power range (for example, the average value of the healthy benchmark sensitivity coefficient is -0.033% / ℃, while it is currently around -0.08% / ℃), then this combination is judged as an abnormal efficiency-temperature correlation characteristic that indicates an abnormal increase in the loss of the PFC inductor.
[0046] The method provided in this embodiment can accurately identify abnormal accelerated decay patterns of efficiency with temperature changes, thereby providing early warning of gradual degradation inside the inductor before the absolute values of efficiency or temperature exceed the standard. This significantly improves the detection capability and timeliness of latent faults, overcoming the shortcomings of traditional independent threshold methods that cannot correlate and analyze coupled variables and have a high false negative rate.
[0047] In some embodiments, precise timestamps are added to the occurrence process of current waveform distortion features and efficiency-temperature correlation anomaly features: the entire timeline of the two from their appearance, change to disappearance is analyzed and strictly compared with the start-stop control signal of the charger; if and only if the active timelines of current waveform distortion features and efficiency-temperature correlation anomaly features are completely wrapped within the effective working time interval of the charger, and the vehicle is in a stationary condition with the charging engine running, it is determined that the current waveform distortion features and efficiency-temperature correlation anomaly features have strict abnormal synchronicity.
[0048] A timestamp can be a time-series label attached by the system to each extracted feature data point (such as a current harmonic amplitude or an efficiency-temperature sensitivity coefficient), precisely recording the time of its generation. It is typically generated based on a unified vehicle network time protocol and used for subsequent timing alignment and correlation analysis. A start-stop control signal can be a hard-wired signal or network message signal (such as a CAN signal) issued by the on-board charger controller to control the on / off state of the charging main circuit. Its effective level (such as a high level) duration precisely corresponds to the complete operating interval of the charger from start to stop. The effective operating time interval can be the entire time span from when the start-stop control signal becomes effective (such as a rising edge) to when it becomes ineffective (such as a falling edge), during which time the power conversion circuit of the on-board charger is active. The stationary charging condition can be a specific vehicle operating state, referring to a state where the vehicle's high-voltage system is powered on and in the charging process, but the vehicle's drive system is activated and the vehicle is stationary. This state is comprehensively determined through the status messages of the vehicle's main controller (VCU).
[0049] Specifically, traditional diagnostic methods, when identifying electrical or thermal anomalies, often lack rigorous examination of the operating conditions under which these characteristics occur. This can easily lead to misjudging interference from other systems or transient anomalies as the target fault, resulting in a high false alarm rate. For example, generator harmonic interference during vehicle operation, external electromagnetic interference when stationary, or instantaneous efficiency fluctuations and temperature rises caused by the start-stop of heavy loads such as air conditioning compressors can all generate isolated feature fragments similar to PFC inductor faults. If these non-charging-condition features are included in the analysis without differentiation, it will severely interfere with the judgment of the true degradation state of the PFC inductor. Therefore, the core reason for constructing strict anomaly synchronization features is the need to establish a precise "time and logic filter" to ensure that the extracted current waveform distortion and efficiency-temperature correlation anomalies are within the PFC inductor's "responsible operating conditions." To address the above issues, this step synchronously records the corresponding high-precision timestamp for each data point generated for the current waveform distortion and efficiency-temperature correlation anomaly features. Subsequently, the diagnostic algorithm calls and parses the start-stop control signal waveform from the charger controller, accurately calibrating the start and end times of each effective operating interval. Simultaneously, the algorithm continuously monitors the status messages of the vehicle controller to confirm that the vehicle is in a charging-enabled, stationary state with "high voltage on, charging enabled, and vehicle speed at zero" without being turned off. When determining synchronization, the algorithm first compares the timestamp sequences of all data points for each of the two types of abnormal features with the effective working range of the charger, time by time. It requires that the occurrence of feature data points must be entirely within a certain effective working range, and no data points are allowed to appear outside that range. For example, if a charging session lasts 30 minutes, and the extracted harmonic amplitude abnormal feature data only covers minutes 10 to 25, and the efficiency-temperature abnormal feature covers minutes 12 to 28, and the vehicle status confirms that the vehicle has been stationary charging throughout these 30 minutes, then the active timelines of both are determined to be completely enclosed within the effective working range. Finally, the system generates an abnormal synchronization feature identifier with a logical value of "true" only when the active periods of the current waveform distortion feature and the efficiency-temperature correlation abnormal feature both meet the above "complete embedding" condition, and the vehicle's operating condition always meets the stationary charging requirements throughout the entire determination period.
[0050] The method provided in this embodiment can strictly distinguish between the actual fault symptoms generated by the PFC inductor itself under charging conditions and interference signals introduced by other systems or operating conditions, thereby greatly reducing the false alarm rate of fault diagnosis, ensuring the accuracy and reliability of subsequent fault isolation and confidence assessment, and laying a solid logical foundation for achieving accurate predictive maintenance.
[0051] In some embodiments, a three-dimensional correlation framework of time, space, and logic is constructed based on the multi-dimensional feature spectrum set of PFC inductors. Based on the logic in the three-dimensional correlation framework of time, space, and logic, cross-validation is performed on current waveform distortion features, efficiency-temperature correlation anomaly features, and anomaly synchronization features to generate a multi-dimensional feature correlation strength index that characterizes the consistency of causal linkage between features. Based on the multi-dimensional feature correlation strength index obtained at different times, the comprehensive confidence score of the motor characterizing the performance degradation of PFC inductors at different times is calculated. All multi-dimensional feature correlation strength indices are integrated to generate a PFC inductor fault isolation and confidence information set.
[0052] The three-dimensional correlation framework of time, space, and logic can refer to an analytical model used for structured analysis and verification of the intrinsic relationships between multi-dimensional characteristics of PFC inductors. Cross-validation refers to the analytical process within this framework, placing the current waveform distortion characteristics, efficiency-temperature correlation anomaly characteristics, and anomaly synchronization characteristics under the three dimensions of time, space, and logic for mutual verification and consistency checks. The multi-dimensional feature correlation strength index can be a quantitative comprehensive indicator used to characterize the extent to which the linkage relationship between various features conforms to the preset fault physical model after three-dimensional cross-validation; a higher value indicates a stronger internal consistency of the feature set pointing to a single fault source in the PFC inductor. The motor comprehensive confidence score can be a final quantitative value that integrates the multi-dimensional feature correlation strength index and the spatial locality verification results, used to characterize the system's degree of certainty (confidence level) in the diagnostic conclusion that "the PFC inductor exhibits performance degradation."
[0053] Specifically, traditional fault diagnosis, after extracting multiple abnormal features, often uses simple "AND / OR" logic for combination and judgment, lacking systematic verification of the deep physical correlation and spatiotemporal consistency between features. This may lead to two types of misjudgments: first, misassociating anomalies with different root causes but occurring simultaneously by chance as the same fault (for example, current distortion caused by power grid harmonics and efficiency decline caused by aging cooling fans may be misjudged as inductor faults); second, the inability to quantify the credibility of diagnostic conclusions, making early warning decisions lack a basis. The true degradation of PFC inductors is a deterministic, strongly correlated process with a specific evolutionary sequence in time (electrical anomalies precede thermal anomalies), locality in space (temperature rise concentrated in the inductor body), and logical conformity to the energy loss transfer law (increased loss → harmonics and temperature rise → decreased efficiency). To address the above problems, this step first cross-validates the multi-dimensional feature spectrum set of PFC inductors based on the constructed three-dimensional correlation framework of time, space, and logic (the specific construction process is described in subsequent claims). In the time dimension, the system verifies whether the aggravation of current waveform distortion begins in the early stages of charging and precedes the appearance of efficiency-temperature anomalies. In the spatial dimension, it verifies whether the abnormal temperature rise is limited to the PFC inductor monitoring point, while other heat sources within the charger (such as the switching transistor heatsink) experience normal temperature rises. In the logical dimension, it verifies whether the dynamic changes among the total harmonic distortion rate of the current, the PFC inductor temperature rise rate, and the efficiency decrease value conform to the preset causal model of "increased losses triggering harmonics and temperature rise, leading to efficiency decrease." Based on the above verification results, the system calculates a multidimensional feature correlation strength index. For example, it weights and fuses the degree of synchronicity of temporal correlation, the significance of spatial locality, and the correlation coefficient of logical causality to generate an index value between 0 and 1. Subsequently, the system inputs this index along with the spatial locality level into a preset performance degradation scoring model. The model employs a piecewise linear mapping algorithm. For example, when the correlation strength index is higher than 0.8 and the spatial locality level is "high," it is directly mapped to a high confidence score (e.g., 90 points). If the correlation strength index is between 0.6 and 0.8, the score is calculated using linear interpolation based on the spatial locality level. Finally, the system outputs the comprehensive confidence score for this motor and integrates it with the intermediate results of verification across various dimensions to form a structured set of PFC inductor fault isolation and confidence information.
[0054] The method provided in this embodiment can effectively eliminate misjudgments caused by occasional interference and related system anomalies, ensuring that the diagnostic conclusions accurately point to the PFC inductor itself, and outputting them in the form of a quantitative confidence score, providing a solid and reliable data basis for implementing differentiated predictive maintenance strategies.
[0055] In some embodiments, based on the start / stop signal of the charger, all feature data in the PFC inductor multi-dimensional feature spectrum set are time-aligned to construct a time-related dataset; the set of monitoring points at different physical locations within the on-board charger is identified and located, and the feature data involving temperature and electrical response in the PFC inductor multi-dimensional feature spectrum set are mapped and grouped according to their source and the set of monitoring points to form a spatial correlation dataset for comparing local and global state differences; based on the basic principles of electricity and thermodynamics, the theoretical causal relationship link between current waveform distortion features, efficiency-temperature correlation anomaly features, and temperature features is defined, which serves as a three-dimensional correlation framework for verifying whether the data linkage relationship in the multi-dimensional feature spectrum set conforms to physical laws.
[0056] A time-related dataset can refer to a standardized dataset formed by performing time-series alignment and interpolation on all time-stamped feature data (such as the total harmonic distortion rate sequence of current, the temperature sequence of PFC inductor, and the efficiency value sequence) in the multi-dimensional feature spectrum set of the PFC inductor, using the start-stop signal of the charger as a unified reference. All feature points are located on the same time axis. A spatially related dataset can refer to a grouped dataset formed by classifying and mapping feature data from different spatial locations in the multi-dimensional feature spectrum set, based on the identification and location of various physical monitoring points within the on-board charger (such as the temperature sensing point of the PFC inductor, the temperature sensing point of the switching transistor heatsink, and the input current sampling point). This dataset is used to compare and analyze the locality and globality of abnormal signals. A theoretical causal relationship chain can be a theoretical path derived in advance based on basic electrical principles (such as Ohm's law and the core loss formula) and thermodynamic principles (such as the heat conduction equation and the efficiency formula), describing a series of chain reactions caused by the performance degradation of the PFC inductor.
[0057] Specifically, traditional fault characteristic analysis often examines time series, spatial location, or physical principles in isolation, lacking a unified analytical structure that organically integrates these three aspects. This results in loose diagnostic logic, making it highly susceptible to misjudgments due to occasional interference or anomalies in related systems. For example, observing only a current harmonic that intensifies over time cannot confirm whether its thermal effect is limited to the PFC inductor (lack of spatial verification); similarly, observing localized temperature rise and efficiency decline, without verifying whether they strictly follow the current harmonic and occur synchronously within the charging cycle (lack of temporal verification), may lead to misjudgment as an independent fault in the cooling system. To address these issues, this step first uses the high-priority charger start / stop control signal as a global time reference. Utilizing timestamp alignment and data interpolation techniques, all characteristic data, including current waveform distortion features, PFC inductor temperature timing, and efficiency timing, are unified onto an absolute time axis defined by this reference, generating a time-correlated dataset to ensure the comparability of all characteristic values at any given time. Next, the system calls upon a pre-stored digital twin model of the on-board charger to identify the physical coordinates and functions of key monitoring points. For example, sensor TS1 is located in the middle of the PFC inductor core, while TS2 is located on the heatsink of the main switching transistor. Based on this mapping relationship, the system categorizes temperature characteristic data into spatial groups such as "PFC inductor part" and "other heat source parts," and electrical characteristic data into logical groups such as "input side" or "output side," forming a spatially correlated dataset. Finally, based on domain knowledge (such as the relationship between core loss and frequency and magnetic flux density, and the relationship between heat accumulation and power loss and thermal resistance), the system incorporates a theoretical causal relationship chain as the logical dimension framework. For example: increased inductor core loss (not directly measurable) → leads to an increase in the current harmonic distortion rate (THD) (measurable) → causes an acceleration in the temperature rise rate ΔT / Δt of local hot spots in the PFC inductor (measurable) → ultimately manifests as an accelerated decrease in system efficiency η with the temperature rise at that point (measurable).
[0058] By using the method provided in this embodiment, discrete feature signals are correlated with the actual physical structure and operation of the device, laying a solid foundation for moving from "feature perception" to "causal diagnosis", and significantly improving the directionality, systematicity and reliability of subsequent fault isolation analysis.
[0059] In some embodiments, time-related verification is performed: verifying that the aggravation of current waveform distortion begins in the early stage of the charging process and precedes the appearance of efficiency-temperature correlation anomalies. Simultaneously, it analyzes whether the intensity of current waveform distortion and the temperature rise rate at the PFC inductor monitoring point show a synchronous aggravation trend throughout the charging process. Spatial-related verification is performed: comparing the temperature rise rate of the PFC inductor with the temperature rise rates of other major heat sources within the on-board charger. If only the PFC inductor exhibits an independent abnormal temperature rise, while other heat sources show normal temperature rises, it confirms that the thermal anomaly has spatial locality. Logical-related verification is performed: based on a preset fault causal chain model, analyzing whether the data linkage relationship of each feature in the multi-dimensional feature spectrum set of the PFC inductor is consistent with the theoretical derivation of this model. The theoretical transmission logic of the fault causal chain model is that inductor performance degradation leads to increased losses and harmonics, causing localized temperature rise anomalies, ultimately resulting in a decrease in system efficiency.
[0060] Temporal correlation verification can be performed within the time dimension of the three-dimensional correlation framework to verify whether the intensification of the current waveform distortion characteristics (such as total harmonic distortion rate) begins in the early stages of the charging process and precedes the significant manifestation of the efficiency-temperature correlation anomaly characteristics (such as the abnormal sensitivity coefficient of efficiency to temperature rise). Simultaneously, it analyzes whether the trends of these two characteristics are synchronous throughout the charging process. Spatial correlation verification can be performed within the spatial dimension of the three-dimensional correlation framework by comparing the temperature rise rate of the monitoring points at the PFC inductor with the temperature rise rates of other major heat sources within the on-board charger (such as the main switch heatsink and output diode heatsink) to confirm whether the thermal anomaly has a localized characteristic limited to the PFC inductor itself. Logical correlation verification can be performed within the logical dimension of the three-dimensional correlation framework, based on a preset fault causal chain model, by comparing the actual data linkage relationships in the multi-dimensional feature spectrum set of the PFC inductor with the theoretical transmission logic derived from the model to verify whether the measured data conforms to the physical causal law caused by the single root cause of "inductor performance degradation." The fault causal chain model can be a pre-defined theoretical model that describes a series of chain reactions triggered by the performance degradation of PFC inductors. Its core logic chain is "inductor performance degradation → increased losses and harmonics → (PFC inductor) abnormal local temperature rise → decreased system efficiency".
[0061] Specifically, traditional fault verification methods often perform isolated time-series analysis or spatial comparisons, lacking a systematic verification that links temporal evolution, spatial distribution, and physical causal logic in a closed loop. For example, if only the synchronous rise of current harmonics and temperature in time is verified, but if it is not further verified whether this temperature rise occurs only at the PFC inductor (lack of spatial locality), the interference of poor overall heat dissipation of the charger cannot be ruled out. Conversely, if only the local overheating of the PFC inductor is verified, but if it is not verified whether this overheating strictly follows the logical sequence of "current harmonics intensify first, then temperature rise becomes abnormal" (lack of causal logic), the inductor temperature rise caused by external heat source radiation may be misjudged as being caused by its own losses. To address the above issues, this step, in the time-correlation verification, systematically analyzes the time-correlation dataset to specifically verify whether the significant increase in the total harmonic distortion (THD) of the current begins in the initial stage after charging starts (such as the first 5 minutes) and is detected before the abnormal acceleration of the PFC inductor temperature rise rate (ΔT / Δt). Simultaneously, the system calculates the dynamic correlation coefficient (e.g., Pearson correlation coefficient) between the THD sequence and the ΔT / Δt sequence throughout the entire charging cycle, requiring its value to remain consistently above a preset threshold (e.g., 0.7) to confirm that the intensity changes of both show a synchronous increasing trend. In spatial correlation verification, the system calls the spatial correlation dataset to compare the temperature rise rate of the PFC inductor monitoring point (TS_PFC) and the switching transistor heatsink monitoring point (TS_Switch). For example, in one charge, if the temperature rise rate of TS_PFC reaches 1.5°C per minute, while the temperature rise rate of TS_Switch is only 0.3°C per minute and consistent with the ambient temperature rise baseline, then the thermal anomaly is determined to have high spatial locality, pointing to the PFC inductor itself. In logical correlation verification, the system compares the observed data linkage sequence of "THD rises first → ΔT / Δt rises later → efficiency η decreases rapidly with temperature rise" with the theoretical transmission logic of the preset fault causal chain model link by link. The system quantifies logical consistency by calculating the statistical relationships between characteristic data of each link (such as the negative correlation between the THD growth slope and the efficiency decline slope). Only when the actual data linkage significantly meets the model's expectations in terms of intensity, timing, and correlation is the logical association verification deemed successful.
[0062] The method provided in this embodiment not only links multi-dimensional anomalies in terms of phenomena, but also verifies their inherent unity in terms of physical essence and evolutionary laws. It can effectively eliminate interference from external interference and related system anomalies, ensuring that the diagnostic conclusion accurately and reliably locks the PFC inductor itself, and provides an irrefutable logical basis for the subsequent generation of high-confidence performance degradation reports.
[0063] In some embodiments, based on the results of time-related verification and logical-related verification, the dynamic correlation degree between the intensity of the current waveform distortion characteristics and the temperature rise rate of the PFC inductor monitoring point, as well as the statistical correlation coefficient between the total harmonic distortion rate of the current and the temperature rise rate, are calculated; the weighted sum of the dynamic correlation degree and the statistical correlation coefficient is used as the multidimensional feature correlation strength index; the multidimensional feature correlation strength index is input into a preset performance degradation scoring model, and the performance degradation scoring model, based on the thermal anomaly locality level confirmed by spatial correlation verification and the multidimensional feature correlation strength index, calculates the comprehensive confidence score of the motor representing the degree of performance degradation of the PFC inductor at different times through a piecewise linear mapping algorithm.
[0064] Dynamic correlation can be a quantitative indicator obtained by analyzing time-correlated datasets and calculating the degree of synchronous change between the intensity sequence of current waveform distortion characteristics (such as total harmonic distortion rate, THD) and the temperature rise rate sequence of the PFC inductor monitoring point throughout the entire charging cycle. Statistical correlation coefficient can be a coefficient obtained in logical correlation verification by using statistical methods (such as Pearson correlation coefficient) to calculate the linear correlation between the two sets of data: total harmonic distortion rate and PFC inductor temperature rise rate. The preset performance degradation scoring model can be a pre-built mathematical model stored in the system. Its inputs are a multidimensional feature correlation strength index and a spatial locality level, and its output is a comprehensive motor confidence score. This model encapsulates the mapping rules from feature correlation to diagnostic confidence. The piecewise linear mapping algorithm can be a calculation method used within the preset performance degradation scoring model. This algorithm selects different linear function segments for mapping calculation based on the different numerical intervals of the multidimensional feature correlation strength index and the spatial locality level, ultimately outputting a continuous confidence score.
[0065] Specifically, traditional fault diagnosis, after obtaining multiple verification indicators, typically employs simple threshold judgments or fixed formula weighting, lacking a confidence quantification model capable of systematically integrating multidimensional evidence and simulating the expert comprehensive judgment process. For example, even if time-correlation verification shows that current distortion and temperature rise are highly synchronized (high dynamic correlation), and logical verification shows a strong correlation between the two (high statistical correlation coefficient), if spatial verification finds that other heat sources also have abnormal temperature rises (low spatial locality level), the reliability of the diagnostic conclusion should be significantly reduced. Simply averaging the first two indicators will severely overestimate the confidence level, leading to misjudgment. To address these issues, this step first calculates the dynamic correlation between the total harmonic distortion (THD) sequence and the PFC inductor temperature rise rate (ΔT / Δt) sequence throughout the entire charging cycle, based on the results of time-correlation verification. For example, by calculating the local correlation coefficient of the two sequences using a sliding window and averaging them, if ΔT / Δt also rises synchronously with THD and the trends are consistent, the dynamic correlation can reach a high value (e.g., 0.85). Simultaneously, based on logical correlation verification, the statistical correlation coefficient (e.g., Pearson correlation coefficient) between the THD sequence and the ΔT / Δt sequence is calculated. If the two show a significant positive linear correlation, the coefficient is high (e.g., 0.78). The system weights and sums the dynamic correlation degree and the statistical correlation coefficient with preset weights (e.g., 0.6 and 0.4) to generate a multidimensional feature correlation strength index (e.g., 0.85×0.6+0.78×0.4=0.822). Subsequently, the system inputs this index along with the thermal anomaly locality level (e.g., "high", "medium", "low") determined by spatial correlation verification into a preset performance degradation scoring model. The model employs a piecewise linear mapping algorithm: for example, if the spatial locality level is "high" and the multidimensional feature association strength index is greater than 0.8, it is directly mapped to a high confidence interval (e.g., 90 to 100 points), with the specific score calculated through interpolation (e.g., an index of 0.822 maps to approximately 92 points); if the index is between 0.6 and 0.8, a linear mapping with different slopes is applied based on the spatial locality level; if the index is below 0.6, the confidence score is suppressed to a low level (e.g., below 60 points) regardless of the locality. Finally, the system calculates the comprehensive confidence score of the motor corresponding to this diagnosis.
[0066] The method provided in this embodiment constructs a refined confidence quantification model that integrates multi-dimensional verification evidence, transforming the qualitative verification results of the three dimensions of time, space, and logic into a continuous, sensitive, and interpretable comprehensive score, thus realizing the leap from "multiple indicators alarm" to "one confidence level decision".
[0067] In some embodiments, a comprehensive health index model is constructed based on historical and current PFC inductor fault isolation and confidence index sets and corresponding PFC inductor multi-dimensional feature spectrum data. A time series analysis method is used to fit the change trajectory of the comprehensive health index model to predict the future decline trend of the comprehensive health index. Based on the predicted decline trend curve of the comprehensive health index and a preset safety threshold line, the remaining reliable service life or the time window for the next recommended maintenance of the on-board charger PFC inductor is estimated. The current health status, decline trend prediction, and maintenance time window are integrated to generate a structured on-board charger inductor health status prediction report.
[0068] The comprehensive health index model can be a mathematical model that comprehensively quantifies the overall health status of PFC inductors, constructed based on historical and current multi-batch diagnostic data through statistical analysis or machine learning methods. The model's output is a health index sequence that evolves over time, typically ranging from [0,1] to [0,100], where high scores represent health and low scores represent decline. The time series analysis method can be a statistical technique used to analyze the time dependence and regularity of the output sequence of the comprehensive health index model. In this embodiment, it specifically refers to an algorithm used to fit historical health change trajectories and extrapolate to predict future trends, such as an autoregressive integral moving average model. The decline trend curve can be a curve describing the possible future change path of the health index, obtained by applying the time series analysis method to fit the historical sequence of the comprehensive health index model and then extending (predicting) it. The preset safety threshold line can be a pre-set numerical boundary within the value range of the comprehensive health index model to define the "safe" and "warning" states. When the predicted decline trend curve intersects with this line, it is considered to have reached a critical point requiring maintenance action.
[0069] Specifically, traditional threshold-based fault alarms or single-diagnostic methods can only provide a binary conclusion of "whether it is currently abnormal," failing to depict the dynamic process and future trajectory of performance degradation in critical components (such as PFC inductors). This leads to maintenance strategies that are either passively delayed (only addressed after a fault occurs) or overly conservative (blindly replacing components periodically, resulting in waste). For example, even if the current diagnostic confidence score is only moderate (e.g., 70 points), a very steep downward trend could trigger a fault in the short term; conversely, a low score (e.g., 65 points) with a gradual trend could allow for a longer buffer period. To address these issues, this step first compiles a set of PFC inductor fault isolation and confidence information (including the motor's comprehensive confidence score) and its corresponding multidimensional feature spectrum data generated during each charging process over a past period (e.g., three months). Using the motor's comprehensive confidence score as the core health indicator, and incorporating key derived indicators from the feature spectrum set (such as the moving average of the efficiency-temperature sensitivity coefficient), a comprehensive health index model that more fully reflects the inductor's performance status is constructed through principal component analysis or weighted fusion algorithms. This model outputs a health index sequence sorted by time (e.g., the index sequence corresponding to the last 90 charging cycles: [92, 91, 90, 88, 85, 83, ...]). Subsequently, the system uses time series analysis methods (such as the ARIMA model) to fit this historical sequence, analyzing its inherent downward trend, seasonality (if any), and random fluctuations. After model training, the system predicts the health index for a future period (e.g., the next 30 charging cycles), forming a decline trend curve. Then, the system calls a preset safety threshold line (e.g., a health index of 60) to calculate the future time point corresponding to the first time the decline trend curve falls below this threshold line (e.g., predicted to fall below at the 128th charging cycle). Based on the current moment (e.g., the 98th charge), the system calculates the remaining reliable service life (i.e., 130-98=32 charging cycles). Finally, combining maintenance experience (e.g., spare parts procurement cycle is about 5 days), the system defines a recommended maintenance time window before the critical point (e.g., "maintenance during the 115th to 125th charging cycles"), and structurally integrates the current health status (e.g., "current index 83, status: early decline"), the predicted decline trend curve visualization, and the clear maintenance time window to generate a graphic and textual on-board charger inductor health status prediction report.
[0070] The method provided in this embodiment realizes a qualitative leap from discrete fault diagnosis to continuous health management, transforming historical and current diagnostic data into predictive health trends and specific remaining life estimates, providing a direct and quantitative basis for decision-making in implementing precise predictive maintenance.
Claims
1. A fault prediction system for the electronic control system of a new energy vehicle, characterized in that, include: A multi-source real-time monitoring dataset is acquired. Based on the multi-source real-time monitoring dataset, time-frequency domain analysis and feature extraction are performed to separate and quantify the current waveform distortion, efficiency-temperature correlation anomaly, and abnormal synchronization characteristics dominated by the performance degradation of PFC inductors, and to generate a multi-dimensional feature spectrum set of PFC inductors. Based on the multi-dimensional feature spectrum set of the PFC inductor, spatiotemporal correlation analysis and fault mode matching between multi-dimensional features are performed. Through cross-validation and fault mode matching, the confidence level characterizing the performance degradation of the PFC inductor is calculated, and a PFC inductor fault isolation and confidence information set is generated. Based on the PFC inductor fault isolation and confidence information set, a quantitative model of health measurement and a degradation trend prediction are performed, and a vehicle charger inductor health status prediction report is generated and output to guide preventive maintenance.
2. The system according to claim 1, characterized in that, The process of generating the multi-dimensional feature spectrum set of the PFC inductor includes: The multi-source real-time monitoring dataset includes AC input current waveform under charging conditions, temperature time-series data of PFC inductor, and real-time charging efficiency data. The AC input current waveform is sampled at high frequency and subjected to spectrum analysis to extract its total harmonic distortion rate and the proportion of specific harmonic amplitude as current waveform distortion characteristics. The dynamic correlation between the temperature change of the PFC inductor and the real-time charging efficiency was analyzed, and the abnormal decay rate of efficiency with increasing temperature was analyzed as an abnormal feature of efficiency-temperature correlation. The current waveform distortion characteristics and efficiency-temperature correlation anomalies are calibrated to appear strictly within the time window from the start to the end of charging, which is used as an abnormal synchronization characteristic. By integrating the current waveform distortion features, the efficiency-temperature correlation anomaly features, and the abnormal synchronization features, a multi-dimensional feature spectrum set of the PFC inductor is generated.
3. The system according to claim 2, characterized in that, The analysis process of the current waveform distortion characteristics includes: Perform a fast Fourier transform on the AC input current waveform to obtain the current spectrum of the AC input current waveform; By comparing the current spectrum with the standard sinusoidal current spectrum, the real-time total harmonic distortion rate of the AC input current waveform is analyzed. Meanwhile, the amplitude sequence of specific subharmonic components related to the switching frequency of the PFC circuit in the current spectrum is extracted, and the linear regression method is used to analyze the growth slope and correlation coefficient of this amplitude over time to quantify its regularity enhancement trend. If the total harmonic distortion rate of the current continues to increase and the regularity of the specific harmonic components increases, then the total harmonic distortion rate of the current and the specific harmonic components are used as the current waveform distortion characteristics characterizing the performance degradation of the PFC inductor core.
4. The system according to claim 3, characterized in that, The analysis process of the efficiency-temperature correlation anomaly features includes: Within the same or similarly preset charging power range, a historical baseline correlation map between the temperature of the PFC inductor and the real-time charging efficiency is established using the Gaussian process regression method. During the real-time acquisition of the multi-source real-time monitoring dataset, the degree of deviation of the current charging efficiency from the expected efficiency in the historical benchmark correspondence map is analyzed under the same temperature rise conditions. Meanwhile, the ratio of the decrease in efficiency to the increase in temperature within a unit time window is analyzed and defined as the efficiency-temperature sensitivity coefficient. If the degree of deviation continues to increase and the absolute value of the efficiency-temperature sensitivity coefficient increases abnormally, then the degree of deviation and the efficiency-temperature sensitivity coefficient are regarded as abnormal features of the efficiency-temperature correlation.
5. The system according to claim 4, characterized in that, The analysis process of the abnormal synchronization characteristics includes: Add precise timestamps to the occurrence processes of the current waveform distortion characteristics and the efficiency-temperature correlation anomaly characteristics: The entire timeline of the two processes from their appearance, change to disappearance was analyzed and rigorously compared with the start and stop control signals of the charger. The current waveform distortion feature and the efficiency-temperature correlation anomaly feature are determined to have strict abnormal synchronization if and only if the active timelines of the current waveform distortion feature and the efficiency-temperature correlation anomaly feature are completely wrapped within the effective working time interval of the charger, and the vehicle is in a stationary condition with the charging engine running.
6. The system according to claim 5, characterized in that, The process of generating the PFC inductor fault isolation and confidence information set includes: Based on the multi-dimensional feature spectrum set of the PFC inductor, a three-dimensional correlation framework of time, space and logic is constructed. Based on the logic in the three-dimensional correlation framework of time, space and logic, the current waveform distortion feature, the efficiency-temperature correlation anomaly feature and the anomaly synchronization feature are cross-validated to generate a multi-dimensional feature correlation strength index that characterizes the causal linkage consistency between features. Based on the multidimensional feature correlation strength index obtained at different times, the comprehensive confidence score of the motor characterizing the performance degradation of PFC inductor at different times is calculated. Integrate all the multidimensional feature correlation strength indices to generate the PFC inductor fault isolation and confidence information set.
7. The system according to claim 6, characterized in that, The construction process of the aforementioned three-dimensional framework of temporal, spatial, and logical association includes: Based on the start / stop signal of the charger, time alignment is performed on all feature data in the multi-dimensional feature spectrum set of the PFC inductor to construct a time-related dataset; Identify and locate a set of monitoring points at different physical locations within the on-board charger. Map and group the feature data related to temperature and electrical response in the multi-dimensional feature spectrum of the PFC inductor according to their source to the set of monitoring points to form a spatial correlation dataset for comparing local and global state differences. Based on the fundamental principles of electricity and thermodynamics, a theoretical causal relationship link is defined between the current waveform distortion characteristics, the efficiency-temperature correlation anomaly characteristics, and the temperature characteristics. This serves as the three-dimensional correlation framework for verifying whether the data linkage relationship in the multi-dimensional feature spectrum set conforms to physical laws.
8. The system according to claim 7, characterized in that, The logic within the three-dimensional framework of time, space, and logic includes: Time-related verification: Verify that the aggravation of the current waveform distortion characteristics begins in the early stage of the charging process and precedes the appearance of the efficiency-temperature correlation anomaly characteristics. At the same time, analyze whether the intensity of the current waveform distortion characteristics and the temperature rise rate of the PFC inductor monitoring point show a synchronous aggravation trend throughout the charging process. Spatial correlation verification: Compare and analyze the temperature rise rate of the PFC inductor with the temperature rise rate of other major heat sources in the on-board charger. If only the PFC inductor shows an independent abnormal temperature rise, while the temperature rise of other heat sources is normal, then the thermal anomaly is confirmed to be spatially localized. Logical correlation verification: Based on the preset fault causal chain model, analyze whether the data linkage relationship of each feature in the multi-dimensional feature spectrum set of the PFC inductor is consistent with the theoretical derivation of this model. The theoretical transmission logic of the fault causal chain model is that the inductor performance degradation leads to an increase in losses and harmonics, causing abnormal local temperature rise, and ultimately causing a decrease in system efficiency.
9. The system according to claim 8, characterized in that, The cross-validation and overall confidence score of the motor include: Based on the results of the time correlation verification and the logic correlation verification, the dynamic correlation between the intensity of the current waveform distortion feature and the temperature rise rate of the PFC inductor monitoring point, as well as the statistical correlation coefficient between the total harmonic distortion rate of the current and the temperature rise rate, are calculated. The weighted sum of the dynamic correlation degree and the statistical correlation coefficient is used as the multidimensional feature correlation strength index; The multidimensional feature correlation strength index is input into a preset performance degradation scoring model. The performance degradation scoring model calculates the motor comprehensive confidence score, which characterizes the degree of performance degradation of the PFC inductor at different times, based on the thermal anomaly locality level confirmed by the spatial correlation verification and the multidimensional feature correlation strength index, through a piecewise linear mapping algorithm.
10. The system according to claim 9, characterized in that, The process of generating the on-board charger inductor health status prediction report includes: Based on the historical and current PFC inductor fault isolation and confidence index set and the corresponding PFC inductor multi-dimensional feature spectrum set data, a comprehensive health index model is constructed. By using time series analysis, the trajectory of the changes in the comprehensive health index model is fitted to predict the future decline trend of the comprehensive health index. Based on the predicted decline trend curve of the comprehensive health index and the preset safety threshold line, estimate the remaining reliable service life or the time window for the next recommended maintenance of the on-board charger PFC inductor. By integrating the current health status, decline trend prediction, and maintenance time window, a structured health status prediction report for the on-board charger inductor is generated.
