Aircraft power supply system no-load abnormity monitoring method and early warning system

Abstract

The invention discloses an aircraft power supply system no-load abnormity monitoring method and early warning system, and relates to the technical field of electrical variable measurement, and the method comprises the steps: executing no-load exclusive reference calibration after confirming that the system enters a stable no-load working condition, and building a no-load operation reference database; the multi-channel unit subjected to nanosecond-level synchronous calibration acquires no-load electric variables and environmental parameters; abnormal features are extracted through time domain and frequency domain conjoint analysis, and data are corrected through a no-load exclusive environment-aging coupling compensation model; a no-load exclusive risk scoring formula is adopted to quantify the risk, and an abnormal type and grade are judged in combination with a no-load exclusive abnormal recognition model; and fault tracing is completed through the combined positioning model, grading early warning is triggered, and storage data is encrypted. The method can accurately capture no-load weak abnormal signals, improve the early fault recognition rate, and guarantee the operation safety of an aircraft power supply system.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and in particular to a method and early warning system for monitoring airborne anomalies in aircraft power systems. Background Technology

[0002] The aircraft power system is a core component of the aircraft avionics system, and its operational stability directly determines the overall power supply safety and flight safety of the aircraft. The no-load operating condition is a critical operational state for the aircraft power system during ground maintenance, pre-flight inspections, and in-flight condition switching. The electrical operating characteristics of the power system under no-load conditions directly reflect the overall health level and potential fault hazards of the system, representing a critical window for early identification of electrical faults and avoidance of power outage risks during loaded operation. Current mainstream aircraft power system anomaly monitoring technologies mostly focus on fault monitoring and protection under loaded operating conditions, with a significant lack of specialized monitoring technologies for no-load conditions. The changes in the system's electrical characteristics under no-load conditions are much smaller than under loaded conditions, making it difficult for conventional threshold monitoring methods to capture weak abnormal signals and identify early, latent faults during the no-load phase. This can easily lead to missed faults and escalation of faults during loaded operation.

[0003] Existing methods for monitoring aircraft power systems have the following key shortcomings: Firstly, the monitoring solutions lack specific adaptability to no-load conditions, mostly adopting general thresholds and monitoring logic for load conditions, which cannot meet the identification requirements of weak fault characteristics under no-load conditions. Existing solutions mostly use a fixed threshold comparison mode for a single electrical variable, which can only identify serious faults exceeding the rated limit. They are severely inadequate in identifying early anomalies and latent faults under no-load conditions, and are easily affected by fluctuations in operating conditions, leading to misjudgments and missed judgments. Secondly, it cannot effectively eliminate the coupling interference between the environment and equipment aging. In aviation scenarios, environmental parameters such as temperature, humidity, and atmospheric pressure vary widely. The aging of equipment throughout its entire life cycle will cause electrical characteristics to drift. Existing technologies only use general fixed coefficient environmental compensation, without considering the coupling effect of aging and environment on weak signals under no-load conditions. This can easily cause weak fault characteristics to be masked by interference, resulting in a continuous decline in monitoring accuracy during long-term operation. Third, there is a lack of ability to quantitatively assess and accurately locate anomalies under no-load conditions. Existing solutions can only output basic anomaly alarm signals and lack a risk quantification mechanism for weak no-load characteristics, making it impossible to accurately distinguish between normal operating disturbances and actual faults. At the same time, due to the limited number of fault characteristics under no-load conditions, conventional fault location methods are prone to location errors, making it impossible to accurately locate abnormal modules and trace fault sources, and thus difficult to provide effective data support for subsequent operation and maintenance. Fourth, a closed-loop monitoring system for the entire process of airborne operation has not been established. Existing technologies have not built a dedicated monitoring system for the entire process of airborne operation, from benchmark calibration, data acquisition, interference compensation, feature recognition to positioning and early warning. This makes it impossible to adapt to the high reliability and high safety requirements of the aviation industry for power systems, and it is difficult to meet the health management needs of the aircraft power system throughout its entire life cycle. Summary of the Invention

[0004] The present invention proposes an airborne anomaly monitoring method and early warning system for aircraft power systems to solve the problems mentioned in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for monitoring airborne anomalies in an aircraft power system, comprising the following steps: Upon receiving the operating condition switching signal from the aircraft power system and entering a stable no-load operating condition, the system performs a calibration operation for no-load specific reference parameters. It collects reference data on phase voltage, line voltage, no-load leakage current, output frequency, total harmonic distortion, phase sequence phase difference, output impedance, and ripple coefficient electrical variables under rated no-load operating conditions, and establishes a reference database for normal no-load operation. The reference threshold is different from the general threshold for load operating conditions and is adapted and set in combination with the aging characteristics of the no-load full life cycle and historical no-load fault data. By using a multi-channel synchronous acquisition unit, nanosecond-level synchronous calibration is performed on each acquisition channel based on the system reference clock to eliminate phase sequence difference under no-load conditions and the time delay-sensitive deviation of impedance calculation; redundant dual-channel parallel acquisition is used for the weak signal channel of no-load leakage current to verify the validity of the data and synchronously acquire environmental temperature, humidity and atmospheric pressure parameters. The collected real-time electrical variable data are subjected to joint time-domain and frequency-domain analysis. In the time domain, the sliding window method is used to extract the characteristics of voltage fluctuation, current change, frequency shift and impedance change. In the frequency domain, the harmonic distribution characteristics are extracted by fast Fourier transform to obtain the abnormal feature parameter set, and normalization processing and invalid feature removal are completed. Based on environmental parameters and the no-load aging state of the equipment, a no-load-specific environment-aging coupling interference compensation model is adopted for correction: conventional electrical variables are compensated independently by the environment, the no-load leakage current is introduced into the aging-environment coupling correction term, the temperature interference coefficient is dynamically adjusted according to the no-load aging degree of the equipment, the masking effect of coupling interference on the weak leakage current characteristics is quantified, and the compensation and correction of electrical variables and characteristic parameters are completed. The corrected feature parameters are compared with the benchmark threshold, and the comprehensive score of abnormal risk is calculated using the empty-load-specific risk scoring formula to amplify the risk weight of the empty-load weak deviation feature; combined with the pre-trained empty-load-specific anomaly recognition model, the anomaly type is identified and the risk level is determined. Based on the system topology, a joint localization model combining Bayesian probabilistic inference and no-load fault feature matching is adopted. By correcting the posterior probability through feature matching degree, the localization deviation caused by insufficient no-load fault features is resolved, and the abnormal module localization and fault source analysis are completed. Based on the determined level of abnormal risk, a corresponding warning signal is triggered, and the entire stream of data is simultaneously encrypted and stored in a closed loop.

[0006] Furthermore, the formula for calculating the comprehensive score of abnormal risk is as follows: ;in, A comprehensive risk assessment of no-load anomalies in the aircraft's power system. The total number of core electrical variable dimensions participating in the evaluation. For the first Weighting coefficients for each electrical variable dimension. For the first Real-time collected values ​​of individual electrical variables, For the first The reference calibration value of each electrical variable, For the first The maximum permissible limit for each electrical variable For the first Nonlinear correction exponent for individual electrical variables, This represents the total number of characteristic frequencies in the harmonic analysis. For the first The influence weight of each characteristic frequency harmonic For the first The difference between the real-time distortion rate of each characteristic frequency harmonic and the reference distortion rate. For the first The maximum permissible distortion difference of each characteristic frequency harmonic. The influence coefficient for the duration of the anomaly. This refers to the duration during which abnormal feature parameters deviate from the baseline threshold.

[0007] Furthermore, it also includes an adaptive update step for the benchmark threshold under no-load conditions. The system collects no-load operation data, environmental parameters, and no-load aging characteristic parameters throughout the entire life cycle at fixed intervals. Combined with historical anomaly identification results and feedback on misjudgments and omissions, the benchmark threshold is adaptively and iteratively updated, and the dynamic coefficient table of the coupled compensation model is updated synchronously. The updated parameters are used for subsequent anomaly monitoring.

[0008] Furthermore, the formula for calculating the posterior probability of anomaly occurrence in the anomaly module location and fault tracing analysis is as follows: ; For the emergence of abnormal feature sets The anomaly occurred at the time. Power module The posterior probability, This represents the total number of topology modules in the aircraft's power system. For the first When a power module malfunctions, an abnormal feature set appears. The conditional probability, For the first Prior fault probability of each power module For abnormal feature set With the The number of matching features in the fault feature library of each power module. For abnormal feature set The total number of features.

[0009] Furthermore, the no-load-specific anomaly identification model adopts a serial fusion architecture of multi-scale convolutional neural network and bidirectional long short-term memory network. The front end extracts spatial dimension features, and the back end extracts time series dimension features. The anomaly classification results are output through a fully connected layer. The model uses full-lifecycle no-load normal data, no-load anomaly simulation data, and historical no-load measured data. In addition, leakage current samples from different aging stages and environments are added to construct the training set. The model is trained using an adaptive moment estimation algorithm with cross-entropy loss as the optimization objective, and supports incremental learning optimization.

[0010] Furthermore, the fitting process of the aging-environment coupling correction term is as follows: the no-load aging stage of the equipment is divided into several intervals, and the interference coefficient of temperature and humidity on the no-load leakage current is fitted for each interval to form a dynamic coupling coefficient table; during compensation, the interference coefficient of the corresponding interval is called to perform compensation according to the current aging state of the equipment.

[0011] Furthermore, the verification and switching rules of the redundant dual-channel acquisition architecture are as follows: the two sets of sensing units of the same precision in the no-load leakage current acquisition channel output data synchronously, calculate the deviation value of the two sets of data, and when the deviation exceeds the preset limit, automatically switch to the backup channel and trigger the channel fault warning.

[0012] Furthermore, an airborne anomaly early warning system for an aircraft power system includes a multi-dimensional electrical variable acquisition module, a reference data management module, an anomaly feature extraction module, an environmental interference compensation module, an intelligent anomaly identification module, an anomaly location and tracing module, an early warning output module, and a full-process data storage module, which are connected in sequence via communication. The multi-dimensional electrical variable acquisition module has a built-in nanosecond-level timing synchronization calibration unit and a redundant dual-channel acquisition architecture configured for the weak signal channel of no-load leakage current, which is used to complete the synchronous acquisition of electrical variables and environmental parameters under no-load conditions. The reference data management module is used to perform reference parameter calibration for the aircraft power system under no-load conditions, establish and maintain a reference database for normal no-load operation dedicated to no-load operation, support adaptive iterative updates of no-load reference thresholds and coupling compensation coefficients based on no-load operation data of the power system throughout its entire life cycle, no-load aging characteristics of equipment, and changes in environmental parameters, and also support the traceability query of historical reference data. The abnormal feature extraction module is used to perform joint time-domain and frequency-domain analysis on the collected real-time electrical variable data to extract the core abnormal feature parameters under the corresponding no-load conditions. The environmental interference compensation module has a built-in pre-fitted no-load exclusive environment-aging coupling interference compensation model. It is configured with a dynamic coupling correction unit for no-load leakage current. It takes synchronously collected environmental parameters and equipment aging status as input, quantifies the amount of interference of coupling interference on no-load weak electrical variable data and weak characteristic parameters, and performs compensation correction on real-time electrical variable data and abnormal characteristic parameters. The intelligent anomaly identification module has a built-in pre-trained idle-specific anomaly identification model and risk quantification assessment unit, which is used to identify the type of idle anomaly by combining real-time anomaly feature parameters and benchmark database data. The anomaly localization and tracing module has a built-in joint anomaly localization model based on Bayesian probabilistic inference and airborne fault feature matching. Combined with the aircraft power system topology, it calculates the posterior probability of anomalies in each power module and performs anomaly module localization and fault tracing analysis. The warning output module is used to trigger the corresponding level of audible and visual warning signal according to the determined abnormal risk level, and simultaneously push the warning information to the aircraft avionics system and the ground operation and maintenance platform. The full-process data storage module is used to encrypt and store the electrical variable data, feature parameters, anomaly identification results, and traceability information collected throughout the process, and to establish an operation audit log.

[0013] Furthermore, the redundant dual-channel acquisition architecture of the multi-dimensional electrical variable acquisition module has a built-in data fusion and verification unit, which is used to compare the idle weak signal acquisition data of the two sets of sensing units in real time. When the deviation between the two sets of data exceeds the preset limit, it automatically switches to the backup acquisition channel and triggers a channel fault warning. The sampling frequency of the multi-channel synchronous sampling unit supports up to 1MHz, and the synchronization error of the nanosecond-level timing synchronization calibration does not exceed 100ns.

[0014] Furthermore, it also includes an airborne edge computing unit and a ground operation and maintenance platform linkage unit. The airborne edge computing unit is connected to a multi-dimensional electrical variable acquisition module, an intelligent anomaly identification module, an early warning output module, and a full-process data storage module to complete the real-time acquisition, feature extraction, anomaly identification, and local early warning of airborne electrical variable data. The ground operation and maintenance platform receives the full amount of airborne monitoring data uploaded by the system through the airborne communication link, supports centralized monitoring of airborne operation data of power systems of multiple aircraft, statistical analysis of historical airborne fault data, centralized training and optimization updates of airborne anomaly identification models and coupling compensation models, and can generate operation and maintenance plans based on the uploaded anomaly data.

[0015] Compared with existing technologies, the beneficial effects of this invention are: First, it fills the technological gap in the specialized monitoring of aircraft power systems under no-load conditions. This invention constructs a closed-loop monitoring system, encompassing no-load-specific benchmark calibration, multi-channel synchronous data acquisition, joint time-frequency domain feature extraction, environmental and aging coupling interference compensation, intelligent anomaly identification, fault location and tracing, and tiered early warning. Unlike existing technologies that focus on general monitoring logic under loaded conditions, this system is specifically adapted to the unique characteristics of no-load conditions, such as small changes in electrical characteristics and weak fault signals. It systematically solves the core problems of existing technologies, such as insufficient ability to identify early-stage latent faults under no-load conditions and the lack of a comprehensive monitoring system. Secondly, it enhances the ability to capture and identify weak abnormal signals under no-load conditions. This invention eliminates the inherent biases of phase sequence difference and output impedance calculation sensitive to channel delay under no-load conditions through nanosecond-level timing synchronization calibration; for weak signal acquisition channels such as no-load leakage current, a redundant dual-channel acquisition architecture is adopted to ensure the continuity and reliability of weak signal acquisition; combined with a dedicated no-load risk scoring mechanism, the risk weight of weak deviation characteristics under no-load conditions is amplified through nonlinear correction, which can effectively capture early weak abnormal signals under no-load conditions, significantly reducing the probability of misjudgment and missed detection during the monitoring process. Third, it effectively improves the environmental adaptability and life-cycle operational stability of the monitoring solution. This invention constructs an idle-specific environment-aging coupling interference compensation model, which uses independent environmental compensation for conventional electrical variables and sets a dynamic interference coefficient for idle leakage current that adjusts according to the degree of equipment aging. This can quantify and eliminate the masking effect of the coupling effect of environment and equipment aging on weak fault characteristics. At the same time, it supports adaptive iterative updates of idle reference thresholds, which can adapt to the electrical characteristic drift throughout the equipment's life cycle, avoid the decrease in monitoring accuracy caused by reference mismatch, and ensure the stable operation of the system in a wide range of aviation environments and throughout the entire service life of the equipment. Fourth, it achieves accurate classification and targeted fault location for no-load anomalies. Employing an anomaly identification model optimized for no-load samples, it can accurately classify various no-load anomalies, such as voltage regulator anomalies, early generator winding faults, and rectifier open circuits. Combined with a joint localization model integrating Bayesian probabilistic inference and no-load fault feature matching, it corrects the localization bias caused by insufficient no-load fault features, accurately locating the specific module where the anomaly occurs. This provides reliable data support for subsequent operation and maintenance, improving the accuracy and efficiency of power system fault handling. Attached Figure Description

[0016] Figure 1 This is a schematic block diagram of the aircraft power system airborne anomaly monitoring method and early warning system proposed in this invention; Figure 2 Logic diagram of high-precision synchronous acquisition and redundancy verification technology; Figure 3 This is a diagram showing the time-frequency joint analysis and the environmental and aging coupled compensation processing. Figure 4 This is a diagram for intelligent anomaly detection and risk assessment based on the CNN-LSTM architecture. Figure 5 Decision graph for Bayesian probabilistic inference and topological tracing. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0019] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0020] Reference Figures 1 to 5 The method for monitoring no-load anomalies in aircraft power systems includes the following steps: After receiving the operating condition switching signal of the aircraft power system and confirming that the system has entered a stable no-load operating condition, the system performs a no-load-specific reference parameter calibration operation. It collects reference data of electrical variables such as phase voltage, line voltage, no-load leakage current, output frequency, total voltage harmonic distortion, phase sequence phase difference, output impedance, and ripple coefficient under the rated no-load operating condition of the power system, and establishes a reference database for normal no-load operation. Among them, the initial setting of the reference threshold is different from the general threshold of the load condition. It is specifically combined with the full life cycle aging characteristics of the aircraft power system under the no-load condition, historical no-load fault records, and no-load operation data to complete the adaptation setting. Through a multi-channel synchronous acquisition unit, full-parameter synchronous acquisition of the aircraft power system under no-load operation is performed: based on the reference clock signal of the aircraft power system, the sampling timing of each electrical variable acquisition channel is synchronously calibrated at the nanosecond level to eliminate the deviation problem of phase sequence difference and output impedance calculation being sensitive to channel delay under no-load conditions; for weak signal acquisition channels such as no-load leakage current, a redundant dual-channel acquisition architecture is adopted to perform parallel synchronous acquisition of key electrical variables, and the validity of the acquired data is verified by data fusion algorithm to obtain real-time data of electrical variables at the power system output terminal, and the environmental parameters of temperature, humidity and atmospheric pressure of the aircraft power system environment are acquired synchronously; Preprocessing and joint time-domain and frequency-domain analysis are performed on the collected real-time electrical variable data: in the time domain, the sliding window method is used to extract voltage fluctuation characteristics, current abrupt change characteristics, frequency offset characteristics, and impedance change characteristics; in the frequency domain, the signal spectrum is converted by fast Fourier transform to extract harmonic distribution characteristics, thereby obtaining the abnormal characteristic parameter set under the corresponding no-load condition. The characteristic parameter set is then subjected to maximum and minimum normalization processing and invalid interference feature removal. Based on synchronously collected environmental parameters and the current no-load aging state of the equipment, a pre-fitted no-load-specific environment-aging coupling interference compensation model is used to perform correction: independent environmental interference compensation is used for conventional electrical variables, and a special aging-environment coupling correction term is introduced for no-load leakage current. The interference coefficient of temperature on leakage current is dynamically adjusted with the no-load aging degree of the equipment, rather than a fixed coefficient, to quantify the masking effect of coupling interference on weak leakage current characteristics. Compensation correction is performed on the real-time collected electrical variable data and abnormal feature parameter set. The compensation model differs from the general compensation model under load conditions. Its parameters are pre-fitted based on power system no-load calibration data under different aging stages and environments. The real-time anomaly characteristic parameters after compensation and correction are compared with the corresponding benchmark thresholds in the no-load normal operation benchmark database. The no-load-specific risk scoring formula is used to calculate the comprehensive risk score of no-load anomalies. The scoring formula is designed for no-load weak fault characteristics and introduces nonlinear correction index of electrical variables, harmonic weight correction and anomaly duration index correction to amplify the risk weight of weak deviation characteristics. Combined with the pre-trained no-load-specific anomaly identification model, the anomaly type of the power system is identified during no-load operation, and the anomaly risk level is determined based on the risk score. Based on the distribution characteristics of the identified anomaly types and feature parameters, and combined with the topology of the aircraft power system, a joint localization model of Bayesian probabilistic inference and airborne fault feature matching is constructed. The Bayesian posterior probability is corrected by the matching degree between the anomaly features and the module airborne fault feature library to solve the localization deviation caused by insufficient airborne fault features. The posterior probability of anomaly occurrence of each power module is calculated to complete the anomaly module localization and fault source analysis. Based on the determined abnormal risk level, a corresponding warning signal is triggered. Simultaneously, the electrical variable data, characteristic parameters, abnormal identification results, and traceability information collected throughout the entire process are encrypted and stored in a closed loop to complete the full-process abnormal monitoring of the aircraft power system during no-load operation.

[0021] In this invention, the calculation process of the comprehensive score for the abnormal risk of no-load operation is achieved through the formula: ;in, A comprehensive risk assessment of no-load anomalies in the aircraft's power system. The total number of core electrical variable dimensions participating in the evaluation. For the first The weight coefficients of each electrical variable dimension, with the sum of all weight coefficients being 1, are pre-set using the analytic hierarchy process based on the degree of influence of the electrical variable on no-load anomalies, wherein the weight coefficient of no-load leakage current is not less than 0.3. For the first Real-time collected values ​​of individual electrical variables, For the first The reference calibration value of each electrical variable, For the first The maximum permissible limit for each electrical variable For the first The nonlinear correction index for each electrical variable ranges from 1 to 3, with the nonlinear correction index for the no-load leakage current set to 3, based on the fault sensitivity of the electrical variable under no-load conditions. This represents the total number of characteristic frequencies in the harmonic analysis. For the first The influence weight of each characteristic frequency harmonic For the first The difference between the real-time distortion rate of each characteristic frequency harmonic and the reference distortion rate. For the first The maximum permissible distortion difference of each characteristic frequency harmonic. The influence coefficient for the duration of the anomaly. To measure the duration of abnormal characteristic parameters deviating from the benchmark threshold, the nonlinear weighting of multi-dimensional electrical variables, the correction of harmonic distortion effects, and the dynamic gain of abnormal duration are used to achieve refined quantification of no-load abnormal risk. This can accurately distinguish between normal operating condition fluctuations and real abnormalities, significantly reducing the misjudgment rate and missed judgment rate of no-load abnormality monitoring, and providing quantitative support for the accurate determination of abnormal risk levels.

[0022] This invention also includes an adaptive update step for the benchmark threshold under no-load operating conditions. The process involves collecting no-load operating data of the aircraft power system throughout its entire lifecycle, environmental parameter change data, and equipment no-load aging characteristic parameters at fixed intervals. Combined with historical no-load anomaly identification results and feedback data on misjudgments and omissions, the benchmark threshold in the no-load normal operation benchmark database is adaptively and iteratively updated. The dynamic coefficient table of the coupled compensation model is updated synchronously, so that the benchmark and compensation coefficients match the actual no-load operating state of the power system in real time. The updated parameters are then used synchronously for subsequent anomaly feature comparison and risk scoring calculation.

[0023] In this invention, the calculation process of the posterior probability of anomaly occurrence in anomaly module location and fault tracing analysis is performed using the formula: ; For the emergence of abnormal feature sets The anomaly occurred at the time. Power module The posterior probability, This represents the total number of topology modules in the aircraft's power system. For the first When a power module malfunctions, an abnormal feature set appears. The conditional probability is obtained based on historical fault data statistics, in which the weight of the no-load leakage current characteristic is not less than 40%; For the first The prior failure probability of each power module is calculated based on the module's service life, historical failure records, and reliability model. For abnormal feature set With the The number of matching features in the fault feature library of each power module. For abnormal feature set By combining the total number of features with Bayesian probabilistic inference and feature matching degree correction, the module for detecting unloaded anomalies can be accurately located, and the probability of anomalies in each module can be quantified, providing accurate location guidance for subsequent operation and maintenance.

[0024] In this invention, the pre-trained airborne anomaly identification model adopts a serial fusion architecture of multi-scale convolutional neural network and bidirectional long short-term memory network. The front-end multi-scale convolutional neural network extracts the spatial dimension features of airborne anomaly features, and the back-end bidirectional long short-term memory network extracts the temporal dimension features of airborne anomaly features. Finally, the anomaly type classification result is output through a fully connected layer. The model training process uses airborne normal operation data throughout the entire life cycle of the aircraft power system, simulation data of various airborne anomaly faults, and historical airborne fault measured data. In addition, leakage current samples from different aging stages and different environments are added to construct the training dataset. The cross-entropy loss function is used as the optimization objective, and the model weights are iteratively updated through an adaptive moment estimation optimization algorithm. After training, the model can identify multiple types of early weak airborne faults, including voltage regulator anomalies, early generator winding faults, rectifier open circuits, line loose connections, and harmonic anomalies. At the same time, incremental learning optimization is performed on the model based on the newly added measured airborne anomaly data.

[0025] In this invention, the fitting process of the aging-environment coupling correction term is as follows: the no-load aging stage of the equipment is divided into several intervals, and for each aging interval, the interference coefficients of temperature and humidity on the no-load leakage current are fitted to form a dynamic coupling coefficient table; during the compensation process, according to the current aging state of the equipment, the interference coefficients of the corresponding aging interval are called to perform compensation to eliminate the coupling interference between aging and the environment.

[0026] In this invention, the verification and switching rules of the redundant dual-channel acquisition architecture are as follows: two sets of sensing units of the same precision are deployed in parallel on the no-load leakage current weak signal acquisition channel. The two sets of units output the acquired data synchronously. The deviation value of the two sets of data is calculated by the data fusion algorithm. When the deviation value exceeds the preset limit, the system automatically switches to the backup acquisition channel and triggers a channel fault warning.

[0027] This invention includes the following modules: The multi-dimensional electrical variable acquisition module consists of a multi-channel synchronous sampling unit, a voltage sensing unit, a current sensing unit, a frequency detection unit, a harmonic analysis unit, an impedance measurement unit, and an environmental parameter acquisition unit, and is deployed at key nodes of the aircraft power system. The multi-channel synchronous sampling unit has a built-in nanosecond-level timing synchronization calibration module, which performs nanosecond-level synchronous calibration on each acquisition channel based on the reference clock signal of the aircraft power system to eliminate phase sequence difference and channel delay deviation in output impedance calculation under no-load conditions. The key no-load weak signal acquisition channel adopts a redundant dual-channel acquisition architecture, with two sets of sensing units of the same precision deployed in parallel to complete the synchronous acquisition, validity verification, and preliminary preprocessing of electrical variable data and environmental parameters under no-load conditions. The benchmark data management module is used to perform benchmark parameter calibration for the aircraft power system under no-load conditions, establish and maintain a benchmark database for no-load operation, and support adaptive iterative updates of no-load benchmark thresholds and coupling compensation coefficients based on no-load operation data throughout the power system's life cycle, equipment no-load aging characteristics, and changes in environmental parameters. It also supports the traceability and query of historical benchmark data. The abnormal feature extraction module has a built-in time-domain and frequency-domain joint analysis unit. In the time domain, it uses the sliding window method to extract voltage fluctuations, current abrupt changes, frequency shifts, and impedance changes. In the frequency domain, it uses fast Fourier transform to extract harmonic distribution features. It is used to perform feature extraction, normalization processing, and invalid feature removal on the collected real-time electrical variable data. The environmental interference compensation module has a built-in pre-fitted no-load exclusive environment-aging coupling interference compensation model. It is configured with a dynamic coupling correction unit for no-load leakage current. It takes synchronously collected environmental parameters and equipment aging status as input, quantifies the amount of interference of coupling interference on no-load weak electrical variable data and weak characteristic parameters, and performs compensation correction on real-time electrical variable data and abnormal characteristic parameters. The intelligent anomaly identification module has a built-in pre-trained idle-specific anomaly identification model and risk quantification assessment unit. It is used to combine the real-time anomaly feature parameters after compensation and correction and the idle normal operation benchmark database data to calculate the comprehensive score of idle anomaly risk using the idle-specific risk scoring formula, identify the type of idle anomaly and determine the anomaly risk level. The anomaly localization and tracing module has a built-in joint anomaly localization model based on Bayesian probabilistic inference and airborne fault feature matching. It is used to calculate the posterior probability of anomaly occurrence of each power module based on the identified anomaly type and feature parameter distribution characteristics, combined with the aircraft power system topology, and to complete the anomaly module localization and fault tracing analysis. The early warning output module is used to trigger the corresponding level of audible and visual early warning signal based on the determined abnormal risk level, and simultaneously push the early warning information to the aircraft avionics system and the ground operation and maintenance platform. The end-to-end data storage module is used to encrypt and store electrical variable data, characteristic parameters, anomaly identification results, and traceability information collected throughout the entire process, and to establish an operation audit log.

[0028] In this invention, the redundant dual-channel acquisition architecture of the multi-dimensional electrical variable acquisition module has a built-in data fusion and verification unit for real-time comparison of the idle weak signal acquisition data of the two sets of sensing units. When the deviation between the two sets of data exceeds the preset limit, it automatically switches to the backup acquisition channel and triggers a channel fault warning. The sampling frequency of the multi-channel synchronous sampling unit supports up to 1MHz, and the synchronization error of the nanosecond-level timing synchronization calibration does not exceed 100ns.

[0029] This invention also includes an airborne edge computing unit and a ground operation and maintenance platform linkage unit. The airborne edge computing unit is communicatively connected to a multi-dimensional electrical variable acquisition module, an intelligent anomaly identification module, an early warning output module, and a full-process data storage module to complete real-time acquisition, feature extraction, anomaly identification, and local early warning of airborne electrical variable data. The ground operation and maintenance platform receives the full amount of airborne monitoring data uploaded by the system through the airborne communication link, supports centralized monitoring of airborne operation data of multiple aircraft power systems, statistical analysis of historical airborne fault data, centralized training and optimization updates of airborne anomaly identification models and coupling compensation models, and can generate operation and maintenance plans based on the uploaded anomaly data.

[0030] The following two examples further illustrate the specific implementation of this system: Example 1: Implementation of No-Load Anomaly Monitoring for Variable Frequency AC Power Supply Systems in Civil Aircraft. This example implements a method for monitoring no-load anomalies in aircraft power supply systems. First, the operating condition switching signal of the aircraft power supply system is received. After confirming that the system has entered a stable no-load operating condition, a calibration operation for no-load specific reference parameters is performed. Reference data of phase voltage, line voltage, no-load leakage current, output frequency, total harmonic distortion, phase sequence phase difference, output impedance, and ripple coefficient under the rated no-load operating condition of the power supply system are collected to establish a no-load normal operation reference database. The initial setting of the reference threshold differs from the general threshold for load conditions. It is specifically designed to adapt to the full life cycle aging characteristics of the aircraft power supply system under no-load conditions, historical no-load fault records, and no-load operating data. The reference threshold and weighting of the no-load leakage current are separately adapted to the requirements for identifying weak no-load faults.

[0031] Subsequently, the multi-channel synchronous acquisition unit of the multi-dimensional electrical variable acquisition module performs full-parameter synchronous acquisition of the aircraft power system under no-load operation. Based on the reference clock signal of the aircraft power system, the sampling timing of each electrical variable acquisition channel is synchronously calibrated at the nanosecond level, with the synchronization error controlled within 100ns, eliminating the deviation problems of phase sequence difference and output impedance calculation being sensitive to channel delay under no-load conditions. For weak signal acquisition channels such as no-load leakage current, a redundant dual-channel acquisition architecture is adopted to perform parallel synchronous acquisition of key electrical variables. Two sets of sensors with the same precision synchronously output the acquired data, and the validity of the acquired data is verified by a data fusion algorithm. When the deviation between the two sets of data exceeds the preset limit, the system automatically switches to the backup acquisition channel and triggers a channel fault warning. During the acquisition process, real-time data of various electrical variables at the power system output are acquired, and environmental parameters such as temperature, humidity, and atmospheric pressure of the aircraft power system's environment are acquired synchronously. The sampling frequency supports up to 1MHz, adapting to the high-precision acquisition requirements of weak signals under no-load conditions.

[0032] After data acquisition, the real-time electrical variable data is preprocessed and subjected to joint time-domain and frequency-domain analysis using the anomaly feature extraction module. In the time domain, a sliding window method is used to extract voltage fluctuation characteristics, current abrupt changes, frequency shift characteristics, and impedance change characteristics. In the frequency domain, a fast Fourier transform is used to convert the signal spectrum and extract harmonic distribution characteristics, resulting in a set of anomaly feature parameters corresponding to the no-load operating condition. This set of feature parameters is then subjected to max-min normalization and invalid interference feature removal, filtering out conventional electromagnetic interference features from the aircraft's avionics system.

[0033] Subsequently, based on synchronously collected environmental parameters and the current no-load aging state of the equipment, a pre-fitted no-load-specific environment-aging coupling interference compensation model is used for correction. Independent environmental interference compensation is applied to conventional electrical variables, while a dedicated aging-environment coupling correction term is introduced for no-load leakage current. The interference coefficient of temperature on leakage current is dynamically adjusted according to the no-load aging degree of the equipment, rather than being a fixed coefficient. This quantifies the masking effect of coupling interference on weak leakage current characteristics and performs compensation correction on the real-time collected electrical variable data and abnormal characteristic parameter set. The compensation model differs from the general compensation model for load conditions. Its parameters are pre-fitted based on no-load calibration data of the power system under different aging stages and environments. The no-load aging stage of the equipment is divided into multiple intervals, and the interference coefficients of temperature and humidity on no-load leakage current are fitted for each aging interval, forming a dynamic coupling coefficient table. During the compensation process, the interference coefficients of the corresponding interval are called according to the current aging state of the equipment to perform compensation.

[0034] After compensation and correction, the intelligent anomaly identification module compares the real-time anomaly feature parameters after compensation and correction with the corresponding benchmark thresholds in the no-load normal operation benchmark database. A risk scoring formula specific to no-load operation is used to calculate the comprehensive risk score for no-load anomalies. This formula, targeting weak no-load fault characteristics, introduces nonlinear correction exponents for electrical variables, harmonic weight correction, and anomaly duration exponent correction to amplify the risk weights of weak deviation features. Combined with a pre-trained no-load specific anomaly identification model, the model identifies the types of anomalies in the power system's no-load operation. The model employs a serial fusion architecture of a multi-scale convolutional neural network and a bidirectional long short-term memory network. The front-end multi-scale convolutional neural network extracts the spatial dimension features of no-load anomaly features, while the back-end bidirectional long short-term memory network extracts the temporal series dimension features. Finally, the fully connected layer outputs the anomaly type classification results, which can identify multiple types of early weak no-load faults, including voltage regulator anomalies, early generator winding faults, rectifier open circuits, line loose connections, and harmonic anomalies. The risk level of anomalies is determined based on risk scoring. At the same time, the full life cycle of the aircraft power system is collected at fixed intervals for no-load operation data. Combined with historical no-load anomaly identification results, the baseline threshold is adaptively iteratively updated, and the dynamic coefficient table of the coupled compensation model is updated synchronously.

[0035] Subsequently, through the anomaly localization and tracing module, based on the distribution characteristics of the identified anomaly types and feature parameters, and combined with the topology of the aircraft power system, a joint localization model of Bayesian probabilistic inference and airborne fault feature matching is constructed. The Bayesian posterior probability is corrected by the matching degree between anomaly features and the module airborne fault feature database, addressing the localization bias caused by insufficient airborne fault features. The posterior probability of anomaly occurrence for each power module is calculated, completing the anomaly module localization and fault tracing analysis. Finally, based on the determined anomaly risk level, a corresponding warning signal is triggered. Simultaneously, the electrical variable data, feature parameters, anomaly identification results, and tracing information collected throughout the entire process are encrypted and stored in a closed loop, completing the full-process anomaly monitoring of the aircraft power system during airborne operation.

[0036] Table 1. Performance Comparison of This Method and Traditional Monitoring Methods for Civil Aircraft Power Systems Performance indicators Method of the present invention Traditional monitoring methods Early-stage weak fault identification rate under no-load conditions high Low Anomaly module location accuracy high middle False alarm rate of no-load condition monitoring Low high No-load condition monitoring false alarm rate Low high Fault warning lead time long short Environmental and aging interference compensation effect excellent Difference Table 1 visually demonstrates the comprehensive advantages of this method in the no-load monitoring scenario of AC power systems for civil aircraft. Traditional monitoring methods often use general thresholds and models for load conditions without specific adaptation to no-load conditions. This results in insufficient ability to identify early-stage weak faults, high false alarm and false negative rates, and failure to consider the coupling interference of environment and aging, making early fault warning and accurate fault location impossible. This method, through no-load-specific benchmark calibration, synchronous data acquisition, coupling compensation, and intelligent identification architecture, can accurately identify early-stage weak faults, significantly improve anomaly location accuracy, reduce the risk of false alarms and false negatives, achieve early fault warning, and fully adapt to the high-safety-level monitoring requirements of civil aviation power systems.

[0037] Example 2: Implementation of Idle Anomaly Monitoring of High-Voltage DC Power Supply System for Reconnaissance and Strike Integrated UAVs. This example implements an idle anomaly monitoring method for aircraft power supply systems, and deploys an idle anomaly early warning system for aircraft power supply systems. The system includes a multi-dimensional electrical variable acquisition module, a benchmark data management module, an anomaly feature extraction module, an environmental interference compensation module, an intelligent anomaly identification module, an anomaly location and tracing module, an early warning output module, and a full-process data storage module. It also includes an airborne edge computing unit and a ground-based operation and maintenance platform linkage unit. First, the multi-dimensional electrical variable acquisition module receives the operating condition switching signal of the UAV power supply system. After confirming that the system has entered a stable idle operating condition, the benchmark data management module performs an idle-specific benchmark parameter calibration operation, collecting benchmark data of electrical variables such as bus voltage, idle leakage current, output ripple, total harmonic distortion rate, line voltage drop, output impedance, and bus voltage fluctuation rate under the rated idle operating condition of the power supply system. An idle normal operation benchmark database is established. The benchmark thresholds are specifically designed to adapt to the aging characteristics and historical fault records of the UAV power supply system under idle operating conditions, differing from the general thresholds for load conditions.

[0038] Subsequently, the multi-channel synchronous sampling unit of the multi-dimensional electrical variable acquisition module performs full-parameter synchronous acquisition of the UAV power system under no-load operation. Based on the reference clock signal of the UAV power system, the sampling timing of each electrical variable acquisition channel is synchronously calibrated at the nanosecond level, with the synchronization error controlled within 100ns, eliminating channel delay deviations in phase difference and output impedance calculation under no-load conditions. For weak signal acquisition channels such as no-load leakage current, a redundant dual-channel acquisition architecture is adopted, with two sets of sensing units of the same precision deployed in parallel to perform parallel synchronous acquisition. The two sets of acquired data are compared in real time through a data fusion and verification unit. When the deviation value exceeds the preset limit, the system automatically switches to the backup acquisition channel and triggers a channel fault warning. The temperature, humidity, and atmospheric pressure environmental parameters of the high-altitude environment where the UAV power system is located are also acquired synchronously.

[0039] After data acquisition, the real-time electrical variable data is preprocessed and subjected to joint time-domain and frequency-domain analysis using the anomaly feature extraction module. In the time domain, a sliding window method is used to extract voltage fluctuations, current abrupt changes, ripple variations, and impedance changes. In the frequency domain, a fast Fourier transform is used to convert the signal spectrum and extract harmonic distribution features, resulting in the corresponding set of anomaly feature parameters under no-load conditions. Normalization and invalid interference feature removal are then performed. The environmental interference compensation module utilizes a built-in no-load-specific environment-aging coupling interference compensation model for correction. An aging-environment coupling correction term is introduced for no-load leakage current, and the temperature interference coefficient on the leakage current is dynamically adjusted according to the no-load aging degree of the equipment. Compensation and correction are applied to the real-time electrical variable data and the set of anomaly feature parameters. The parameters of the compensation model are pre-fitted based on no-load calibration data from different aging stages and high-altitude environments of the UAV power system.

[0040] Subsequently, the intelligent anomaly identification module compares the compensated and corrected real-time anomaly feature parameters with the benchmark thresholds of the no-load normal operation benchmark database. A comprehensive no-load anomaly risk score is calculated using a no-load-specific risk scoring formula, amplifying the risk weights of weak no-load deviation features. This is combined with a pre-trained no-load-specific anomaly identification model to identify anomaly types. The model employs a serial fusion architecture of a multi-scale convolutional neural network and a bidirectional long short-term memory network. Based on the entire lifecycle no-load operation data and fault simulation data of the UAV power system, it completes training and incremental learning optimization. It can identify multiple no-load anomalies, such as voltage regulation unit anomalies, early generator winding faults, rectifier diode open circuits, line loose connections, and no-load leakage of the energy storage battery, while simultaneously determining the anomaly risk level. The benchmark data management module collects full lifecycle no-load operation data at fixed intervals and performs adaptive iterative updates to the benchmark thresholds and coupling compensation coefficients, ensuring real-time matching between the benchmark, compensation coefficients, and the actual operating state of the power system.

[0041] Subsequently, through the anomaly localization and tracing module, combined with the topology of the UAV power system, a joint localization model of Bayesian probabilistic inference and airborne fault feature matching is constructed to calculate the posterior probability of anomalies in each power module, completing the anomaly module localization and fault tracing analysis. The early warning output module triggers corresponding early warning signals based on the determined anomaly risk level, simultaneously pushing the early warning information to the UAV flight control system and the ground maintenance platform. The full-process data storage module performs encrypted storage of all process data and establishes an operation audit log. The airborne edge computing unit completes real-time acquisition, feature extraction, anomaly identification, and local early warning of airborne electrical variable data. The ground maintenance platform receives all monitoring data through the airborne communication link, supporting centralized monitoring of airborne operation data of multiple UAV power systems, historical data statistical analysis, centralized model training and optimization updates, and generating maintenance and repair plans based on anomaly data.

[0042] Performance indicators Method of the present invention Traditional monitoring methods Monitoring accuracy in complex high-altitude environments high Low Early leakage fault identification capability under no-load conditions excellent Difference Airborne terminal real-time monitoring of response speed quick slow Multi-flight centralized operation and maintenance adaptability excellent Difference Abnormal positioning accuracy under no-load conditions high Low Full lifecycle monitoring adaptability excellent middle Table 2 clearly demonstrates the comprehensive advantages of this method in the airborne monitoring scenario of UAV high-voltage DC power supply systems. Traditional monitoring methods are not adapted to the complex high-altitude environment of UAVs, lack the ability to identify early-stage airborne leakage faults, have slow real-time response speeds at the airborne end, and cannot meet the centralized operation and maintenance needs of multiple UAVs. Furthermore, the monitoring accuracy declines rapidly throughout the entire lifecycle as the equipment ages. This method adapts to the complex high-altitude environment through an airborne-specific coupling compensation model, achieves rapid response by combining airborne edge computing, and adapts to the aging characteristics of the equipment throughout its lifecycle through baseline adaptive updates. It can accurately identify early-stage airborne faults and complete precise location, while supporting centralized operation and maintenance management of multiple UAVs, fully meeting the long-endurance and high-reliability monitoring needs of UAVs.

[0043] refer to Figure 1 This diagram illustrates the complete operational logic of anomaly monitoring in an aircraft power system under no-load conditions. The process begins with real-time identification of the power system's operating condition. After confirming a stable no-load state, the system first performs calibration of dedicated baseline parameters and database construction. Subsequently, it acquires multi-dimensional raw data, including voltage, current, and environmental parameters, through a high-precision synchronous acquisition unit. In the core processing stage, the system sequentially performs time-frequency feature extraction and environmental and aging coupling interference compensation to eliminate background noise. The corrected feature parameters are then fed into an intelligent identification model for risk scoring and level determination. If an anomaly is identified, the system performs Bayesian probabilistic inference based on the topology to accurately locate the fault point. Finally, based on the risk level, an early warning is triggered, and encrypted storage of all process data is executed, forming a complete closed loop from state perception to decision output.

[0044] Reference Figure 2 This diagram details the technical aspects of the system at the data acquisition layer, focusing on addressing the susceptibility of weak signals to interference and phase sequence deviation under no-load conditions. The system utilizes a reference clock signal to perform nanosecond-level timing synchronization calibration on all electrical variable acquisition channels, ensuring the accuracy of output impedance and phase difference calculations. For weak signals such as no-load leakage current, which are highly susceptible to electromagnetic interference, the system employs a special redundant dual-channel parallel acquisition architecture. Two sets of sensors with equal precision synchronously output data, and real-time deviation verification is performed using a built-in data fusion algorithm. If the deviation between the two sets of data exceeds the limit, the system automatically switches to the backup channel and triggers a channel fault warning, ensuring high reliability and high fidelity of the data entering subsequent analysis stages. This hardware-level synchronization and redundancy mechanism lays a solid foundation for the extraction of weak fault characteristics.

[0045] Reference Figure 3This figure illustrates the deep transformation process from the original signal to standardized feature parameters. In the feature extraction stage, the system employs a parallel analysis strategy in both the time and frequency domains: in the time domain, a sliding window is used to capture transient features such as voltage fluctuations and current surges; in the frequency domain, a Fast Fourier Transform is used to extract the distortion distribution of each harmonic. Because the no-load leakage current is extremely sensitive to the environment and dynamically changes with equipment aging, the system introduces a crucial "aging-environment coupling compensation" step. This step does not use fixed coefficients but instead calls a pre-fitted dynamic coupling coefficient table based on the current aging stage of the equipment, quantifying and offsetting the masking effect of temperature and humidity on weak signals. Through this multi-dimensional compensation correction and feature normalization processing, the system can eliminate invalid interference components, significantly improve the identification of early-stage weak fault characteristics under no-load conditions, and avoid false alarms caused by environmental changes.

[0046] Reference Figure 4 This diagram illustrates the internal architecture and decision-making logic of the intelligent anomaly identification module. This module employs a serial fusion architecture of a multi-scale convolutional neural network and a bidirectional long short-term memory network, aiming to simultaneously capture the deep features of no-load anomalies in both spatial and temporal dimensions. During the identification process, the system simultaneously runs a dedicated risk quantification and evaluation engine, amplifying the deviation weights of key indicators such as leakage current through a nonlinear correction exponent, and calculating a comprehensive risk score by combining it with a duration exponent. This method, combining quantitative scoring with an AI classification model, can not only accurately identify various weak fault types such as early generator winding faults and rectifier open circuits, but also classify risks into different levels based on the scoring results. This multi-dimensional judgment mode solves the pain point of insufficient sensitivity of the traditional single threshold method under no-load conditions.

[0047] refer to Figure 5 This figure illustrates the localization and feedback logic in the final stage of the anomaly monitoring system. To address the localization bias caused by the relative scarcity of unloaded fault features, the system constructs a joint localization model. This model first acquires the aircraft's power topology and the historical prior fault probabilities of each module. When the intelligent identification module detects an anomaly feature set, the system corrects the Bayesian posterior probability by calculating the matching degree between the anomaly features and the unloaded fault feature database of each module. By calculating and ranking the probability of anomalies occurring in each power module in real time, the system can accurately pinpoint the source of the fault and output a source tracing analysis report. Finally, the system performs encrypted data storage and incremental learning, feeding the measured data back to the benchmark database for adaptive updates, ensuring that monitoring performance continuously optimizes as the equipment's service life increases.

[0048] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring no-load anomalies in an aircraft power system, characterized in that, Includes the following steps: After receiving the operating condition switching signal of the aircraft power system and entering the stable no-load operating condition, the system performs the no-load exclusive reference parameter calibration operation, collects the phase voltage, line voltage, no-load leakage current, output frequency, total harmonic distortion, phase sequence phase difference, output impedance, and ripple coefficient electrical variable reference data under the rated no-load operating condition, and establishes a reference database for normal no-load operation. The baseline threshold differs from the general threshold under load conditions and is adapted and set by combining the aging characteristics of the entire life cycle under no-load conditions and historical no-load fault data. By using a multi-channel synchronous acquisition unit, nanosecond-level synchronous calibration is performed on each acquisition channel based on the system reference clock to eliminate phase sequence difference under no-load conditions and the time delay-sensitive deviation of impedance calculation; redundant dual-channel parallel acquisition is used for the weak signal channel of no-load leakage current to verify the validity of the data and synchronously acquire environmental temperature, humidity and atmospheric pressure parameters. The collected real-time electrical variable data are subjected to joint time-domain and frequency-domain analysis. In the time domain, the sliding window method is used to extract the characteristics of voltage fluctuation, current change, frequency shift and impedance change. In the frequency domain, the harmonic distribution characteristics are extracted by fast Fourier transform to obtain the abnormal feature parameter set, and normalization processing and invalid feature removal are completed. Based on environmental parameters and the no-load aging state of the equipment, a no-load-specific environment-aging coupling interference compensation model is adopted for correction: conventional electrical variables are compensated independently by the environment, the no-load leakage current is introduced into the aging-environment coupling correction term, the temperature interference coefficient is dynamically adjusted according to the no-load aging degree of the equipment, the masking effect of coupling interference on the weak leakage current characteristics is quantified, and the compensation and correction of electrical variables and characteristic parameters are completed. The corrected feature parameters are compared with the benchmark threshold, and the comprehensive score of abnormal risk is calculated using the vacant load-specific risk scoring formula, thus amplifying the risk weight of the vacant load weak deviation feature. By combining a pre-trained, unloaded, dedicated anomaly recognition model, the anomaly type is identified and the risk level is determined; Based on the system topology, a joint localization model combining Bayesian probabilistic inference and no-load fault feature matching is adopted. By correcting the posterior probability through feature matching degree, the localization deviation caused by insufficient no-load fault features is resolved, and the abnormal module localization and fault source analysis are completed. Based on the determined level of abnormal risk, a corresponding warning signal is triggered, and the entire stream of data is simultaneously encrypted and stored in a closed loop.

2. The method for monitoring no-load anomalies in an aircraft power system according to claim 1, characterized in that, The formula for calculating the comprehensive score of abnormal risk is as follows: ;in, A comprehensive risk assessment of no-load anomalies in the aircraft's power system. The total number of core electrical variable dimensions participating in the evaluation. For the first Weighting coefficients for each electrical variable dimension. For the first Real-time collected values ​​of individual electrical variables, For the first The reference calibration value of each electrical variable, For the first The maximum permissible limit for each electrical variable For the first Nonlinear correction exponent for individual electrical variables, This represents the total number of characteristic frequencies in the harmonic analysis. For the first The influence weight of each characteristic frequency harmonic For the first The difference between the real-time distortion rate of each characteristic frequency harmonic and the reference distortion rate. For the first The maximum permissible distortion difference of each characteristic frequency harmonic. The influence coefficient is the duration of the anomaly. This refers to the duration during which abnormal feature parameters deviate from the baseline threshold.

3. The method for monitoring no-load anomalies in an aircraft power system according to claim 1, characterized in that, It also includes an adaptive update step for the benchmark threshold under no-load conditions. The system collects no-load operation data, environmental parameters, and no-load aging characteristic parameters throughout the entire life cycle at fixed intervals. Combined with historical anomaly identification results and feedback on misjudgments and omissions, the benchmark threshold is adaptively and iteratively updated. The dynamic coefficient table of the coupled compensation model is updated synchronously. The updated parameters are used for subsequent anomaly monitoring.

4. The method for monitoring no-load anomalies in an aircraft power system according to claim 1, characterized in that, The formula for calculating the posterior probability of anomaly occurrence in the anomaly module location and fault tracing analysis is as follows: ; For the emergence of abnormal feature sets The anomaly occurred at the time. Power module The posterior probability, This represents the total number of topology modules in the aircraft's power system. For the first When a power module malfunctions, an abnormal feature set appears. The conditional probability, For the first Prior fault probability of each power module For abnormal feature set With the The number of matching features in the fault feature library of each power module. For abnormal feature set The total number of features.

5. The method for monitoring no-load anomalies in an aircraft power system according to claim 1, characterized in that, The no-load-specific anomaly identification model adopts a serial fusion architecture of multi-scale convolutional neural network and bidirectional long short-term memory network. The front end extracts spatial dimension features, and the back end extracts time series dimension features. The anomaly classification results are output through a fully connected layer. The model uses full-lifecycle no-load normal data, no-load anomaly simulation data, and historical no-load measured data. In addition, leakage current samples under different aging stages and environments are added to construct the training set. The model is trained by an adaptive moment estimation algorithm with cross-entropy loss as the optimization objective, and supports incremental learning optimization.

6. The method for monitoring no-load anomalies in an aircraft power system according to claim 1, characterized in that, The fitting process of the aging-environment coupling correction term is as follows: the no-load aging stage of the equipment is divided into several intervals, and the interference coefficient of temperature and humidity on the no-load leakage current is fitted for each interval to form a dynamic coupling coefficient table; during compensation, the interference coefficient of the corresponding interval is called to perform compensation according to the current aging state of the equipment.

7. The method for monitoring no-load anomalies in an aircraft power system according to claim 1, characterized in that, The verification and switching rules of the redundant dual-channel acquisition architecture are as follows: the two sets of sensing units with the same precision in the no-load leakage current acquisition channel output data synchronously, calculate the deviation value of the two sets of data, and when the deviation exceeds the preset limit, automatically switch to the backup channel and trigger the channel fault warning.

8. An aircraft power system in-flight anomaly early warning system according to any one of claims 1-7, characterized in that, It includes a multi-dimensional electrical variable acquisition module, a baseline data management module, an anomaly feature extraction module, an environmental interference compensation module, an intelligent anomaly identification module, an anomaly location and source tracing module, an early warning output module, and a full-process data storage module, which are connected in sequence via communication. The multi-dimensional electrical variable acquisition module has a built-in nanosecond-level timing synchronization calibration unit and a redundant dual-channel acquisition architecture configured for the weak signal channel of no-load leakage current, which is used to complete the synchronous acquisition of electrical variables and environmental parameters under no-load conditions. The reference data management module is used to perform reference parameter calibration for the aircraft power system under no-load conditions, establish and maintain a reference database for normal no-load operation dedicated to no-load operation, support adaptive iterative updates of no-load reference thresholds and coupling compensation coefficients based on no-load operation data of the power system throughout its entire life cycle, no-load aging characteristics of equipment, and changes in environmental parameters, and also support the traceability query of historical reference data. The abnormal feature extraction module is used to perform joint time-domain and frequency-domain analysis on the collected real-time electrical variable data to extract the core abnormal feature parameters under the corresponding no-load conditions. The environmental interference compensation module has a built-in pre-fitted no-load exclusive environment-aging coupling interference compensation model. It is configured with a dynamic coupling correction unit for no-load leakage current. It takes synchronously collected environmental parameters and equipment aging status as input, quantifies the amount of interference of coupling interference on no-load weak electrical variable data and weak characteristic parameters, and performs compensation correction on real-time electrical variable data and abnormal characteristic parameters. The intelligent anomaly identification module has a built-in pre-trained idle-specific anomaly identification model and risk quantification assessment unit, which is used to identify the type of idle anomaly by combining real-time anomaly feature parameters and benchmark database data. The anomaly localization and tracing module has a built-in joint anomaly localization model based on Bayesian probabilistic inference and airborne fault feature matching. Combined with the aircraft power system topology, it calculates the posterior probability of anomalies in each power module and performs anomaly module localization and fault tracing analysis. The warning output module is used to trigger the corresponding level of audible and visual warning signal according to the determined abnormal risk level, and simultaneously push the warning information to the aircraft avionics system and the ground operation and maintenance platform. The full-process data storage module is used to encrypt and store the electrical variable data, feature parameters, anomaly identification results, and traceability information collected throughout the process, and to establish an operation audit log.

9. The aircraft power system no-load anomaly early warning system according to claim 8, characterized in that, The redundant dual-channel acquisition architecture of the multi-dimensional electrical variable acquisition module has a built-in data fusion and verification unit, which is used to compare the idle weak signal acquisition data of the two sets of sensing units in real time. When the deviation between the two sets of data exceeds the preset limit, it automatically switches to the backup acquisition channel and triggers a channel fault warning. The multi-channel synchronous sampling unit supports a sampling frequency of up to 1MHz, and the synchronization error of the nanosecond-level timing synchronization calibration does not exceed 100ns.

10. The aircraft power system in-flight anomaly early warning system according to claim 8, characterized in that, It also includes an airborne edge computing unit and a ground operation and maintenance platform linkage unit. The airborne edge computing unit is connected to a multi-dimensional electrical variable acquisition module, an intelligent anomaly identification module, an early warning output module, and a full-process data storage module to complete the real-time acquisition, feature extraction, anomaly identification, and local early warning of airborne electrical variable data. The ground operation and maintenance platform receives all the airborne monitoring data uploaded by the system through the airborne communication link. It supports centralized monitoring of airborne operation data of power systems of multiple aircraft, statistical analysis of historical airborne fault data, centralized training and optimization of airborne anomaly identification model and coupling compensation model, and can generate operation and maintenance plans based on the uploaded anomaly data.

Images