Method and system for predicting insulation deterioration trend of lightweight high-voltage high-power aviation motor

Abstract

This invention belongs to the field of aviation electrical equipment condition monitoring and life prediction technology, specifically relating to a lightweight method and system for predicting the insulation degradation trend of high-voltage, high-power aviation motors. The method includes: constructing a quasi-steady-state segment using flight parameter stage switching tags and calibrating the blind zone of insulation degradation trend; filling the blind zone of insulation degradation trend using weighted trend extrapolation based on dynamic confidence weights and forward-reverse multi-layer consistency verification; and asymptotically correcting multi-physics coupling degradation by constructing a lightweight physical surrogate function, ultimately generating a remaining life prediction interval with confidence intervals. This invention achieves highly robust insulation degradation trend prediction under intermittent missing multi-source signals and complex operating conditions, significantly reducing computational complexity while ensuring physical consistency, and providing accurate decision-making basis for condition-based maintenance and flight safety assurance of high-voltage, high-power aviation motor insulation systems.

Description

Technical Field

[0001] This invention belongs to the field of aviation electrical equipment condition monitoring and life prediction technology, specifically relating to a method and system for predicting the insulation degradation trend of lightweight high-voltage high-power aviation motors. Background Technology

[0002] With the rapid development of multi-electric / all-electric aircraft technology, high-voltage, high-power aero-motors have become core power actuators for flight control, environmental control, and propulsion systems. These motors operate under a complex stress environment of thin air at high altitudes, intense temperature fluctuations, strong mechanical vibrations, and high-frequency pulse voltage impacts, making their stator winding insulation systems highly susceptible to irreversible degradation. Insulation failure is one of the main causes of catastrophic failures in aero-motors, directly threatening flight safety. Therefore, accurately predicting insulation degradation trends and remaining service life is crucial for achieving condition-based maintenance and ensuring flight safety. However, the real-time operating conditions of aero-motors are extremely complex and variable, with frequent flight phase transitions, leading to strong non-stationarity and transient fluctuations in multi-source characteristic signals reflecting insulation status, such as partial discharge, vibration spectrum, and hotspot temperature. Simultaneously, airborne sensors often experience intermittent operation, data loss, or drift in harsh airborne environments, inevitably creating large blind spots in the continuous condition monitoring data stream regarding insulation degradation trends. Traditional prediction methods based on ideal continuous sequences suffer from significant accuracy degradation or even model failure in these insulation degradation trend blind spots.

[0003] Existing insulation lifetime prediction technologies mostly rely on threshold alarms for a single parameter or simplified extrapolations based on the Arrhenius thermal lifetime formula, making it difficult to fully reflect the microscopic nature of insulation degradation under the coupling effects of multiple physics. While a few multi-parameter fusion methods offer richer feature extraction dimensions, they typically employ simple interpolation strategies to fill in the blind spots of insulation degradation trends, ignoring the nonlinear abrupt changes in degradation characteristics within these blind spots caused by the dynamic coupling of electro-thermal-mechanical factors. This results in poor smoothness and insufficient physical consistency in the prediction results at the boundaries of the insulation degradation trend blind spots. Furthermore, while existing high-fidelity finite element and other multiphysics simulations can accurately calculate the insulation degradation process, their computational load is enormous and cannot meet the requirements of airborne edge computing platforms for lightweight real-time inference. Summary of the Invention

[0004] To overcome the problems of low prediction accuracy of insulation degradation trend blind spots in existing technologies, insufficient integration of multi-physics field coupling degradation mechanisms and data-driven approaches, and limited airborne computing resources, this invention provides a lightweight method and system for predicting insulation degradation trends in high-voltage, high-power aircraft motors. It achieves highly robust insulation degradation trend prediction under intermittent missing multi-source signals and complex operating conditions, significantly reducing computational complexity while ensuring physical consistency. This provides accurate decision-making basis for condition-based maintenance and flight safety assurance of high-voltage, high-power aircraft motor insulation systems.

[0005] The technical solution of this application specifically includes: According to one aspect of this application, a method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors is provided, comprising: By using the combined criteria of flight parameter stage switching tags and operating condition change rate, the synchronously acquired multi-source signal streams are segmented into quasi-steady-state operation segments and the blind zone of insulation degradation trend is identified. For each blind zone of insulation degradation trend, multidimensional degradation feature sequences are extracted from the forward and backward quasi-steady-state segments. Dynamic reliability weights are calculated based on the physical consistency criterion, and complete forward boundary degradation feature sequences and backward boundary degradation feature sequences are generated. A mapping relationship between the multidimensional degradation feature sequences and the overall insulation degradation index is established. Starting from the forward and backward boundary degradation feature sequences, we perform forward and backward weighted trend recursion, and generate a preliminary multidimensional degradation feature prediction sequence and a preliminary overall insulation degradation index within the insulation degradation trend blind zone through multi-layer consistency iterative verification. The preliminary prediction sequence is converted into a comprehensive insulation damage energy density to identify weak hot spots in the insulation. A lightweight physical surrogate function is constructed using the LP perturbation method to iteratively physical correct the preliminary multidimensional degradation feature prediction sequence. Based on the converged overall insulation degradation index, the remaining flight hours and reference values ​​are calculated, a prediction interval is generated, and the final remaining service life prediction value is output adaptively according to the location of the weak area.

[0006] As a further option of the method of the present invention, the method for forming the quasi-steady-state operating segment includes: Align multi-source signal streams with a unified time base; Within a preset sliding time window, the first condition is that the flight parameter stage switching label remains unchanged and the center of the window is not in the flight stage switching transition interval. The second condition is that the mean of the operating condition change rate sequence within the window is lower than the preset mean change rate threshold and the maximum value is lower than the preset peak change rate threshold. Starting from the origin of the multi-source signal stream, the sliding time window is moved with a preset step size, and the continuous time interval that simultaneously satisfies the first and second conditions is marked as a quasi-steady-state operation segment.

[0007] As a further option of the method of the present invention, the insulation degradation trend blind zone calibration step includes: Identify the operating condition change intervals, which include the preset transition time before and after the flight phase switching time and the time period when the operating condition change rate exceeds the preset change threshold. Identify the intermittent sampling interval of the sensor, and determine the interval in which the sampling interval exceeds a preset multiple of the nominal sampling period and two or more sensors show intermittent sampling at the same time as the intermittent sampling interval of the sensor. Identify data packet loss intervals and mark the intervals in which packet sequence numbers are continuously missing and the time span of the missing segment is greater than a preset multiple of a single sampling period as data packet loss intervals. The union of the operating condition change interval, the sensor intermittent sampling interval, and the data packet loss interval is used to form a blind zone of insulation degradation trend. For each blind zone of insulation degradation trend, the most recent quasi-steady-state operating segment is searched forward along the time axis as the forward quasi-steady-state operating segment, and the most recent quasi-steady-state operating segment is searched backward as the backward quasi-steady-state operating segment.

[0008] As a further option of the method of the present invention, the multidimensional degradation feature sequence extraction includes: The amplitude variation coefficient and pulse repetition rate of partial discharge are calculated step by step from the amplitude and phase sequence of partial discharge; the proportion of vibration fundamental frequency energy is calculated step by step from the broadband vibration acceleration spectrum; the temperature rise rate of hot spots is calculated step by step from the distribution of hot spots in fiber optic temperature measurement; and the third harmonic distortion rate of leakage current is calculated step by step from the total harmonic of leakage current to ground. The partial discharge amplitude variation coefficient, pulse repetition rate, vibration fundamental frequency energy ratio, hot spot temperature rise rate, and leakage current third harmonic distortion rate at each time step are combined to form a multidimensional degradation feature vector.

[0009] As a further option of the method of the present invention, the step of calculating dynamic reliability weights and generating complete forward boundary degradation feature sequences and backward boundary degradation feature sequences based on the physical consistency criterion includes: For each dimension of degradation feature component, a physical consistency reference model is established. Based on whether the feature component exceeds the preset reasonable range, the sensor baseline drift trend, and the detection of instantaneous abnormal peaks in the signal, dynamic confidence weights with values ​​between 0 and 1 are generated. The physical consistency reference model is based on the Weibull model of the statistical distribution of partial discharge. Pair the multidimensional degradation feature vector of each time step with the corresponding dynamic confidence weight vector and arrange them in chronological order to form the forward boundary degradation feature sequence; perform the same operation on the backward quasi-steady-state running segment to obtain the backward boundary degradation feature sequence.

[0010] As a further option of the method of the present invention, the step of establishing the mapping relationship between the multidimensional degradation feature sequence and the overall insulation deterioration index includes: An overall insulation degradation index is constructed using a weighted linear combination. : ,in, For the first Degenerative eigencomponents, These correspond to the coefficient of variation of partial discharge amplitude, pulse repetition rate, proportion of vibration fundamental frequency energy, hot spot temperature rise rate, and third harmonic distortion rate of leakage current, respectively. The fusion coefficient is... For the intercept term; By using historical degradation characteristic data and insulation diagnostic test results, the prediction error is minimized to determine and .

[0011] As a further option of the method of the present invention, the forward and reverse weighted trend recursion execution steps include: For the forward boundary degradation feature sequence, select the last one closest to the boundary of the insulation degradation trend blind zone. Using the feature data from each time step as the extrapolation modeling window, and with dynamic confidence as the weight, a weighted local linear model is established for each dimension of degradation feature component to predict the first time step within the insulation degradation trend blind zone. eigenvalues ,in The weighted average of time indices within the extrapolation modeling window. For the first The weighted average of the dimensional degenerate eigenvalues. The weighted slope; Add the predicted values ​​to the sequence and move the window to generate a positive prediction sequence point by point; After time reversing the backward boundary degradation feature sequence, the same operation is performed to generate the inverse prediction sequence.

[0012] As a further option of the method of the present invention, the multi-layer consistency iterative verification includes: The first layer of verification involves performing first-order difference detection on the predicted sequence of each degradation feature component within the blind zone of insulation deterioration trend, and using weighted median filtering to correct predicted oscillation points that exceed the difference threshold. The second layer of verification establishes cross-correlation constraints by utilizing the physical relationship between the hot spot temperature rise rate and the proportion of vibration fundamental frequency energy, and simultaneously corrects the predicted values ​​of the hot spot temperature rise rate and the proportion of vibration fundamental frequency energy. The third layer of verification calculates the weighted Euclidean distance between the forward and reverse prediction sequences and the midpoint of the insulation degradation trend blind zone as the midpoint consistency. When the midpoint consistency exceeds the preset consistency threshold, the order of the trend extrapolation algorithm is adjusted and the fixed step size is switched to a time-varying step size.

[0013] As a further option of the method of the present invention, the steps of generating the preliminary multidimensional degradation feature prediction sequence and the preliminary overall insulation degradation index within the insulation degradation trend blind zone include: Define the weighted residual of the blind zone of the total insulation degradation trend. ,in This represents the total number of time steps within the blind zone of insulation degradation trend. For the degenerate feature dimension, and The first The first time step Forward and backward predicted values ​​of the dimensional degenerate eigencomponents. This corresponds to the dynamic credibility weights; Repeatedly perform forward and reverse trend extrapolation and first to third layer verification until the weighted residual of the blind zone of the full insulation degradation trend is lower than the convergence threshold. Take the weighted average of the forward and reverse prediction sequences as the preliminary multidimensional degradation feature prediction sequence. By utilizing the mapping relationship between the multidimensional degradation feature sequence and the overall insulation degradation index, the feature vectors of each time step of the preliminary multidimensional degradation feature prediction sequence are substituted into... A preliminary sequence of overall insulation degradation indicators was obtained, among which, For the first Degenerative eigencomponents, These correspond to the coefficient of variation of partial discharge amplitude, pulse repetition rate, proportion of vibration fundamental frequency energy, hot spot temperature rise rate, and third harmonic distortion rate of leakage current, respectively. The fusion coefficient is... This is the intercept term.

[0014] As a further option of the method of the present invention, the step of converting the preliminary prediction sequence into a comprehensive insulation damage energy density includes: The preliminary multidimensional degradation feature prediction sequence was combined with synchronously acquired partial discharge amplitude, fiber optic thermography hotspot values, and effective vibration acceleration values ​​to calculate the partial discharge energy density step by step. Dielectric loss, heating power density, and cumulative energy density and thermomechanical stress shear energy density The combined insulation damage energy density is obtained by superposition. .

[0015] As a further option of the method of the present invention, the iterative physical correction step of the lightweight physical surrogate function on the preliminary multidimensional degradation feature prediction sequence includes: A three-layer progressive scan of the motor insulation structure is performed, consisting of the single-turn coil micro-element layer, the single-slot winding middle layer, and the overall phase winding macro-layer. Micro-element aggregates that exceed the preset upper limit threshold for comprehensive insulation damage energy density are identified as weak insulation hotspots. For each weakly insulating hotspot region, a small parameter is introduced to represent the coupling effect strength. The electro-thermal-mechanical multiphysics coupling degradation equation is asymptotically expanded, retaining the zero-order uniform degradation term and the first-order key coupling term, and a lightweight physical surrogate function is constructed. ,in This is the amount of degradation amplitude correction. , , These are temperature, electric field strength, and stress, respectively. The zero-order uniform degradation amplitude, , , These are the thermo-electric coupling coefficient, the thermo-mechanical coupling coefficient, and the electro-mechanical coupling coefficient, respectively.

[0016] As a further option of the method of the present invention, the step of iteratively physical correcting the preliminary multidimensional degradation feature prediction sequence by the lightweight physical surrogate function further includes: The degradation amplitude of the corresponding weak insulation hotspot region in the preliminary multidimensional degradation feature prediction sequence is corrected point by point using a lightweight physical surrogate function to obtain the physically corrected multidimensional degradation feature sequence. Calculate the state transition deviation sequence of the predicted sequence before and after physical correction; The root mean square value of the state transition deviation sequence is fed back into the trend extrapolation algorithm. The weighting coefficient decay factor and step size adjustment factor are finely adjusted, and the forward and reverse trend extrapolation and consistency verification are re-executed. The process is iterated until the L2 norm of the state transition deviation sequence converges, and the converged overall insulation degradation index benchmark sequence is obtained.

[0017] As a further option of the method of the present invention, the step of calculating the main remaining flight hours and reference value includes: The converged overall insulation degradation index benchmark sequence is extrapolated to the preset insulation failure threshold to obtain the main remaining flight hours; The first reference remaining flight hours are obtained by extrapolating the preliminary overall insulation degradation index sequence without physical correction. The average temperature of the fiber optic thermometer hotspots in the forward and backward boundary degradation feature sequences is extracted. Based on the Arrhenius thermal lifetime formula, the empirical overall degradation index sequence is calculated and extrapolated to obtain the second reference remaining flight hours.

[0018] As a further option of the method of the present invention, the step of obtaining the final remaining useful life prediction value includes: The half median deviation is calculated by constructing a sample set using the primary remaining flight hours, the first reference remaining flight hours, and the second reference remaining flight hours. Using the main remaining flight hours as the center, and twice the half-width of the median deviation of half the data, a remaining service life prediction interval is generated; When the calibrated weak hot spot area is located in the main insulation or a critical part of the slot, the lower limit of the prediction interval is output as the final remaining service life prediction value; otherwise, the midpoint of the prediction interval is output.

[0019] Another aspect of this application provides a lightweight high-voltage high-power aviation motor insulation degradation trend prediction system, the system comprising: The quasi-steady-state segment cutting and blind zone calibration module is used to cut the synchronously acquired multi-source signal stream into quasi-steady-state operating segments and calibrate the insulation degradation trend blind zone by using the flight parameter stage switching label and the operating condition change rate joint criterion. The degradation feature extraction and mapping module is used to extract multidimensional degradation feature sequences from the forward and backward quasi-steady-state segments of each insulation degradation trend blind zone, calculate dynamic confidence weights based on the physical consistency criterion, generate complete forward boundary degradation feature sequences and backward boundary degradation feature sequences, and establish a mapping relationship between the multidimensional degradation feature sequences and the overall insulation degradation index. The trend recursion and consistency verification module is used to start from the forward and backward boundary degradation feature sequences, perform forward and backward weighted trend recursion, and generate a preliminary multidimensional degradation feature prediction sequence and a preliminary overall insulation degradation index within the insulation degradation trend blind zone through multi-layer consistency iteration verification. The physical correction module is used to convert the preliminary prediction sequence into a comprehensive insulation damage energy density, identify weak hot spots in the insulation, and use the LP perturbation method to construct a lightweight physical surrogate function to iteratively correct the preliminary multidimensional degradation feature prediction sequence. The life prediction and output module is used to calculate the main remaining flight hours and reference value based on the converged overall insulation degradation index, generate prediction intervals, and adaptively output the final remaining life prediction value according to the location of the weak area.

[0020] The beneficial effects of this invention are: This invention addresses the problems of low reconstruction accuracy in data blind zones, insufficient fusion of multi-physics coupling mechanisms, and excessive computational complexity in existing aerospace motor insulation prediction methods. It proposes a lightweight insulation degradation trend prediction method. Through quasi-steady-state segmentation and bidirectional weighted trend extrapolation driven by dynamic credibility weights, combined with three-layer iterative verification of smoothness, physical correlation consistency, and midpoint fit, it effectively overcomes the boundary oscillations and physical mismatch defects caused by simple interpolation within the blind zone. The average weighted residual across the entire blind zone is reduced to the order of 0.0018, and the reconstruction accuracy is improved by more than an order of magnitude compared to traditional interpolation methods. The lightweight physical surrogate function constructed using the LP perturbation method retains only the zero-order uniform degradation term and the first-order key coupling term, transforming the multi-field coupled partial differential equation system into algebraic operations. While maintaining physical consistency, the computation time is reduced by more than two orders of magnitude compared to the finite element method, meeting the lightweight real-time inference requirements of airborne edge computing platforms. By fusing the master prediction sequence with the Arrhenius empirical index and the half median deviation of the uncorrected index, a remaining lifetime prediction interval with confidence interval is generated. Combined with the adaptive decision-making of weak hot spots, a conservative or mean lifetime is output. In the field test, the deviation between the lower bound of the prediction interval and the actual insulation abnormality alarm time is less than 5%, which significantly improves the robustness of insulation degradation trend prediction and the safety of maintenance decision-making under complex interferences such as sensor intermittent and sudden changes in operating conditions. Attached Figure Description

[0021] Figure 1 A schematic diagram of the overall method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors; Figure 2 S100 Flowchart for Predicting Insulation Deterioration Trends of Lightweight High-Voltage High-Power Aviation Motors; Figure 3 S200 Flowchart for Predicting Insulation Deterioration Trends of Lightweight High-Voltage High-Power Aviation Motors; Figure 4 S300 Flowchart for Predicting Insulation Degradation Trends of Lightweight High-Voltage High-Power Aviation Motors; Figure 5 S400 flowchart for predicting insulation degradation trends of lightweight high-voltage high-power aviation motors; Figure 6 S500 flowchart for predicting insulation degradation trends of lightweight high-voltage high-power aviation motors; Figure 7 A diagram showing the distribution of overall insulation damage energy density along slot number; Figure 8 A comparative curve of the overall insulation degradation index trends; Figure 9 , Figure 10 , Figure 11 , Figure 12 and Figure 13Diagram of the operating interface of a lightweight, high-voltage, high-power aviation motor insulation degradation trend prediction system. Detailed Implementation

[0022] 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 embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The theoretical basis of this invention is built upon the degradation dynamics of multi-source insulation characterization quantities, the weighted trend extrapolation theory, the multi-physics field coupled insulation degradation mechanism, and the LP perturbation asymptotic expansion method. By utilizing the flight parameter stage switching label and the operating condition change rate as joint criteria, the multi-source signal stream synchronously acquired during the operation of high-voltage, high-power aircraft motors is segmented into quasi-steady-state operating segments, and insulation degradation trend blind zones are identified. For each blind zone's forward and backward quasi-steady-state segments, multi-dimensional degradation features including the partial discharge amplitude variation coefficient, pulse repetition rate, vibration fundamental frequency energy ratio, hot spot temperature rise rate, and leakage current third harmonic distortion rate are extracted. Dynamic reliability weights are generated based on the physical consistency criterion, forming a boundary degradation feature sequence. Weighted trend extrapolation is performed from both forward and reverse time directions. After intersecting at the middle of the blind zone, a three-layer iterative verification mechanism is initiated, including smoothness verification, physical correlation consistency verification, and midpoint coincidence verification, with the dynamic reliability weights as... The weighted coefficients are repeatedly iterated to obtain a preliminary multidimensional degradation feature prediction sequence and map it to a preliminary overall insulation degradation index. Then, the preliminary prediction sequence is combined with the measured signal to transform it into a comprehensive insulation damage energy density sequence. The weak hot spots in the insulation are identified by a three-layer progressive scan. A lightweight physical surrogate function that retains the zero-order uniform degradation term and the first-order critical coupling term is constructed using the LP perturbation method to physically correct the degradation amplitude. The state transition deviation is fed back to the trend extrapolation stage to fine-tune the parameters and iterate until the deviation converges. Finally, the main remaining flight hours are extrapolated from the converged overall degradation index. The uncorrected index and the Arrhenius empirical index are fused to generate a half median deviation prediction interval. The final remaining service life prediction value is output based on the hot spot location. The core theoretical conception includes: an insulation degradation state transition equation considering multi-factor coupling, whose general form is that the derivative of the degradation feature vector with respect to time is jointly described by the homogeneous degradation rate matrix, the coupled degradation increment matrix, and the random perturbation term; robust estimation of the degradation trend within the blind zone is achieved by introducing dynamic confidence weights to perform weighted least squares extrapolation on the degradation feature observation sequence; a comprehensive insulation damage energy density expression is established using the energy conservation of electro-thermal-mechanical multiphysics fields; small parameters are introduced into the degradation equation to expand the state variables into an asymptotic series, separating the zero-order term and the first-order coupling term, thereby obtaining a lightweight surrogate model that can be solved quickly. This theoretical framework unifies blind zone feature reconstruction, physical consistency constraints, and computational lightweighting within the same processing flow, providing a physical and mathematical foundation for specific implementation.

[0024] The specific embodiments of the present invention will be described in detail below.

[0025] Example 1: Please see Figure 1 This invention illustrates a method for predicting the insulation degradation trend of lightweight high-voltage high-power aviation motors, as provided in an embodiment of the present invention, comprising: S100: Quasi-steady-state cutting of multi-source signal streams and blind zone calibration of insulation degradation trend; S200: Construction of Boundary Degradation Feature Sequences and Generation of Dynamic Credibility Weights; S300: Blind zone degradation feature estimation based on weighted forward and reverse prediction and three-layer verification; S400: Weak region identification and physical proxy correction based on comprehensive insulation damage energy density; S500: Generation of remaining useful life prediction range and output of final prediction value.

[0026] The specific details are as follows: Please refer to Figure 2 The flowchart of S100 in an exemplary lightweight high-voltage high-power aviation motor insulation degradation trend prediction method of this application is shown, and the content includes: S110: Obtain the flight parameter phase switching tag sequence from the aviation motor operation monitoring system, and synchronously collect operating parameters from the motor controller and sensor network to calculate the operating condition change rate.

[0027] S120: Align the partial discharge amplitude and phase sequence, broadband vibration acceleration spectrum, fiber optic temperature measurement hotspot distribution, and ground leakage current full harmonic multi-source signal stream with a unified time base, and use the flight parameter stage switching label and operating condition change rate joint criterion to cut the multi-source signal stream into quasi-steady-state operation segments.

[0028] S130: Identify the intervals of sudden changes in operating conditions, the intervals of intermittent sampling by sensors, and the intervals of data loss, and merge these intervals into the insulation degradation trend blind zone, and record the forward quasi-steady-state operating segment index and the backward quasi-steady-state operating segment index corresponding to each blind zone.

[0029] In one possible implementation of step S110, the process of obtaining the flight parameter stage switching label and the rate of change of operating condition further includes: S111: Extract the flight phase transition tag sequence from the flight parameter recording system or onboard maintenance computer. Flight phase transition tags include standard phases such as taxiing, takeoff, climb, cruise, descent, approach, and landing. Each phase tag is accompanied by a world standard timestamp, marking the start and end times of the phase. The extraction process involves reading data frames conforming to the ARINC717 or ARINC429 protocol via the onboard data bus, parsing the binary-encoded flight phase identifiers, and forming a flight phase transition tag sequence arranged in ascending chronological order.

[0030] S112: Synchronously acquires the instantaneous values ​​of the high-voltage DC bus voltage, three-phase AC phase current, motor speed command and actual speed feedback, and torque command signals from the digital signal processor or monitoring unit of the motor controller. Simultaneously, it acquires temperature readings from each channel of the temperature-sensing optical fiber arranged axially and circumferentially along the stator winding from the airborne distributed optical fiber temperature measurement system; acquires triaxial vibration acceleration signals from the bearing housing and casing measuring points from the broadband vibration acceleration sensor; acquires the amplitude and phase angle information of the partial discharge pulse from the partial discharge coupling sensor, with the phase angle reference point synchronized with the zero-crossing point of the fundamental wave of the motor terminal voltage; and acquires the time-domain waveform of the leakage current from the ground leakage current transformer. The sampling frequency of each signal is set according to the motor's rated speed, insulation characteristic frequency band, and partial discharge pulse resolution requirements to ensure sufficient feature extraction.

[0031] S113: For the synchronously acquired operating condition signals, calculate the operating condition change rate step by step. The operating condition change rate is a weighted synthesis of the change rate of the high-voltage DC bus voltage, the change rate of the effective value of the phase current, the change rate of the rotational speed, and the change rate of the torque command. The construction method is as follows: calculate the ratio of the standard deviation to the mean of each operating condition parameter within the sliding time window, multiply each by the corresponding normalized weighting coefficient, and then sum them. The weighting coefficients are experimentally calibrated based on the sensitivity of each operating condition parameter to the influence of insulation stress. The higher weights are chosen for the voltage change rate and the rotational speed change rate to reflect the combined influence of high-frequency pulse voltage amplitude fluctuations and centrifugal force changes on the insulation. The final output is a scalar operating condition change rate sequence, with the sampling time synchronized with the multi-source signals.

[0032] This step S120 further includes: S121: For four types of signal streams—partial discharge amplitude-phase sequence, broadband vibration acceleration spectrum, fiber optic temperature measurement hotspot distribution, and ground leakage current full harmonic—time base alignment is performed based on the GPS or airborne network time stamp. For heterogeneous signals with inconsistent sampling rates, cubic spline interpolation is used to resample to a unified time grid. The grid spacing is taken as the greatest common divisor of the original sampling intervals of each signal, and the synchronization of the aligned signals is verified by cross-correlation analysis.

[0033] S122: Define the joint criterion for quasi-steady-state operation segments. Within a sliding time window of preset length, a time segment that simultaneously meets the following two conditions is determined to be in quasi-steady-state operation: Condition 1: The flight parameter phase switching label within the window remains unchanged, and the center of the window never falls within the transition range of any flight phase switching. Condition 2: The mean value of the operating condition rate of change sequence within the window is lower than the preset mean value threshold, and the maximum value of the operating condition rate of change sequence within the window is lower than the preset peak value threshold. These thresholds are set based on the motor's rated operating conditions and the results of the accelerated aging test of the insulation.

[0034] S123: Starting from the beginning of the multi-source signal stream, the sliding time window is moved with a preset step size, and continuous time intervals that satisfy the joint criterion S122 are marked as quasi-steady-state operating segments. Each quasi-steady-state operating segment records the segment start and end timestamps, data block pointers of the multi-source signals within the segment, and statistical characteristics of the rate of change of operating conditions within the segment. Adjacent quasi-steady-state operating segments are naturally separated by non-quasi-steady-state intervals.

[0035] S124: When the duration of a non-quasi-steady-state interval exceeds the preset minimum interval, the most stable sub-interval within that interval is found as an additional quasi-steady-state segment for recording; if the shortest duration is less than the preset minimum interval, it is merged with the adjacent non-quasi-steady-state interval to ensure the continuity of the quasi-steady-state segments. Output a list of all quasi-steady-state running segments and the time-domain boundaries of each segment.

[0036] This step S130 further includes: S131: Identify abrupt change intervals in operating conditions. Traverse all flight phase switching labels, marking a preset transition time period before and after each flight phase switching moment as an abrupt change interval in operating conditions; simultaneously, scan the operating condition change rate sequence, marking the moment when the operating condition change rate exceeds a preset abrupt change threshold and the short extended time period before and after it as an abrupt change interval in operating conditions. Merge all overlapping or adjacent abrupt change intervals in operating conditions to form a continuous non-quasi-steady-state period.

[0037] S132: Identify intermittent sampling intervals for sensors. For each sensor signal, check the difference between adjacent sampling timestamps. When the sampling interval exceeds a preset multiple of the sensor's nominal sampling period, mark the interval as a suspected intermittent sampling interval. If two or more sensors simultaneously exhibit intermittent sampling within the same time period, determine the interval as a sensor intermittent sampling interval, and record the start and end times of the interval and the sensor channels involved.

[0038] S133: Identify data packet loss intervals. Analyze the transmission protocol checksum, packet sequence number continuity, and timestamp jump characteristics of the airborne data bus or acquisition and storage system. When consecutive packet sequence number gaps are detected and the time span of the gap is greater than a preset multiple of a single sampling period, mark that time span as a data packet loss interval. For partial discharge amplitude-phase sequences, if the phase encoding is broken or the number of sampling points is incomplete, the corresponding time period is also marked as a packet loss interval.

[0039] S134: The union of the operating condition abrupt change intervals obtained in S131, the sensor intermittent sampling intervals obtained in S132, and the data packet loss intervals obtained in S133 is spatially merged and filled with minimal intervals to form a set of insulation degradation trend blind zones. Each insulation degradation trend blind zone includes the start time of the blind zone, the end time of the blind zone, and a marker for missing or unreliable data types within the blind zone. For each blind zone, the nearest quasi-steady-state operating segment is searched forward along the time axis and recorded as the forward quasi-steady-state operating segment of the blind zone; the nearest quasi-steady-state operating segment is searched backward along the time axis and recorded as the backward quasi-steady-state operating segment of the blind zone. If the time interval between the forward or backward quasi-steady-state operating segment and the blind zone boundary exceeds a preset maximum extrapolation span, the specified duration data closest to the blind zone boundary is extracted as the boundary segment.

[0040] Please refer to Figure 3 The flowchart of S200 in an exemplary lightweight high-voltage high-power aviation motor insulation degradation trend prediction method of this application is shown, and the content includes: S210: For each forward quasi-steady-state operating segment and backward quasi-steady-state operating segment corresponding to each insulation degradation trend blind zone, calculate the coefficient of variation of partial discharge amplitude, pulse repetition rate, proportion of vibration fundamental frequency energy, hot spot temperature rise rate and third harmonic distortion rate of leakage current step by step from four signals: partial discharge amplitude-phase sequence, broadband vibration acceleration spectrum, fiber optic temperature measurement hot spot distribution and ground leakage current full harmonic distortion rate, to form a multidimensional degradation characteristic sequence.

[0041] S220: Based on the physical consistency criterion, a dynamic confidence weight is automatically calculated for each dimension of the degraded feature component at each time step, which is used to quantify the degree to which the feature component is affected by sensor drift, transient interference or data quality degradation.

[0042] S230: Concatenate the multidimensional degradation feature sequence with the corresponding dynamic confidence weights along the time axis to generate complete forward boundary degradation feature sequences and backward boundary degradation feature sequences.

[0043] S240: Utilize historically accumulated degradation characteristic data and insulation diagnostic test results to establish a mapping relationship from multidimensional degradation characteristic sequences to overall insulation deterioration indicators, and store the mapping relationship parameters.

[0044] In one possible implementation of step S210, the calculation process of the multidimensional degradation features further includes: S211: Calculation of the coefficient of variation of partial discharge amplitude. The partial discharge amplitude-phase sequence is divided into phase windows according to the power frequency cycle, and the width of the phase window is set according to the phase resolution requirements. Within each phase window, the standard deviation and mean of the discharge pulse amplitude are statistically analyzed, and the coefficient of variation of amplitude is defined as the ratio of the standard deviation to the mean. If the number of discharge pulses in a certain phase window is less than the minimum number required for statistical stability, interpolation is performed using the coefficients of variation of adjacent phase windows to complete the result. The coefficient of variation of amplitude reflects the broadening of the partial discharge amplitude distribution. As insulation deteriorates, the dispersion of discharge amplitude increases, and the coefficient of variation rises.

[0045] S212: Calculation of pulse repetition rate. The total number of partial discharge pulses is counted within a unit time window, and divided by the window duration to obtain the pulse repetition rate. The pulse repetition rate directly reflects the intensity of partial discharge activity; during the insulation degradation process, the discharge repetition rate typically shows a monotonically increasing trend.

[0046] S213: Calculation of the fundamental frequency energy proportion of vibration. A short-time Fourier transform is performed on the broadband vibration acceleration spectrum signal to obtain the time-varying spectrum. The spectral energy components corresponding to the motor rotation frequency and its lower harmonics are extracted. The proportion of the sum of these fundamental frequency and harmonic components to the total vibration energy within the analysis frequency band is calculated as the fundamental frequency energy proportion. With the development of insulation thermal aging and mechanical loosening, the high-frequency random vibration component increases, and the fundamental frequency energy proportion gradually decreases.

[0047] S214: Calculation of Hotspot Temperature Rise Rate. Extract the temperature value of the highest measuring point from the fiber optic temperature measurement hotspot distribution sequence. Perform a first-order difference on this temperature time series and divide by the sampling time step to obtain the instantaneous temperature rise rate. To avoid measurement noise interference, a median filter is used to smooth the temperature rise rate sequence; the filter window width is selected based on the noise level.

[0048] S215: Calculation of the third harmonic distortion rate of leakage current. A discrete Fourier transform is performed on the total harmonic signal of the leakage current to extract the fundamental amplitude and the third harmonic amplitude. The third harmonic distortion rate is defined as the percentage of the third harmonic amplitude to the fundamental amplitude. Insulation aging leads to increased dielectric nonlinearity of the material, resulting in an increase in the third harmonic component in the leakage current.

[0049] The five degradation feature components calculated from S211 to S215 are used to form a multidimensional degradation feature vector, denoted as the first... The feature vectors at each time step are ,in This is the coefficient of variation of the partial discharge amplitude. The pulse repetition rate, The percentage of the fundamental frequency energy. For the hot spot temperature rise rate, The third harmonic distortion rate of the leakage current.

[0050] In one possible implementation of step S220, the calculation process of the dynamic credibility weight further includes: S221: Establish a physical consistency reference model for each dimension of degradation characteristics. For the partial discharge amplitude variation coefficient, the physical consistency reference model is based on the Weibull model of the statistical distribution of partial discharge. The reasonable range of characteristic components is defined by the baseline value when the insulation is intact and the saturation value when near breakdown. When the instantaneous characteristic value exceeds the reasonable range, the confidence weight decays exponentially. For the proportion of vibration fundamental frequency energy, the reference model is associated with the stability of motor speed. When the speed fluctuation exceeds the calibration threshold, the confidence of the fundamental frequency energy proportion at that moment decreases. For the hot spot temperature rise rate, the self-consistency of the measured value is evaluated using a simplified one-dimensional thermal balance equation. The larger the equation residual, the lower the confidence. For the third harmonic distortion rate of leakage current, the nonlinear volt-ampere characteristic model of the insulation material is referenced. When the operating voltage at the measuring point deviates from the rated value, the expected range of the distortion rate is corrected according to the nonlinear coefficient.

[0051] S222: Integrates sensor drift and transient interference detection into confidence calculation. The baseline drift of each sensor signal is monitored in real time. A high-pass filter is used to extract the signal trend term. When the slope of the trend term exceeds a set threshold, the confidence weight of the corresponding time step is multiplied by an attenuation factor. Simultaneously, isolated spikes or pulse interference in the signal are identified, and outliers are determined using a statistical thresholding method. The confidence weights of outliers and their neighboring time steps are reduced.

[0052] S223: Combining the evaluation results of S221 and S222, generate dynamic confidence weights for each dimension of the degenerate feature component at each time step. The weights range from 0 to 1, where 1 represents complete confidence and 0 represents complete unconfidence. The first time step The dynamic reliability weights of the degenerate feature components are denoted as: , The five weights are combined to form a dynamic credibility weight vector. .

[0053] In one possible implementation of step S230, the process of generating the boundary degradation feature sequence includes: S231: The multidimensional degenerate feature vector of each time step within the forward quasi-steady-state running segment. With dynamic credibility weight vector Pair them up to form weighted feature tuples. Arrange all weighted feature tuples in ascending time order to form the forward boundary degenerate feature sequence.

[0054] S232: Perform the same operation on the backward quasi-steady-state running segments, arranging them in ascending time order to form a backward boundary degradation feature sequence. Since backward prediction requires backward extrapolation from the end of the blind zone to past times, the backward boundary degradation feature sequence will undergo time reversal index mapping before entering the extrapolation module, so that the time independent variable of the sequence logically points from the blind zone boundary to the inside of the blind zone.

[0055] In one possible implementation of step S240, the process of establishing the mapping relationship between degradation characteristics and overall insulation deterioration index includes: S241: Collect a large number of degradation characteristic data samples accumulated during the service of high-voltage, high-power aircraft motors of the same model, as well as insulation condition level or residual breakdown voltage calibration values ​​obtained through periodic insulation diagnostic tests. Train a fusion model with multidimensional degradation feature vectors as input and insulation condition level as output.

[0056] S242: The fusion model adopts a weighted linear combination form, and the overall insulation degradation index... Represented as: ; in, For the first The fusion coefficients of the dimensional degenerate feature components. This is the intercept term. The fusion coefficients are determined by optimization algorithms that minimize the mean squared prediction error on the calibration data or maximize the ranking correlation.

[0057] S243: The fusion coefficients after training With intercept This data is stored as a mapping parameter. Subsequently, given any multidimensional degradation feature vector, the corresponding overall insulation degradation index can be quickly calculated through the above linear combination.

[0058] Please refer to Figure 4 The flowchart of S300 in an exemplary lightweight high-voltage high-power aviation motor insulation degradation trend prediction method of this application is shown, and the content includes: S310: Starting from the end of the forward boundary degradation feature sequence, a weighted trend extrapolation algorithm is used to predict the changes in each dimension of degradation features within the blind zone point by point along the positive time direction; simultaneously, starting from the end of the backward boundary degradation feature sequence, reverse prediction is performed point by point along the negative time direction. The initial forward and reverse recursions are both based on the weighted linear interpolation of their respective boundary degradation feature sequences to generate the initial prediction sequence.

[0059] S320: When the forward prediction sequence and the reverse prediction sequence intersect for the first time at the middle time point of the blind zone, a multi-layer consistency iterative verification mechanism is initiated, which includes the first layer of smoothness verification, the second layer of physical correlation consistency verification, and the third layer of midpoint coincidence verification.

[0060] S330: Using dynamic confidence weights as weighting coefficients, repeatedly execute the inner loop iteration of forward and reverse recursion and three-layer verification until the weighted residual of the forward and reverse prediction sequences on the full blind zone scale is lower than the set convergence threshold, thus obtaining the preliminary multidimensional degradation feature prediction sequence within the insulation degradation trend blind zone.

[0061] S340: The preliminary multidimensional degradation feature prediction sequence is converted into preliminary overall insulation degradation index step by step through the mapping relationship established by S240, forming a preliminary overall insulation degradation index sequence inside the blind zone.

[0062] In one possible implementation of step S310, the execution process of the weighted trend extrapolation algorithm further includes: S311: Suppose the forward boundary degradation feature sequence contains Each time step is denoted by a time index. The corresponding multidimensional degenerate feature vector is The dynamic credibility weight vector is Select the last one closest to the boundary of the blind zone. The feature data at each time step is used as the extrapolation modeling window. The value is adaptively determined based on the sequence length and is not less than the minimum number of points required to ensure the model can be identified. Using the time indices of these time steps as independent variables, a weighted local linear model is established for each dimension of the degenerate feature component. The model parameters are obtained by minimizing the weighted sum of squared residuals, and the weights are the dynamic confidence weights of the feature component at the corresponding time step. .

[0063] S312: Using the fitted weighted local linear model, predict the first time step within the blind zone. eigenvalues. For the eigenvalues ​​of . The predicted value of the dimensional degenerate feature component is given by the following formula: ; in, The weighted average of time indices within the extrapolation modeling window. For the first The weighted average of the eigenvalues. The weighted slope is used as the weighted average. The weighted average and weighted slope are obtained by... The weights are obtained using the weighted least squares method. The confidence weights of the predicted values ​​are calculated by the weighted average of the weights within the extrapolation modeling window, and multiplied by a decay factor that decreases with the number of extrapolation steps to reflect the cumulative effect of prediction uncertainty with the extrapolation distance.

[0064] S313: Concatenate the newly predicted feature vector and the corresponding prediction confidence weight to the end of the forward boundary degenerate feature sequence. At the same time, move the extrapolation modeling window forward by one step along the time axis. Repeat the prediction process of S312 to generate a positive prediction sequence point by point until it is filled to the predetermined intersection point in the middle of the blind zone.

[0065] S314: For inverse prediction, the backward boundary degradation feature sequence is time-reversed so that the time index of the sequence increases in the same direction as the forward direction but points into the blind zone. Using the same weighted local linear modeling and extrapolation process as S312, the inverse prediction sequence is generated by predicting point by point from the end of the backward boundary sequence to the middle of the blind zone.

[0066] In one possible implementation of step S320, the specific execution process of the three-layer consistency check includes: S321: First-level verification – Smoothness verification of single-dimensional degenerate feature. Smoothness verification is performed on the forward and reverse prediction sequences of each degenerate feature component within the blind zone. The first-order difference sequence of the prediction sequence is calculated, and time steps where the absolute value exceeds a preset difference threshold are identified as prediction oscillation points. For oscillation points, a locally weighted median filter is used to replace the original prediction value, with the filter weights also taken from the confidence weights of the corresponding time steps. After smoothness verification, the corrected forward and reverse prediction sequences are output.

[0067] S322: Second-level verification – Physical correlation consistency verification. During the insulation degradation process of high-power-density motors, there is a physical correlation between the hotspot temperature rise rate and the proportion of vibration fundamental frequency energy: when insulation thermal aging leads to increased material loss and accelerated hotspot temperature rise, the stiffness of the insulation material decreases, the stator winding vibration mode shifts, and the proportion of fundamental frequency energy changes regularly. The cross-correlation coefficient sequence between the hotspot temperature rise rate and the proportion of vibration fundamental frequency energy is calculated using historical degradation data samples, and a cross-correlation constraint function between the two is established. For the first... At each time step, the physical correlation consistency constraint is calculated using the cross-correlation function: when there is a deviation between the predicted value of the vibration fundamental frequency energy ratio and the expected value calculated by the hot spot temperature rise rate through the cross-correlation model, the predicted values ​​of the hot spot temperature rise rate and the vibration fundamental frequency energy ratio are simultaneously corrected using the cross-correlation coupling coefficient as a scaling factor, so that the degradation characteristics of the two dimensions maintain a physically consistent trend.

[0068] S323: Third-layer verification – Midpoint consistency verification. At the midpoint of the blind zone, the weighted Euclidean distance between the multidimensional degenerate feature vectors given by the forward and backward prediction sequences is calculated as the midpoint consistency index. The weighted Euclidean distance uses the harmonic mean of the prediction confidence weights of each feature component as the weights. When the midpoint consistency exceeds a preset consistency threshold, it is determined that the current order or step size of the trend extrapolation algorithm is not suitable for the nonlinearity of the blind zone feature, triggering the order and step size adjustment procedure. Order adjustment includes upgrading the local linear model to a local quadratic model, and step size adjustment switches the fixed extrapolation step size to a time-varying step size. The time-varying step size is inversely proportional to the local curvature of the current prediction point; the greater the curvature, the smaller the step size, in order to improve the prediction accuracy in the nonlinear region.

[0069] In one possible implementation of step S330, the inner loop iteration convergence process includes: S331: Define the full-blind-zone weighted residual as the sum of the squares of the prediction residuals of all time steps and all dimensions of degenerate feature components within the blind zone, multiplied by the corresponding dynamic confidence weights. Let the total number of time steps within the blind zone be... The dimension of the degradation feature is Then the weighted residual of the entire blind zone Calculated by the following formula: ; in, and The first The first time step Forward and backward predicted values ​​of the dimensional degenerate eigencomponents. This is the fusion confidence weight for this dimension of features at this time step. The convergence threshold is set based on statistical experiments.

[0070] S332: In each round of inner loop iteration, the following steps are executed sequentially: forward and reverse trend extrapolation (S310), smoothness verification (S321), physical correlation consistency verification (S322), and midpoint fit verification and order step size adjustment (S323). After each round of three-layer verification, the current full blind zone weighted residual is calculated. If the weighted residual is greater than the convergence threshold, the adjusted order and step size are fed back to the extrapolation module, and the sequence corrected for physical correlation consistency is used as the benchmark for the next round of extrapolation for a new round of iteration.

[0071] S333: The inner loop terminates when the weighted residual in the entire blind zone falls below the convergence threshold, or when the relative change in the weighted residual over two consecutive iterations is less than the minimum improvement ratio. The weighted average of the forward and reverse prediction sequences in the blind zone at each time step in the last iteration is taken as the initial multidimensional degradation feature prediction sequence, with the weights being the confidence weights for their respective directions.

[0072] In one possible implementation of step S340, the calculation process for the preliminary overall insulation degradation index includes: Substitute the feature vector of each time step of the preliminary multidimensional degradation feature prediction sequence into the fusion mapping formula determined in S242. The overall insulation degradation index is calculated point by point to form a preliminary overall insulation degradation index sequence that is continuous within the blind zone period.

[0073] Please refer to Figure 5 The flowchart of S400 in an exemplary lightweight high-voltage high-power aviation motor insulation degradation trend prediction method of this application is shown, and the content includes: S410: The preliminary multidimensional degradation feature prediction sequence obtained from S300 is combined with the partial discharge amplitude, fiber optic temperature hotspot values ​​and effective vibration acceleration values ​​collected from S100. The degradation features at each time step are converted into partial discharge energy density, dielectric loss heating power density and thermomechanical stress shear energy through physical mapping relationship, and then superimposed to obtain the comprehensive insulation damage energy density sequence.

[0074] S420: Based on the thermal-electric-mechanical tolerance characteristics of the insulation material, an upper limit threshold for energy density is set. The motor insulation structure corresponding to the blind zone of insulation degradation trend is scanned in three progressive layers from the micro-element layer of single-turn coil, the middle layer of single-slot winding, to the macro-layer of the whole phase winding. The insulation weak hot spot areas where the comprehensive insulation damage energy density sequence exceeds the threshold are identified and calibrated.

[0075] S430: For each weak insulation hot spot region, the LP perturbation method is applied to perform an asymptotic expansion of the electro-thermal-mechanical multi-physics coupling degradation equation, retaining the zero-order uniform degradation term and the first-order key coupling term while ignoring higher-order minor terms, and constructing a lightweight physical surrogate function.

[0076] S440: Use a lightweight physical surrogate function to perform point-by-point physical correction on the degradation amplitude of the corresponding weak insulation hotspot region in the preliminary multidimensional degradation feature prediction sequence, and calculate the state transition deviation sequence of the prediction sequence before and after physical correction.

[0077] S450: Feed the state transition bias sequence back to the trend extrapolation algorithm of S300, fine-tune the weighting coefficients and step size adjustment factors, re-execute S300 to obtain the updated preliminary multidimensional degradation feature prediction sequence, perform physical correction of S400 again, and iterate until the state transition bias converges.

[0078] In one possible implementation of step S410, the calculation process for the overall insulation damage energy density further includes: S411: Calculation of Partial Discharge Energy Density. Based on the coefficient of variation of partial discharge amplitude and pulse repetition rate in the preliminary multidimensional degradation feature prediction sequence, combined with the mean partial discharge amplitude collected by S100 and the instantaneous phase voltage at the time of discharge, the energy release of partial discharge within the insulation element at each time step is inverted. The partial discharge energy density is obtained by integrating the product of apparent discharge charge and voltage over the number of discharges and dividing by the volume of the insulation element.

[0079] S412: Calculation of dielectric loss heating power density. Using the third harmonic distortion rate of the leakage current and the fundamental voltage, combined with the empirical relationship between the dielectric loss tangent and the relative permittivity of the insulating material, the dielectric loss heating power at this time step is calculated. Then, based on the regional volume of the optical fiber temperature measurement hotspot distribution, it is converted into power density.

[0080] S413: Calculation of thermomechanical stress shear energy density. Using the effective value of vibration velocity obtained by integrating the broadband vibration acceleration spectrum, combined with the stiffness matrix of the motor winding ends and slot structure, the amplitude of alternating stress induced by vibration is calculated. The mechanical energy dissipation per unit volume per cycle is estimated based on the area of ​​the fatigue stress-strain hysteresis loop of the material. Multiplying this by the number of cycles, jointly characterized by the fundamental vibration frequency and the temperature rise rate, yields the thermomechanical stress shear energy density.

[0081] S414: The partial discharge energy density, dielectric loss heating power density, and thermomechanical stress shear energy density are added together to form a comprehensive insulation damage energy density sequence. The overall insulation damage energy density at each time step Represented as: ; in, For partial discharge energy density, This represents the cumulative energy density of the dielectric loss heating power density over a time step. This represents the thermomechanical stress shear energy density. All three energy densities are normalized under a uniform insulating infinitesimal reference volume.

[0082] In one possible implementation of step S420, the process of three-layer progressive scanning to identify weak hotspot areas in the insulation includes: S421: Single-turn coil micro-element layer scanning. A three-dimensional geometric model and finite element mesh of the motor stator winding are constructed, dividing each turn of the coil into several micro-elements. The comprehensive insulation damage energy density sequence is interpolated and mapped to the center of each micro-element according to the sensor placement location and the heat conduction model. An upper limit threshold for energy density is set, determined by a short-time electro-thermal-mechanical combined aging test of the insulation material. All micro-elements are traversed, and those with a comprehensive energy density exceeding the threshold are marked as high-risk micro-elements.

[0083] S422: Single-slot winding mid-layer scan. Aggregate analysis is performed on all micro-elements in each stator slot. If the proportion of high-risk micro-elements in a continuous region within a slot exceeds a preset ratio, the continuous region is designated as a weak insulation hotspot region within the slot, and the center coordinates, volume range, and average energy density exceeding the standard of the region are recorded.

[0084] S423: Macroscopic layer scan of the entire phase winding. This function integrates information on weak hot spots in each slot at the phase winding level. Based on the macroscopic distribution gradient of the electric field, temperature field, and stress field within the motor, it assesses the potential risk of cascading degradation between hot spots. The final output is a complete list of weak insulation hot spots, each containing its slot number, turn number, micro-element coordinate range, and the maximum value of the comprehensive insulation damage energy density over all time periods.

[0085] In one possible implementation of step S430, the construction process of the lightweight physics proxy function further includes: S431: Write the general form of the electro-thermal-mechanical multiphysics coupling degradation equation. Using the local degradation state variables of the insulation... The dielectric strength retention rate or degree of polymerization of a material element is represented by the degradation equation, which includes thermally driven terms, electric field driven terms, mechanical stress driven terms, and cross-coupling terms between the three terms. It is a set of nonlinear partial differential equations, and the computational cost of solving it directly is extremely high.

[0086] S432: Introducing small parameters Representing the relative strength of the coupling effect, degenerate state variables Temperature field electric field strength Stress field All expanded into The asymptotic series of the equation. Substituting into the degenerate equation, we collect the zeroth-order and first-order terms respectively, and obtain the zeroth-order uniform degenerate equation and the first-order coupled degenerate equation. The zeroth-order uniform degenerate equation only contains the independent degenerate contributions of each physical field, and its form is simplified; the first-order coupled degenerate equation contains pairwise coupling terms such as thermo-electric, thermo-mechanical, and electro-mechanical, but does not contain higher-order coupling terms of three fields or more.

[0087] S433: Ignore For small quantities of order and above, construct a lightweight physical surrogate function. The output of the surrogate function is the degradation magnitude correction of the weak insulation hotspot region at a given time step. The inputs are the preliminary predicted partial discharge energy density, dielectric loss heating power density, thermomechanical stress shear energy density, and local hot spot temperature. The surrogate function has the following linear combination form: ; in, The zero-order uniform degradation amplitude, Thermo-electric coupling coefficient, Thermo-mechanical coupling coefficient, These are the electromechanical coupling coefficients, calibrated by least-squares fitting of the finite element solution results from the digital twin model. The computation of the surrogate function involves only algebraic operations and can be performed in real time on an airborne edge computing platform.

[0088] In one possible implementation of step S440, the calculation process of the physical correction and state transition deviation sequence includes: S441: Extract the local degradation feature subsequence corresponding to a certain weak insulation hotspot region from the preliminary multidimensional degradation feature prediction sequence, input it into the lightweight physical surrogate function, and calculate the physical correction degradation amplitude at each time step. Map the corrected degradation amplitude back to the corresponding component of the multidimensional degradation feature vector, replace the preliminary prediction value, and form the physically corrected multidimensional degradation feature sequence.

[0089] S442: Calculate the state transition deviation sequence of the predicted sequence before and after physical correction. The state transition bias vector at each time step is defined as the difference vector between the physically corrected feature vector and the initially predicted feature vector. The state transition bias sequence is the bias vectors of all time steps within the blind zone arranged chronologically.

[0090] In one possible implementation of step S450, the feedback iteration process includes: S451: The root mean square values ​​of each dimension in the state transition deviation sequence are used as feedback indicators and input into the parameter regulator of the S300 trend extrapolation algorithm. The parameter regulator fine-tunes the weighting coefficient decay factor and step size adjustment factor in the weighted trend extrapolation based on the sign and magnitude of the deviation. If the physical correction generally lowers the hotspot temperature rise rate dimension, the extrapolation weighting coefficient decay factor corresponding to the temperature rise rate feature is reduced to enhance the influence of recent data; if the nonlinearity of the corrected sequence intensifies, the step size adjustment factor is appropriately reduced to shrink the overall time-varying step size.

[0091] S452: After parameter fine-tuning, return to S300. Starting from the original forward and backward boundary degradation feature sequences, re-execute forward and backward trend extrapolation, three-layer consistency verification, and preliminary degradation index mapping to generate an updated preliminary multidimensional degradation feature prediction sequence. Then repeat the physical mapping, hotspot scanning, surrogate function correction, and state transition bias calculation from S410 to S440.

[0092] S453: The iteration termination condition is that the relative change of the L2 norm of the state transition deviation sequence in two consecutive iterations is lower than the preset convergence ratio. When the convergence condition is met, the current preliminary overall insulation degradation index sequence is locked as the benchmark sequence for the overall insulation degradation index after convergence.

[0093] Please refer to Figure 6 The flowchart of S500 in an exemplary lightweight high-voltage high-power aviation motor insulation degradation trend prediction method of this application is shown, and the content includes: S510: Based on the baseline sequence of overall insulation degradation index obtained after S450 convergence, extrapolate to the pre-calibrated insulation failure threshold according to the time change trend of the degradation index, and calculate the remaining operating time from the current moment as the main remaining flight hours.

[0094] S520: Obtain the preliminary overall insulation degradation index sequence after the end of the S300 internal cycle but before the S400 physical correction, and the empirical overall degradation index sequence calculated directly from the front and back boundary degradation characteristic sequence based on the Arrhenius thermal lifetime formula. Extrapolate the failure threshold to obtain two reference remaining flight hours.

[0095] S530: Using the primary remaining flight hours and two reference remaining flight hours to form a three-element sample set, calculate the half median deviation, and generate the remaining service life prediction interval with the primary remaining flight hours as the center and twice the half median deviation as the half width.

[0096] S540: Based on the location of the weak hot spot area of ​​the insulation as specified in S420, determine the lower limit or the median of the output prediction interval as the final remaining service life prediction value.

[0097] In one possible implementation of step S510, the calculation process for the main remaining flight hours includes: S511: The insulation failure threshold is set based on the destructive test of the insulation structure of the same type of motor and relevant airworthiness standards, corresponding to the limit value of the overall insulation degradation index, such as a preset proportion of the remaining dielectric strength dropping to the initial value.

[0098] S512: For the baseline sequence of overall insulation degradation indicators obtained after convergence of S450, a linear extrapolation or power-law model is used to fit the trend of degradation indicators over time. If the sequence shows an accelerating degradation trend, an exponential function model is automatically selected for extrapolation. The predicted time point when the degradation indicators reach the failure threshold is calculated.

[0099] S513: Subtract the current time from the predicted failure time to obtain the remaining flight hours. .

[0100] In one possible implementation of step S520, the calculation process for the remaining flight hours includes: S521: Take the preliminary overall insulation degradation index sequence at the end of the S300 inner cycle, use the same extrapolation model as S512 to predict the time when the failure threshold is reached, and obtain the first reference remaining flight hours. .

[0101] S522: Calculation of empirical overall degradation indices based on the Arrhenius thermal lifetime formula. The average temperature of the fiber optic thermometer hotspots in the quasi-steady-state segments at the forward and backward boundaries is extracted as the reference temperature. The Arrhenius formula describes the exponential relationship between lifetime and temperature, and the activation energy is calibrated by thermal aging tests on the insulation material. The equivalent thermal lifetime consumption is calculated based on the reference temperature and operating time. Combined with the insulation state assessment values ​​at the boundary moments, an empirical degradation index sequence is constructed and extrapolated to obtain the second reference remaining flight hours. .

[0102] In one possible implementation of step S530, the process of generating the remaining useful life prediction interval includes: S531: Main remaining flight hours First reference remaining flight hours Second reference remaining flight hours Form a sample set. Calculate the half median deviation of the sample set, which is defined as the median of the absolute values ​​of the differences between each sample point and the sample median.

[0103] S532: Based on remaining flight hours Using the center of the interval as the upper bound, take twice the median deviation plus half as the upper bound, and twice the median deviation minus half as the lower bound, to form the remaining useful life prediction interval. .

[0104] In one possible implementation of step S540, the output rule for the final remaining useful life prediction value includes: S541: Retrieve the list of weak insulation hotspots output by S420 to determine if there are any hotspots located in critical areas such as the main insulation to ground, inter-turn insulation, or slot openings of the motor. The determination of critical areas is based on the motor insulation structure design drawings and historical failure statistics.

[0105] S542: If there are hotspot areas in key parts, then the lower bound of the prediction interval will be determined. Output the final remaining useful life prediction value to provide a conservative basis for maintenance decisions; otherwise, output the median of the prediction interval. This is to provide the most likely lifespan estimate.

[0106] In this embodiment, through the organic combination of S100 to S500, high-fidelity reconstruction and physical consistency prediction of insulation degradation trend in the blind zone are achieved under the conditions of intermittent missing signals from multiple sources and drastic fluctuations in operating conditions of high-voltage, high-power aviation motors. Furthermore, lightweight and robust estimation of remaining service life is completed under the constraints of airborne computing resources.

[0107] Example 2: This embodiment uses a high-voltage, high-power permanent magnet synchronous motor for the main propulsion of a certain type of multi-electric aircraft as an example to verify the insulation degradation trend prediction method provided by this invention. The motor has a rated power of 250kW, a rated voltage of 540V DC bus, a maximum speed of 12000r / min, and adopts a scattered winding structure. The main insulation is polyimide film-mica tape composite insulation, with a ground withstand voltage level of 3.8kV. The motor has accumulated approximately 4200 flight hours. In the last 150 flight hours, abnormal fluctuations in partial discharge amplitude and intermittent sensor data loss occurred multiple times, triggering the need for insulation degradation trend prediction.

[0108] Flight phase switching tags were extracted from the flight data recorder, covering nine complete takeoff and landing cycles. Partial discharge amplitude and phase sequences, broadband vibration acceleration spectra, distributed fiber optic temperature measurement, and leakage current full harmonic data were simultaneously acquired, all with a unified time base of 0.1 seconds. Based on a joint criterion of flight data recorder tags and operating condition change rate, the pseudo-steady-state operating segments and insulation degradation trend blind zones were obtained, as shown in Table 1.

[0109] Table 1. Results of Quasi-steady-state segment and blind zone identification Five-dimensional degradation features were extracted from the forward and backward quasi-steady-state segments of each blind zone, and dynamic reliability weights were calculated. Weighted trend extrapolation and a three-layer consistency check were used to iteratively fill the blind zone. The inner loop converged after an average of 6 iterations, and the average weighted residual for the entire blind zone was 0.0018. Table 2 compares the mean values ​​of each degradation feature within the blind zone with the mean values ​​of the boundary segments after blind zone filling; the feature change trends are continuous and the physical consistency is good.

[0110] Table 2 Comparison of mean values ​​of multidimensional degradation features The preliminary multidimensional degradation feature prediction sequence generates a preliminary overall insulation degradation index through linear fusion mapping. This index is then converted into a comprehensive insulation damage energy density through physical mapping. The peak energy density along the stator slot number is shown below. Figure 8 As shown, the upper limit threshold for energy density is set at 1.8 × 10⁻⁶. 6 J / m 3 The three-layer progressive scanning identified a weak hot spot in the inter-turn insulation near the slot openings of slots 7 to 9 of phase U.

[0111] Figure 7 In the process, the energy density of slots 7, 8, and 9 exceeded the threshold, and the corresponding hotspot regions were targeted for correction in subsequent iterations. A lightweight physical surrogate function constructed using the LP perturbation method was applied to correct the degradation amplitude in these regions. After three iterations in the outer loop, the state transition bias converged to 0.7%. A comparison of the converged overall insulation degradation index benchmark sequence with the uncorrected sequence and the Arrhenius empirical sequence is shown below. Figure 8 As shown. By Figure 8 As can be seen, the baseline sequence is smoother than the uncorrected sequence within the blind zone and exhibits an accelerated degradation inflection point near the current time, which is consistent with subsequent actual detection results. Based on the converged baseline sequence, the uncorrected sequence, and the Arrhenius empirical sequence, the insulation failure threshold was extrapolated to obtain the remaining service life prediction results, as shown in Table 3.

[0112] Table 3. Predicted Remaining Useful Life Since the identified weak insulation hotspot was located in the critical inter-turn insulation at the slot opening, the method output a prediction range lower limit of 90 flight hours as the final predicted remaining service life. During subsequent monitoring of the motor's operation, when the accumulated flight hours reached 4310 hours, the online partial discharge monitoring system triggered a level one alarm. Endoscopic inspection revealed microcracks and localized discoloration on the insulation surface at the stator winding slot opening, confirming that the insulation condition had entered the maintenance phase. The actual occurrence of the observable insulation anomaly closely matched the lower limit of the prediction range, verifying the effectiveness of the method in situations with missing data and multi-physics coupling, as well as the rationality of conservative decision-making.

[0113] Example 3: A lightweight high-voltage high-power aviation motor insulation degradation trend prediction system, with an interface as shown in Figure 1. Figure 9 , Figure 10 , Figure 11 , Figure 12 and 13 As shown, it contains five interfaces, which are as follows: Real-time status interface: Displays a 3D model of the aircraft motor, with pulsating red spheres marking the winding hot spots and weak insulation areas.

[0114] Multidimensional feature interface: The evolution trend of multidimensional degradation features such as partial discharge and temperature rise rate over time is presented in a colored three-dimensional curved surface.

[0115] Blind zone reconstruction interface: The blue normal segment is compared with the gray blind zone, and a cyan prediction curve is superimposed to show the feature reconstruction results of the missing data segment.

[0116] Physical field coupling interface: The energy density distribution of the electro-thermal-mechanical multi-physics field is represented by a semi-transparent purple sphere as the core, surrounded by an energy ring and hot spot spheres.

[0117] Lifetime range interface: The main forecast and reference remaining flight hours are compared through a 3D bar chart, and the lifetime prediction range is marked with semi-transparent interval blocks.

[0118] The system includes: The quasi-steady-state segment cutting and blind zone calibration module is used to cut the synchronously acquired multi-source signal stream into quasi-steady-state operating segments and calibrate the insulation degradation trend blind zone by using the flight parameter stage switching label and the operating condition change rate joint criterion. The degradation feature extraction and mapping module is used to extract multidimensional degradation feature sequences from the forward and backward quasi-steady-state segments of each insulation degradation trend blind zone, calculate dynamic confidence weights based on the physical consistency criterion, generate complete forward boundary degradation feature sequences and backward boundary degradation feature sequences, and establish a mapping relationship between the multidimensional degradation feature sequences and the overall insulation degradation index. The trend recursion and consistency verification module is used to start from the forward and backward boundary degradation feature sequences, perform forward and backward weighted trend recursion, and generate a preliminary multidimensional degradation feature prediction sequence and a preliminary overall insulation degradation index within the insulation degradation trend blind zone through multi-layer consistency iteration verification. The physical correction module is used to convert the preliminary prediction sequence into a comprehensive insulation damage energy density, identify weak hot spots in the insulation, and use the LP perturbation method to construct a lightweight physical surrogate function to iteratively correct the preliminary multidimensional degradation feature prediction sequence. The life prediction and output module is used to calculate the main remaining flight hours and reference value based on the converged overall insulation degradation index, generate prediction intervals, and adaptively output the final remaining life prediction value according to the location of the weak area.

[0119] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0120] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0121] These computer program instructions are also stored in a computer read-memory that can direct a computer or other programmed data processing device to operate in a particular manner, such that the instructions stored in the computer read-memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart or multiple flowcharts and / or block diagram blocks or multiple block diagrams.

[0122] These computer program instructions are also loaded onto a computer or other programming data processing device to cause a series of operational steps to be performed on the computer or other programming device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programming device, provide steps for implementing the functions specified in the flowchart flow or multiple flows and / or the block diagram blocks or multiple blocks.

[0123] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0124] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors, characterized in that, include: By using the combined criteria of flight parameter stage switching tags and operating condition change rate, the synchronously acquired multi-source signal streams are segmented into quasi-steady-state operation segments and the blind zone of insulation degradation trend is identified. For each blind zone of insulation degradation trend, multidimensional degradation feature sequences are extracted from the forward and backward quasi-steady-state segments. Dynamic reliability weights are calculated based on the physical consistency criterion, and complete forward boundary degradation feature sequences and backward boundary degradation feature sequences are generated. A mapping relationship between the multidimensional degradation feature sequences and the overall insulation degradation index is established. Starting from the forward and backward boundary degradation feature sequences, we perform forward and backward weighted trend recursion, and generate a preliminary multidimensional degradation feature prediction sequence and a preliminary overall insulation degradation index within the insulation degradation trend blind zone through multi-layer consistency iterative verification. The preliminary prediction sequence is converted into a comprehensive insulation damage energy density to identify weak hot spots in the insulation. A lightweight physical surrogate function is constructed using the LP perturbation method to iteratively physical correct the preliminary multidimensional degradation feature prediction sequence. Based on the converged overall insulation degradation index, the remaining flight hours and reference values ​​are calculated, a prediction interval is generated, and the final remaining service life prediction value is output adaptively according to the location of the weak area.

2. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 1, characterized in that, The methods for forming the quasi-steady-state operation segment include: Align multi-source signal streams according to a unified time base; Within a preset sliding time window, the first condition is that the flight parameter stage switching label remains unchanged and the center of the window is not in the flight stage switching transition interval. The second condition is that the mean of the operating condition change rate sequence within the window is lower than the preset mean change rate threshold and the maximum value is lower than the preset peak change rate threshold. Starting from the origin of the multi-source signal stream, the sliding time window is moved with a preset step size, and the continuous time interval that simultaneously satisfies the first and second conditions is marked as a quasi-steady-state operation segment.

3. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 2, characterized in that, The insulation degradation trend blind zone calibration steps include: Identify the operating condition change intervals, which include the preset transition time before and after the flight phase switching time and the time period when the operating condition change rate exceeds the preset change threshold. Identify the intermittent sampling interval of the sensor; the interval in which the sampling interval exceeds a preset multiple of the nominal sampling period and two or more sensors show intermittent sampling at the same time is determined as the intermittent sampling interval of the sensor. Identify data packet loss intervals and mark the intervals in which packet sequence numbers are continuously missing and the time span of the missing segment is greater than a preset multiple of a single sampling period as data packet loss intervals. The union of the operating condition change interval, the sensor intermittent sampling interval, and the data packet loss interval is used to form a blind zone of insulation degradation trend. For each blind zone of insulation degradation trend, the most recent quasi-steady-state operating segment is searched forward along the time axis as the forward quasi-steady-state operating segment, and the most recent quasi-steady-state operating segment is searched backward as the backward quasi-steady-state operating segment.

4. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 1, characterized in that, The multidimensional degradation feature sequence extraction includes: The amplitude variation coefficient and pulse repetition rate of partial discharge are calculated step by step from the amplitude and phase sequence of partial discharge; the proportion of vibration fundamental frequency energy is calculated step by step from the broadband vibration acceleration spectrum; the temperature rise rate of hot spots is calculated step by step from the distribution of hot spots in fiber optic temperature measurement; and the third harmonic distortion rate of leakage current is calculated step by step from the total harmonic of leakage current to ground. The partial discharge amplitude variation coefficient, pulse repetition rate, vibration fundamental frequency energy ratio, hot spot temperature rise rate, and leakage current third harmonic distortion rate at each time step are combined to form a multidimensional degradation feature vector.

5. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aviation motors according to claim 4, characterized in that, The process of calculating dynamic reliability weights and generating complete forward boundary degradation feature sequences and backward boundary degradation feature sequences based on the physical consistency criterion includes: For each dimension of degradation feature component, a physical consistency reference model is established. Based on whether the feature component exceeds the preset reasonable range, the sensor baseline drift trend, and the detection of instantaneous abnormal peaks in the signal, dynamic confidence weights with values ​​between 0 and 1 are generated. The physical consistency reference model is based on the Weibull model of the statistical distribution of partial discharge. Pair the multidimensional degradation feature vector of each time step with the corresponding dynamic confidence weight vector and arrange them in chronological order to form the forward boundary degradation feature sequence; perform the same operation on the backward quasi-steady-state running segment to obtain the backward boundary degradation feature sequence.

6. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 4, characterized in that, The steps for establishing the mapping relationship between the multidimensional degradation feature sequence and the overall insulation deterioration index include: An overall insulation degradation index is constructed using a weighted linear combination. : ,in, For the first Degenerative eigencomponents, These correspond to the coefficient of variation of partial discharge amplitude, pulse repetition rate, proportion of vibration fundamental frequency energy, hot spot temperature rise rate, and third harmonic distortion rate of leakage current, respectively. The fusion coefficient is... For the intercept term; By using historical degradation characteristic data and insulation diagnostic test results, the prediction error is minimized to determine and .

7. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 5, characterized in that, The forward and reverse weighted trend recursion execution steps include: For the forward boundary degradation feature sequence, select the last one closest to the boundary of the insulation degradation trend blind zone. Using the feature data from each time step as the extrapolation modeling window, and with dynamic confidence as the weight, a weighted local linear model is established for each dimension of degradation feature component to predict the first time step within the insulation degradation trend blind zone. eigenvalues ,in The weighted average of time indices within the extrapolation modeling window. For the first The weighted average of the dimensional degenerate eigenvalues. The weighted slope; Add the predicted values ​​to the sequence and move the window to generate a positive prediction sequence point by point; After time reversing the backward boundary degradation feature sequence, the same operation is performed to generate the inverse prediction sequence.

8. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 7, characterized in that, The multi-layer consistency iteration verification includes: The first layer of verification involves performing first-order difference detection on the predicted sequence of each degradation feature component within the blind zone of insulation deterioration trend, and using weighted median filtering to correct predicted oscillation points that exceed the difference threshold. The second layer of verification establishes cross-correlation constraints by utilizing the physical relationship between the hot spot temperature rise rate and the proportion of vibration fundamental frequency energy, and simultaneously corrects the predicted values ​​of the hot spot temperature rise rate and the proportion of vibration fundamental frequency energy. The third layer of verification calculates the weighted Euclidean distance between the forward and reverse prediction sequences and the midpoint of the insulation degradation trend blind zone as the midpoint consistency. When the midpoint consistency exceeds the preset consistency threshold, the order of the trend extrapolation algorithm is adjusted and the fixed step size is switched to a time-varying step size.

9. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 8, characterized in that, The steps for generating the preliminary multidimensional degradation feature prediction sequence and the preliminary overall insulation degradation index within the insulation degradation trend blind zone include: Define the weighted residual of the blind zone of the total insulation degradation trend. ,in This represents the total number of time steps within the insulation degradation trend blind zone. For the degenerate feature dimension, and The first The first time step Forward and backward predicted values ​​of the dimensional degenerate eigencomponents. This corresponds to the dynamic credibility weights; Repeatedly perform forward and reverse trend extrapolation and first to third layer verification until the weighted residual of the blind zone of the full insulation degradation trend is lower than the convergence threshold. Take the weighted average of the forward and reverse prediction sequences as the preliminary multidimensional degradation feature prediction sequence. By utilizing the mapping relationship between the multidimensional degradation feature sequence and the overall insulation degradation index, the feature vectors of each time step of the preliminary multidimensional degradation feature prediction sequence are substituted into... A preliminary sequence of overall insulation degradation indicators was obtained, among which, For the first Degenerative eigencomponents, These correspond to the coefficient of variation of partial discharge amplitude, pulse repetition rate, proportion of vibration fundamental frequency energy, hot spot temperature rise rate, and third harmonic distortion rate of leakage current, respectively. The fusion coefficient is... This is the intercept term.

10. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 1, characterized in that, The steps for converting the preliminary predicted sequence into a comprehensive insulation damage energy density include: The preliminary multidimensional degradation feature prediction sequence was combined with synchronously acquired partial discharge amplitude, fiber optic thermography hotspot values, and effective vibration acceleration values ​​to calculate the partial discharge energy density step by step. Dielectric loss, heating power density, and cumulative energy density and thermomechanical stress shear energy density The combined insulation damage energy density is obtained by superposition. .

11. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 10, characterized in that, The lightweight physical surrogate function performs iterative physical correction steps on the preliminary multidimensional degenerate feature prediction sequence, including: A three-layer progressive scan of the motor insulation structure is performed, consisting of the single-turn coil micro-element layer, the single-slot winding middle layer, and the overall phase winding macro-layer. Micro-element aggregates that exceed the preset upper limit threshold for comprehensive insulation damage energy density are identified as weak insulation hotspots. For each weak insulation hotspot region, a small parameter is introduced to represent the coupling effect strength. The electro-thermal-mechanical multiphysics coupling degradation equation is asymptotically expanded, retaining the zero-order uniform degradation term and the first-order key coupling term, and a lightweight physical surrogate function is constructed. ,in This is the amount of degradation amplitude correction. , , These are temperature, electric field strength, and stress, respectively. The zero-order uniform degradation amplitude, , , These are the thermo-electric coupling coefficient, the thermo-mechanical coupling coefficient, and the electro-mechanical coupling coefficient, respectively.

12. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 11, characterized in that, The iterative physical correction step of the lightweight physical surrogate function for the preliminary multidimensional degenerate feature prediction sequence also includes: The degradation amplitude of the corresponding weak insulation hotspot region in the preliminary multidimensional degradation feature prediction sequence is corrected point by point using a lightweight physical surrogate function to obtain the physically corrected multidimensional degradation feature sequence. Calculate the state transition deviation sequence of the predicted sequence before and after physical correction; The root mean square value of the state transition deviation sequence is fed back into the trend extrapolation algorithm. The weighting coefficient decay factor and step size adjustment factor are finely adjusted, and the forward and reverse trend extrapolation and consistency verification are re-executed. The process is iterated until the L2 norm of the state transition deviation sequence converges, and the converged overall insulation degradation index benchmark sequence is obtained.

13. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 12, characterized in that, The calculation steps for the main remaining flight hours and reference value include: The converged overall insulation degradation index benchmark sequence is extrapolated to the preset insulation failure threshold to obtain the main remaining flight hours; The first reference remaining flight hours are obtained by extrapolating the preliminary overall insulation degradation index sequence without physical correction. The average temperature of the fiber optic thermometer hotspots in the forward and backward boundary degradation feature sequences is extracted. Based on the Arrhenius thermal lifetime formula, the empirical overall degradation index sequence is calculated and extrapolated to obtain the second reference remaining flight hours.

14. The method for predicting the insulation degradation trend of lightweight high-voltage high-power aircraft motors according to claim 13, characterized in that, The steps for obtaining the final remaining useful life prediction value include: The half median deviation is calculated by constructing a sample set using the primary remaining flight hours, the first reference remaining flight hours, and the second reference remaining flight hours. Using the main remaining flight hours as the center, and twice the half-width of the median deviation of half the data, a remaining service life prediction interval is generated; When the calibrated weak hot spot area is located in the main insulation or a critical part of the slot, the lower limit of the prediction interval is output as the final remaining service life prediction value; otherwise, the midpoint of the prediction interval is output.

15. The lightweight high-voltage high-power aviation motor insulation degradation trend prediction system according to any one of claims 1-14, characterized in that the system include: The quasi-steady-state segment cutting and blind zone calibration module is used to cut the synchronously acquired multi-source signal stream into quasi-steady-state operating segments and calibrate the insulation degradation trend blind zone by using the flight parameter stage switching label and the operating condition change rate joint criterion. The degradation feature extraction and mapping module is used to extract multidimensional degradation feature sequences from the forward and backward quasi-steady-state segments of each insulation degradation trend blind zone, calculate dynamic confidence weights based on the physical consistency criterion, generate complete forward boundary degradation feature sequences and backward boundary degradation feature sequences, and establish a mapping relationship between the multidimensional degradation feature sequences and the overall insulation degradation index. The trend recursion and consistency verification module is used to start from the forward and backward boundary degradation feature sequences, perform forward and backward weighted trend recursion, and generate a preliminary multidimensional degradation feature prediction sequence and a preliminary overall insulation degradation index within the insulation degradation trend blind zone through multi-layer consistency iteration verification. The physical correction module is used to convert the preliminary prediction sequence into a comprehensive insulation damage energy density, identify weak hot spots in the insulation, and use the LP perturbation method to construct a lightweight physical surrogate function to iteratively correct the preliminary multidimensional degradation feature prediction sequence. The life prediction and output module is used to calculate the main remaining flight hours and reference value based on the converged overall insulation degradation index, generate prediction intervals, and adaptively output the final remaining life prediction value according to the location of the weak area.

Images