EVTOL rudder health evaluation method fusing life cycle characteristics and multi-dimensional parameters

Abstract

This invention discloses a health assessment method for eVTOL servos that integrates lifecycle characteristics and multidimensional parameters. The method includes: S1, constructing a health assessment parameter index system that integrates a multidimensional monitoring parameter library and an evolutionary feature index library; S2, acquiring sample sequence data of the servo's complete lifecycle and including health scores according to the health assessment parameter index system, constructing a sequence sample dataset of the servo's lifecycle, and inputting it into a periodic distribution health prediction model for model learning and training; S3, collecting real-time sequence data of the eVTOL servo according to the health assessment parameter index system, inputting it into the periodic distribution health prediction model, and outputting the servo's current lifecycle stage and health assessment value. This invention combines monitoring parameters and corresponding evolutionary indexes to accurately identify and assess the health of servos according to their lifecycle aging stages, providing highly reliable technical decision support for aircraft flight safety and predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of eVTOL servo health management, and in particular to an eVTOL servo health assessment method that integrates lifecycle characteristics and multidimensional parameters. Background Technology

[0002] An electric vertical take-off and landing (eVTOL) aircraft is an aircraft that combines the vertical take-off and landing capabilities of a helicopter with the efficient cruise performance of a fixed-wing aircraft. The servo motors within an eVTOL aircraft are the core execution system for flight operations, and their performance directly impacts the aircraft's safety, maneuverability, and reliability. As the core execution unit for attitude control and take-off and landing stability, the servo motors must operate continuously under complex conditions such as frequent starts and stops, varying loads, and high-altitude temperature and humidity fluctuations. This can directly lead to loss of attitude control, take-off and landing deviations, and even crashes. Under complex conditions, key parameters of the aircraft's servo motors are prone to high-frequency random fluctuations. Static thresholds cannot distinguish between "load fluctuations" and "fault signals," resulting in a high false alarm rate. This not only increases maintenance redundancy costs but may also lead to "alarm fatigue," reducing the efficiency of responding to genuine faults. The progression of aircraft servos from health to aging or failure is often a "gradual degradation." Existing static threshold methods have a zero recognition rate for these "micro-evolution signals," often only triggering alarms when the servo jams or torque decays by more than 30%, at which point the maintenance window has been lost, directly threatening flight safety. Furthermore, aircraft servos exhibit individual differences, which existing static thresholds cannot effectively cover, leading to oversensitivity to new servos (over-alarming) and delayed response to aging servos (missed alarms), resulting in insufficient adaptability. To address these technical issues, there is an urgent need for a cutting-edge health or fault assessment technology that can overcome the limitations of static thresholds, possess dynamic operating condition self-adaptation capabilities, and capture micro-degradation signals to meet the core requirements of high reliability and low maintenance costs for aircraft. Summary of the Invention

[0003] The purpose of this invention is to provide a health assessment method for eVTOL servos that integrates life cycle characteristics and multi-dimensional parameters. By combining monitoring parameters with corresponding evolution indicators, the method accurately identifies and assesses the health of servos according to their life cycle aging stages. This provides highly reliable technical decision support for flight safety and predictive maintenance of aircraft, filling the gap in eVTOL servo-specific health monitoring technology.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A health assessment method for eVTOL servos that integrates lifecycle characteristics and multi-dimensional parameters, the method comprising:

[0006] S1. Construct a health assessment parameter index system that integrates a multi-dimensional monitoring parameter library and an evolutionary characteristic index library. The multi-dimensional monitoring parameter library consists of monitoring parameter items associated with health faults of eVTOL servos, and the evolutionary characteristic index library consists of evolutionary index items of the monitoring parameter items.

[0007] S2. Construct a periodic distribution health prediction model divided by life cycle stages. Obtain sample sequence data of the complete life cycle of the servo motor and including health scores according to the health evaluation parameter index system to construct a sequence sample dataset of the servo motor life cycle. The periodic distribution health prediction model uses the sequence sample dataset for model learning and training.

[0008] S3. Collect real-time sequence data of the eVTOL servo motor in real time according to the health assessment parameter index system, input the data into the periodic distribution health prediction model, and output the life cycle stage and health assessment value of the servo motor.

[0009] To better realize the present invention, the life cycle stage is divided into four stages in sequence: no aging, mild aging, moderate aging, and severe aging. The adjacent stages in the life cycle stage are the mutation evolution feature intervals. The periodic distribution health prediction model uses the sequence sample dataset to extract and identify the mutation evolution feature intervals of the life cycle stage and the adjacent stages. The periodic distribution health prediction model uses the sequence sample dataset and health score labels to evaluate and train the model.

[0010] Preferably, the data sources for the monitoring parameters in the multi-dimensional monitoring parameter library include collected eVTOL servo operating data and / or monitoring data of the servo obtained through a servo monitoring device. The monitoring parameters in the multi-dimensional monitoring parameter library include output angle deviation, drive current fluctuation value, torque feedback value, response delay time and / or gearbox vibration frequency.

[0011] Preferably, the evolutionary indicator items in the evolutionary feature indicator library include time-dimensional evolutionary indicators, correlation indicators, and / or statistical dimension indicators. The time-dimensional evolutionary indicators include mean, mean change, median, median change, trend slope, and / or trend slope change. The correlation indicators include correlation matrix, covariance matrix, and / or mutual information entropy. The statistical dimension indicators include skewness, skewness change, kurtosis, kurtosis change, and / or quantile mobility of the data distribution statistics of the monitoring parameter items.

[0012] Preferably, in method S1, an evolutionary feature index calculation module corresponding to the evolutionary feature index library is constructed. The evolutionary feature index calculation module calculates and obtains the data of the monitoring parameter items according to the global time span and the time span window, respectively, according to the evolutionary index item. The sequence data of the evolutionary index item is output sequentially according to the time span window, and each time span window contains the local data corresponding to the evolutionary index item.

[0013] Preferably, the periodic distribution health prediction model is constructed with a dynamic weight evaluation mechanism, and the dynamic weight evaluation mechanism of the periodic distribution health prediction model is trained by evaluating the dynamic weights in stages according to the life cycle stage.

[0014] This invention provides a first method for health early warning notification as follows: The health assessment value P of the periodic distribution health prediction model is a standard percentage score. The periodic distribution health prediction model makes the following judgments on the health assessment value P and outputs a health early warning notification:

[0015] If the health assessment value P is in Within the specified range, a Level 3 health warning is issued; the health assessment value P is within the specified range. Within the specified range, a Level 2 health warning is issued; the health assessment value P is within the specified range. Within the specified range, a Level 1 health warning is issued; the health assessment value P is less than... It outputs an emergency shutdown warning.

[0016] The present invention provides a second method for health early warning notification as follows:

[0017] S5. The health assessment value P of the periodic distribution health prediction model is a standard score on a percentage scale, and health warnings are output in stages according to the following method:

[0018] If the periodic distribution health prediction model determines that the servo motor is in the non-aging stage and the health assessment value P≤A1, then a health warning will be output.

[0019] If the servo motor is in a mild aging stage, the health assessment value P is at... Within the specified range, a Level 3 warning is issued; the health assessment value P is at... Within the specified range, a Level II warning is issued; the health assessment value P is within the specified range. Within the specified range, issue a Level 1 warning;

[0020] If the servo motor is in the moderate aging stage, and the health assessment value P is at... Within the specified range, a Level 3 warning is issued; the health assessment value P is at... Within the specified range, a Level II warning is issued; the health assessment value P is within the specified range. Within the specified range, issue a Level 1 warning;

[0021] If the servo motor is in a severely aged stage, the health assessment value P is at... Within the specified range, a Level 3 warning is issued; the health assessment value P is at... Within the specified range, a Level II warning is issued; the health assessment value P is within the specified range. Within the specified range, issue a Level 1 warning.

[0022] Preferably, the sample sequence data is obtained by jointly collecting eVTOL servo aging test data and eVTOL servo actual flight data during the complete life cycle stage; both the sample sequence data and the real-time sequence data are processed by Kalman filtering algorithm for noise reduction.

[0023] Preferably, the evolutionary characteristic index calculation module sets an interference filtering mechanism to filter out abnormal interference data in the health assessment parameter index system, and then uses the Gaussian process regression algorithm (GPR) to perform trend curve regression fitting.

[0024] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0025] (1) This invention constructs a health assessment parameter index system that integrates monitoring parameter items and monitoring parameter items. The periodic distribution health prediction model uses a sequence sample dataset containing the complete life cycle of the servo motor and health scores to train the model for the life cycle stage and health assessment value. The key time series data of the servo motor monitoring parameter items are collected in real time and input into the periodic distribution health prediction model, which can accurately identify and assess the life cycle stage and health assessment value of the servo motor. This provides highly reliable technical decision support for the flight safety and predictive maintenance of eVTOL servo motors and fills the gap in eVTOL servo motor-specific health monitoring technology.

[0026] (2) The present invention combines the monitoring parameter items and the evolution index items corresponding to the monitoring parameter items to accurately identify and assess the health of the servo motor according to the aging stage of the life cycle. The periodic distribution health prediction model has the ability to identify multi-dimensional monitoring parameter items and their associated coupling features such as time sequence features, distribution features, and trend features. It can identify the slight deviation features of the life cycle stage and health assessment value, and realize the feature identification, health assessment and health risk classification early warning of early aging signals.

[0027] (3) This invention can adapt to various health and aging cycle stages of eVTOL servo motors, capture micro aging signals and health assessment feature signals of 5%-8% performance degradation of servo motors, realize accurate identification and assessment of life cycle stages and health assessment values, provide early warning of health risk levels, provide sufficient window period for predictive maintenance, and reduce the risk of flight accidents caused by servo motor failures; integrate deep learning and multi-feature attention mechanism to output the probability distribution of health status and maintenance suggestions, and achieve the goal of intelligent closed-loop management "from monitoring to decision". Attached Figure Description

[0028] Figure 1 This is a flowchart of the eVTOL servo health assessment method of the present invention;

[0029] Figure 2 This is an example of the output angle deviation distribution diagram of the servo motor at different health stages;

[0030] Figure 3 for Figure 2 A comparative diagram showing the distribution of output angle deviation across different health stages of the servo motor;

[0031] Figure 4 As an example in the embodiment, the skewness feature map of the output angle deviation statistics is shown.

[0032] Figure 5 The example outputs a peak feature map of the angle deviation statistics. Detailed Implementation

[0033] The present invention will be further described in detail below with reference to embodiments:

[0034] Example

[0035] like Figure 1 As shown, an eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters is presented. The method includes:

[0036] S1. Construct a health assessment parameter index system that integrates a multi-dimensional monitoring parameter library and an evolutionary characteristic index library. The multi-dimensional monitoring parameter library contains monitoring parameter items related to health faults of the eVTOL servo, and the evolutionary characteristic index library contains evolutionary index items of the monitoring parameter items. Preferably, the data sources for the monitoring parameter items in the multi-dimensional monitoring parameter library include collected eVTOL servo operating data and / or monitoring data obtained through servo monitoring devices (including various monitoring sensors). The monitoring parameter items in the multi-dimensional monitoring parameter library include output angle deviation, drive current fluctuation value, torque feedback value, response delay time, and / or gearbox vibration frequency. The output angle deviation is the difference between the angle of the target deflection angle command output by the eVTOL servo and the actual deflection angle (it is the actual deviation of the eVTOL servo's controlled deflection angle). The servo monitoring device (such as an angle sensor) monitors the actual deflection angle and then calculates the difference (i.e., the output angle deviation) between it and the angle of the target deflection angle command. The drive current fluctuation value is the fluctuation difference between the eVTOL servo and the current at different times (or the difference between the actual operating current and the rated stable current can be used instead), which is obtained and calculated by a current monitoring device (such as a current recorder). The torque feedback value is the output torque of the eVTOL servo, which is monitored and obtained by a torque sensor (or it can be calculated from current data). The response delay time is the time interval between the eVTOL servo issuing a deflection command and the actual effective deflection, which can be obtained by a combination of an angle sensor (which monitors the actual effective deflection) and a timer (used to time the time interval). The gearbox vibration frequency is the vibration frequency of the eVTOL servo's gearbox, which is monitored by a vibration sensor installed on the eVTOL servo's gearbox.

[0037] In some embodiments, the evolutionary indicator items in the evolutionary feature indicator library include time-dimensional evolutionary indicators, correlation indicators (including correlation matrix, covariance matrix and / or mutual information entropy, where the correlation matrix is ​​the correlation matrix obtained by training each monitoring parameter item on sample data, and similarly, the covariance matrix and mutual information entropy are obtained) and / or statistical dimension indicators (statistically calculating the curves of each monitoring parameter item's data over the global time span and time span window, and obtaining skewness evolution, kurtosis evolution, and quantile mobility, respectively; statistical dimension indicators can effectively capture slight distortions in the distribution pattern). The time-dimensional evolution indicators include the mean (calculated using all data within the global time span when calculating the global mean; and the moving average within a local time span window when calculating the local mean), mean change, median (calculated using all data within the global time span when calculating the global median; and the moving average within a local time span window when calculating the local median), median change, and trend slope (calculated using all data within the global time span when calculating the global median). When calculating the trend slope, the trend slope is calculated using all data across the global time span; for the local trend slope within a time span window, the local trend slope is calculated using all data within that window, or / and the change in trend slope. Correlation indicators include the correlation matrix, covariance matrix, and / or mutual information entropy. Statistical dimension indicators include the skewness of the distribution statistics of the monitoring parameter items (when calculating the global skewness across the global time span, the global skewness is calculated using all data across the global time span; for the local skewness within a time span window, the local skewness is calculated using all data within that window). The invention obtains local skewness, skewness variation, kurtosis (when calculating global kurtosis over a global time span, all data within the global time span are used to calculate global kurtosis; in local kurtosis within a time span window, all data within the time span window are used to calculate local kurtosis), kurtosis variation, and / or quantile mobility (when calculating global quantile mobility over a global time span, all data within the global time span are used to calculate global quantile mobility; in local quantile mobility within a time span window, all data within the time span window are used to calculate local quantile mobility). Preferably, the invention constructs an evolutionary feature index calculation module corresponding to the evolutionary feature index library. The evolutionary feature index calculation module calculates and obtains the data of the monitoring parameter items according to the evolutionary index items based on the global time span and the time span window, and outputs the sequence data of the evolutionary index items sequentially according to the time span window. Each time span window contains the local data corresponding to the evolutionary index item.In some embodiments, the evolutionary characteristic index calculation module sets an interference filtering mechanism to filter out abnormal interference data in the health assessment parameter index system (the interference filtering mechanism judges, identifies and filters out abnormal interference data), and then uses the Gaussian process regression algorithm (GPR) to perform trend curve regression fitting to identify micro-evolutionary trends; for example, if the evolution curve of the power battery voltage average shows a linear decrease for 15 consecutive windows and the covariance curve with the internal resistance shows an exponential increase, it is determined to be an aging-driven distribution evolution.

[0038] Taking the output angle deviation as an example of a servo motor monitoring parameter, the statistical curve of the output angle deviation data over the global time span is as follows: Figure 2 As shown, in the non-aging stage, its output angle deviation curve distribution is symmetrical, and its skewness is close to 0 (see...). Figure 2 (See the upper curve); in the middle aging stage, the skewness gradually increases (right skewness increases, and positive extreme errors begin to increase; see also...) Figure 4 See also Figure 2 The middle curve; in the severe aging stage (failure degradation period), see [reference needed]. Figure 4 The deviation increases rapidly (strong right deviation, frequent large positive errors, reflecting a tendency for the servo to "stick"). Figure 3 Given Figure 2 A comparative diagram of the output angle deviation distribution. Figure 2 Three scenarios were given: in the four stages of no aging, mild aging, moderate aging, and severe aging, the output angle deviation curve will show different curve characteristics.

[0039] S2. Construct a periodic distribution health prediction model divided by life cycle stages. Obtain sample sequence data containing health scores for the complete life cycle of the servo motor (the complete life cycle is defined as the entire life cycle from factory use to the end of severe aging, including non-aging, mild aging, moderate aging, and severe aging) according to the health assessment parameter index system. Construct a sequence sample dataset for the servo motor life cycle. The sample sequence data is obtained by jointly collecting eVTOL servo motor aging test (ALT, Accelerated Life Test) and eVTOL servo motor actual flight data for the complete life cycle stages. Preferably, at least 10 complete life cycle data are collected. The sample sequence data is mainly based on eVTOL servo motor aging test data and supplemented by eVTOL servo motor actual flight data (the eVTOL servo motor actual flight data is used to enrich the sample sequence data for the life cycle stages). The periodic distribution health prediction model uses the sequence sample dataset for model learning and training. In some embodiments, the life cycle is divided into four consecutive stages: non-aging, mild aging, moderate aging, and severe aging. Adjacent stages within the life cycle are considered mutational evolutionary feature intervals. The periodic distribution health prediction model uses a sequence sample dataset to extract and identify features of these mutational evolutionary feature intervals within the life cycle stages and between adjacent stages. The periodic distribution health prediction model uses the sequence sample dataset and health score labels for model evaluation and training. This invention's periodic distribution health prediction model integrates a temporal attention mechanism to focus on the temporal features of evolutionary indicators, and also integrates a multi-dimensional attention mechanism to jointly focus on the temporal features of multiple monitoring parameters and evolutionary indicators. In life cycle stages and health assessments, it comprehensively identifies multiple features, including monitoring parameters, evolutionary indicators, and temporal relationships, improving the accuracy of the model's life cycle segmentation and health assessment.

[0040] In obtaining health assessment values, this invention uses a consistent percentage-based standard score for all lifecycle stages. The health assessment value obtained at a specific point in time is the result of the health assessment at that specific point in time within the lifecycle stage. The health assessment results have comparability across different time points in the lifecycle stage (the health assessment results quantitatively evaluate the health status of the servo at all time points, and the health assessment results at different time points can be directly compared quantitatively). For example, the health assessment value in the non-aging stage is 98, and the health assessment value in the mild aging stage is 89 (lower than 98 in the non-aging stage, indicating a decrease in servo health). For example, with an angle deviation threshold of ±1.0° and a current threshold of ≤5A, when the servo enters the mild aging stage (5% gear wear), the average angle deviation during the cruise phase increases from 0.2° to 0.3°, and the average current increases from 2A to 2.2A. When the servo enters severe aging (20% gear wear), the angle deviation instantaneously reaches 1.1° during a certain takeoff phase, indicating that the servo has reached a dangerous state and would be difficult to recover if it is currently performing a mission. For example, during the mild aging stage of the servo motor, if the cruise angle is detected to drift from 0.2° to 0.3° (drift amount +50%), and the correlation coefficient between torque feedback and current drops from 0.1 to 0.04 (-60%), and the model output health assessment threshold is 88, triggering a level 3 warning, ground maintenance will complete gear lubrication maintenance and record it in the database before the next flight. If the servo motor subsequently enters the moderate aging stage (e.g., health assessment threshold 68), the flight control system will automatically limit the load to ensure flight safety and arrange a replacement plan to completely prevent the fault from escalating.

[0041] In some embodiments, the periodic distribution health prediction model is constructed with a dynamic weight evaluation mechanism. The dynamic weight evaluation mechanism of the periodic distribution health prediction model is trained and evaluated in stages according to the life cycle stage. The dynamic weight evaluation mechanism of the periodic distribution health prediction model adopts different weight parameters in different life cycle stages. When it is determined that the servo is in a certain life cycle stage, the dynamic weight evaluation mechanism activates the weight parameters of that life cycle stage accordingly (when the life cycle stage of the servo is identified, the dynamic weight evaluation mechanism dynamically adopts the appropriate weight parameters according to the life cycle stage. Different life cycle stages are set with a dynamic adaptation mechanism through the dynamic weight evaluation mechanism, which can better adapt to the characteristics of the life cycle stage and make accurate evaluation of health results) to obtain the corresponding health evaluation value.

[0042] S3. Real-time sequence data of the eVTOL servo is collected according to the health assessment parameter index system and input into the periodic distribution health prediction model to output the servo's life cycle stage and health assessment value. Preferably, the sample sequence data and real-time sequence data of methods S1 to S2 of this invention are all processed by Kalman filtering algorithm to ensure that the data signal-to-noise ratio is >40dB.

[0043] This invention provides two health early warning methods. The first health early warning method is as follows: The health assessment value P of the periodic distribution health prediction model is a percentage standard score. The periodic distribution health prediction model makes the following judgment on the health assessment value P and outputs a health early warning: If the health assessment value P is within a certain range... Within the specified range, a Level 3 health warning is issued. The health assessment value P is within the specified range. Within the specified range, a level-two health warning is output, and maintenance plan suggestions can be generated. The health assessment value P is within... Within the specified range, a Level 1 health warning will be issued, initiating an emergency maintenance procedure to replace the component within 24 hours. The health assessment value P is less than... The system outputs an emergency shutdown warning. The first health warning method mainly relies on health assessment values ​​to divide different levels of health warning intervals and then provides health warnings and emergency shutdown warnings. This can be linked to the equipment control system to reduce load or shut down, thus avoiding safety risks.

[0044] The output angle deviation curve for the statistical time span window range, such as... Figures 4-5 As shown, the skewness characteristics change within a certain time span window as follows: Figure 4 As shown, during the healthy phase, the skewness is close to 0 (symmetrical distribution); during the transition phase of the warning (i.e., the healthy warning prompt phase), the skewness gradually increases (right skew strengthens, and positive extreme errors begin to increase); during the emergency stop warning prompt phase (i.e., the pre-fault phase), the skewness increases rapidly (strong right skew, frequent occurrence of large positive errors, reflecting the tendency of the servo to "stick"). The peak characteristic changes within a certain time span window are as follows: Figure 5As shown, in the healthy phase, the excess kurtosis is close to 0 (approaching a normal distribution with thin tails); in the transitional phase of the warning (i.e., the healthy warning stage), the kurtosis slowly increases (the distribution begins to steepen, and the probability of extreme values ​​increases); in the emergency shutdown warning stage (i.e., the pre-fault stage), the kurtosis increases significantly (sharp peaks and thick tails, errors are concentrated in a specific range and extreme values ​​occur frequently, and wear leads to a decrease in error stability). Analysis of quantile characteristic changes within a certain time span window shows that in the healthy phase, the mobility is low (approximately 0.2-0.3, individual errors are stable within the quartile interval, rarely moving across intervals); in the transitional phase of the warning (i.e., the healthy warning stage), the mobility increases (approximately 0.4-0.5, error fluctuations increase, and individuals move frequently within the interval); in the emergency shutdown warning stage (i.e., the pre-fault stage), the mobility increases sharply (exceeding 0.6, error stability collapses, individuals fluctuate violently between high and low intervals, indicating an impending fault). During the health assessment process, different health scores will also exhibit different curve characteristics (skewness, peak value, and quantile are features of health assessment, which facilitate model learning and training). This embodiment takes the output angle deviation curve as an example, which can effectively extract the skewness, peak value, quantile and other feature data of the output angle deviation curve, making it easier to achieve accurate division and assessment of the cycle stage and health assessment.

[0045] The present invention provides a second health early warning method as follows:

[0046] S5. The health assessment value P of the periodic distribution health prediction model is a percentage-based standard score. Health warnings are output in stages according to the following method: If the periodic distribution health prediction model determines that the servo is in the non-aging stage and the health assessment value P ≤ A1, then a health warning is output. If the servo is in the mild aging stage and the health assessment value P is within the range of A1, then a health warning is output. Within the specified range, a Level 3 warning is issued; the health assessment value P is at... Within the specified range, a Level II warning is issued; the health assessment value P is within the specified range. Within the specified range, a Level 1 warning is issued. If the servo motor is in a moderate aging stage, and the health assessment value P is within the specified range... Within the specified range, a Level 3 warning is issued; the health assessment value P is at... Within the specified range, a Level II warning is issued; the health assessment value P is within the specified range. Within the specified range, a Level 1 warning will be issued. If the servo motor is in a severely aged stage, and the health assessment value P is within the specified range... Within the specified range, a Level 3 warning is issued; the health assessment value P is at... Within the specified range, a Level II warning is issued; the health assessment value P is within the specified range. Within the specified range, a Level 1 warning is issued. The second health warning method divides the health warning range into different levels according to the life cycle stage (including non-aging, mild aging, moderate aging, and severe aging), and outputs warnings at different levels by combining the life cycle stage with the health assessment value.

[0047] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for health assessment of eVTOL servos that integrates lifecycle characteristics and multi-dimensional parameters, characterized in that: The methods include: S1. Construct a health evaluation parameter index system that integrates a multi-dimensional monitoring parameter library and an evolutionary characteristic index library. The multi-dimensional monitoring parameter library contains monitoring parameter items related to health faults of eVTOL servos. The monitoring parameter items in the multi-dimensional monitoring parameter library include output angle deviation, drive current fluctuation value, torque feedback value, response delay time, and gearbox vibration frequency. The evolutionary feature index library consists of evolutionary indexes for monitoring parameters. These indexes include time-dimensional evolutionary indicators, correlation indicators, and statistical indicators. The time-dimensional evolutionary indicators include mean, mean change, median, median change, trend slope, and trend slope change. The correlation indicators include correlation matrix, covariance matrix, and mutual information entropy. The statistical indicators include skewness, skewness change, kurtosis, kurtosis change, and quantile mobility of the monitoring parameter data distribution statistics. S2. Construct a periodic distribution health prediction model divided by life cycle stages. Obtain sample sequence data containing health scores for the complete life cycle of the servo motor according to the health evaluation parameter index system to construct a sequence sample dataset of the servo motor life cycle. The life cycle stages are divided into four consecutive stages: no aging, mild aging, moderate aging, and severe aging. The adjacent stages in the life cycle stage are the mutation evolution feature intervals. The periodic distribution health prediction model uses the sequence sample dataset to extract and identify the features of the mutation evolution feature intervals of the life cycle stage and adjacent stages. The periodic distribution health prediction model uses the sequence sample dataset for model learning and training. S3. Collect real-time sequence data of the eVTOL servo motor in real time according to the health assessment parameter index system, input the data into the periodic distribution health prediction model, and output the life cycle stage and health assessment value of the servo motor.

2. The eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters according to claim 1, characterized in that: In method S1, an evolutionary feature index calculation module corresponding to the evolutionary feature index library is constructed. The evolutionary feature index calculation module calculates and obtains the data of the monitoring parameter items according to the global time span and the time span window, respectively, according to the evolutionary index item. The sequence data of the evolutionary index item is output sequentially according to the time span window, and each time span window contains the local data corresponding to the evolutionary index item.

3. The eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters according to claim 1, characterized in that: The periodic distribution health prediction model is equipped with a dynamic weight evaluation mechanism, which evaluates and trains the dynamic weights in stages according to the life cycle stage.

4. The eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters according to claim 1, characterized in that: The health assessment value P of the periodic distribution health prediction model is a standard score on a percentage scale. The periodic distribution health prediction model makes the following judgments on the health assessment value P and outputs a health warning prompt: If the health assessment value P is in Within the specified range, a Level 3 health warning will be issued; Health assessment value P is at Within the specified range, a Level 2 health warning will be issued; Health assessment value P is at Within the specified range, a Level 1 health warning will be issued; Health assessment value P less than It outputs an emergency shutdown warning.

5. The eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters according to claim 1, characterized in that: It also includes the following methods: S5. The health assessment value P of the periodic distribution health prediction model is a standard score on a percentage scale, and health warnings are output in stages according to the following method: If the periodic distribution health prediction model determines that the servo motor is in the non-aging stage and the health assessment value P≤A1, then a health warning will be output. If the servo motor is in a mild aging stage, the health assessment value P is at... Within the specified range, a Level 3 warning will be issued; Health assessment value P is at Within the specified range, issue a Level II warning; Health assessment value P is at Within the specified range, issue a Level 1 warning; If the servo motor is in the moderate aging stage, and the health assessment value P is at... Within the specified range, a Level 3 warning will be issued; Health assessment value P is at Within the specified range, issue a Level II warning; Health assessment value P is at Within the specified range, issue a Level 1 warning; If the servo motor is in a severely aged stage, the health assessment value P is at... Within the specified range, a Level 3 warning will be issued; Health assessment value P is at Within the specified range, issue a Level II warning; Health assessment value P is at Within the specified range, issue a Level 1 warning.

6. The eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters according to claim 1, characterized in that: The sample sequence data is obtained by jointly collecting eVTOL servo aging test data and eVTOL servo actual flight data during the complete life cycle stage; both the sample sequence data and the real-time sequence data have been processed by Kalman filtering algorithm for noise reduction.

7. The eVTOL servo health assessment method integrating lifecycle characteristics and multi-dimensional parameters according to claim 2, characterized in that: The evolutionary characteristic index calculation module sets up an interference filtering mechanism to filter out abnormal interference data in the health assessment parameter index system, and then uses the Gaussian process regression algorithm (GPR) to perform trend curve regression fitting.

Images