A multi-source uncertainty adaptive life prediction method for quadrotor unmanned aerial vehicle

Abstract

The application discloses a kind of multi-source uncertainty under quadrotor unmanned aerial vehicle adaptive life prediction method, by constructing the degradation model of each actuator, the EM algorithm is used to update the degradation model parameter, and finally the probability density function of residual life is calculated based on the updated parameter;Wherein model parameter updating includes the following iteration: E step, determine state space model, estimate the mean and variance of drift parameter based on Kalman filtering;The conditional expectation corresponding to the complete data set log-likelihood function is calculated according to the mean and variance of drift parameter using RTS smoothing algorithm;M step, update parameters by searching algorithm to maximize the profile log-likelihood function.The application effectively improves the modeling and residual life prediction accuracy of quadrotor unmanned aerial vehicle under multi-source uncertainty, enhances maintenance pertinence, prolongs the service life of body, and guarantees the stable operation of unmanned aerial vehicle in complex environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) maintenance technology, and more specifically to an adaptive life prediction method for quadrotor UAVs under multi-source uncertainty. Background Technology

[0002] With the widespread application of drone technology, quadcopter drones play an important role in aerial photography, logistics, inspection, and other fields. However, during actual flight, the degradation of the actuators can severely affect the performance and lifespan of drones, and they also face multiple uncertainties.

[0003] Traditional UAV lifespan prediction and maintenance methods often fail to adequately consider the multi-source uncertainties in actuator degradation processes, such as measurement noise, scarcity of degradation data, and maintenance interventions. When modeling actuator degradation, they do not sufficiently consider the correlations between actuators, making it difficult to accurately describe the degradation process. Furthermore, existing methods lack adaptive capabilities in updating model parameters, failing to adjust the model promptly based on real-time monitoring data, resulting in low lifespan prediction accuracy and poor targeting and effectiveness of maintenance measures.

[0004] Due to the lack of effective handling of multi-source uncertainties, existing methods cannot accurately estimate the degradation state and parameters of actuators, resulting in large errors in the prediction of the remaining lifespan of UAVs and making it difficult to implement maintenance measures at the appropriate time. This not only affects the normal operation of UAVs but may also lead to mission failure, increasing operating costs and safety risks. Summary of the Invention

[0005] In view of this, the present invention provides an adaptive life prediction method for quadrotor UAVs under multi-source uncertainty. By introducing Kalman filtering, RTS and EM algorithms, the method realizes real-time updating of the parameters of the degradation model of each actuator and the comprehensive degradation model, thereby improving the modeling and remaining life prediction accuracy of the UAV and enhancing the maintenance effect.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] First, an adaptive lifetime prediction method for quadrotor UAVs under multi-source uncertainties, including: Construct degradation models for each actuator; Update the parameters of the degenerate model using the EM algorithm, including performing the following iterations: E-step: Determine the state-space model and estimate the mean and variance of the drift parameters based on Kalman filtering; The RTS smoothing algorithm is used to calculate the corresponding conditional expectation based on the mean and variance of the drift parameters, which is then used to obtain the conditional expectation of the log-likelihood function of the complete dataset. M-step: Update the parameters by maximizing the number of likelihood functions of the cross-section using a search algorithm; The remaining lifetime probability density function is calculated based on the updated parameters.

[0008] Second, an autonomous maintenance strategy for quadcopter UAVs, including: The remaining lifespan of a quadrotor UAV under multi-source uncertainty is predicted using any of the above-described adaptive lifespan prediction methods for UAVs. When the predicted remaining lifespan needs to be extended, relevant parameters are adjusted based on the LF-MPC life extension mechanism to implement autonomous maintenance control.

[0009] Third, an adaptive life prediction system for quadrotor UAVs under multi-source uncertainty, applying any of the above-described adaptive life prediction methods for quadrotor UAVs under multi-source uncertainty, including: The model building module is used to construct the degenerate model of each actuator and determine the state space model. The parameter update module is used to estimate the mean and variance of the drift parameters in the state-space model based on Kalman filtering, and to update the parameters of the degenerate model using the EM algorithm. The lifetime prediction module calculates the remaining lifetime probability density function based on the updated parameters.

[0010] As can be seen from the above technical solution, this invention discloses an adaptive lifetime prediction method for quadrotor UAVs under multi-source uncertainty, which, compared with the prior art, This invention considers multi-source uncertainties and uses Kalman filtering, RTS, and EM algorithms to update model parameters in real time, making the prediction results of the remaining lifespan of UAVs more accurate and better able to fit the actual situation.

[0011] The method of this invention can adapt to various uncertainties during the operation of UAVs and adjust the model and maintenance strategy in a timely manner to ensure the stable operation of UAVs in complex environments, improve their reliability and safety, and reduce operating costs.

[0012] Accurate lifespan prediction provides a reliable basis for maintenance decisions. Autonomous maintenance control based on prediction results can more precisely determine maintenance intervention time, adjust relevant parameters, improve the pertinence of maintenance measures, and extend the service life of the airframe. Compared with methods that do not employ parameter adaptation, this invention can significantly extend the lifespan of the UAV airframe. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0014] Figure 1 Flowchart for autonomous maintenance of a quadcopter drone; Figure 2 Adaptive graph update for degradation parameters; Figure 3 To synthesize the degradation amount and the degradation amount change curves of each actuator; Figure 4 The graph is updated adaptively to reflect the parameters of the comprehensive degradation model. Figure 5 Update the health status change graph for the parameters; Figure 6 Map showing the distribution of quadcopter UAVs with RUL (Radio UAV) coverage; Figure 7 The response curve of a quadcopter UAV when updating actuator degradation parameters; Figure 8 This is a map showing the RUL distribution of a quadcopter UAV after parameter adaptive maintenance. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] Example 1: This embodiment discloses an adaptive lifetime prediction method for quadrotor UAVs under multi-source uncertainties, mainly including actuator degradation modeling, parameter updating, and lifetime prediction: First, actuator degradation modeling: This application first constructs degradation models for each actuator and determines the state-space model; the modeling steps include: The actuator degradation drift coefficients are described as random walk models, and then each actuator degradation model is modeled using a Wiener process driven by standard Brown motion, described as follows: (1) In the formula: Indicates the drift coefficient. is the diffusion coefficient. It is an unknown parameter Regarding time The nonlinear function, using Characterizing the time-varying and nonlinear properties of brushless DC motor degradation; parameters and These are used to determine the rate and shape of nonlinear degradation, respectively.

[0018] In order to estimate the actuator degradation parameters, at the monitoring time Discretize the degradation model to obtain the implicit degradation state sequence of each actuator. At this point, the state-space model of the Wiener process degradation model can be expressed as: (2) In the formula, and Let these represent the drift coefficients of the m-th actuator in the k-th and (k-1)-th iterations, respectively. This represents the effect of other actuators on the degradation of the m-th actuator in the k-th iteration. , and These represent the observation data of the m-th actuator in the k-th and (k-1)-th iterations, respectively. Indicates the diffusion coefficient. The random error term of Brown's motion at time k, discretized by the m-th actuator, has a mean of 0 and a variance of . The normal distribution reflects the stochastic fluctuation characteristics of degradation driven by the Wiener process, and is a concrete manifestation of the stochastic factors in the model. It concerns the m-th actuator at time k with respect to unknown parameters. A difference function, by and The difference is used to obtain the discretized state-space model, which correlates the degenerate state of the actuator with the drift coefficient at adjacent time points, characterizing the cumulative change of the degradation process over time in the model. Wherein: , , , Used to describe the Brownian motion corresponding to the m-th actuator at time t. Similarly, the state value, Is the Brownian motion at time... The state value. It is a trend term used to accumulate the degradation process of the m-th actuator. It is the core trend function of the m-th actuator degradation, describing the degradation process over time. Unknown parameters The nonlinear laws of change.

[0019] Second, parameter update: 1. As can be seen from the state-space model in equation (2), the drift parameter is treated as a hidden state and can only be determined based on historical degradation information. Make an estimate. Because... It follows a Gaussian distribution, therefore it can be obtained by... The recursive filtering estimation, i.e., once the degraded observation data is obtained... Then, the mean can be obtained using the Kalman recursive filtering algorithm. and .

[0020] It is a proportional parameter that controls the nonlinear degradation rate of the actuator, and its initial value is And satisfy , It is the drift parameter of the m-th actuator. The posterior variance is estimated, and the estimation steps include: initialization , ; The state at time step is estimated as follows: , ; Calculate the Kalman gain: ; Update drift parameter estimates:

[0021] The updated variance is: .

[0022] 2. Adaptive update of model parameters: First, define the unknown parameter vector. ,and Time drift parameters .

[0023] It is the drift coefficient of the m-th actuator. The initial prior mean; It is the drift coefficient of the m-th actuator. The uncertainty measure of the initial prior mean, i.e., the initial prior variance; The effect of other actuator degradation on the degradation of the m-th actuator is a term related to the degradation of other actuators. The variance; It is the degradation trend function of the m-th actuator. The set of unknown parameters in; It is the standard Brownian motion that drives the m-th actuator to degenerate. The diffusion coefficient; To achieve unknown parameters and drift coefficient Simultaneously, it is estimated that new actuator degradation observations should be obtained. Then, calculate The maximum likelihood estimate; This application is based on degradation data of various actuators. The log-likelihood function is used express time exist Maximum likelihood under given conditions, where the degradation data of each actuator is... The log-likelihood function is: (3) The log-likelihood function as a whole is used to measure parameters. Observational data Explanatory power; It is the joint likelihood probability, which is based on the parameters. Below, a complete degraded data sequence was observed. The probability of; It is a one-step likelihood probability. It is the degenerate observation of the m-th actuator at time i; It is the degenerate data sequence of the first i-1 time steps of the m-th actuator; In the model, because the state-space model uses parameters related to drift... Data As a state, causing Includes missing data. Generally, when latent variables exist, the function... The unknown parameters are difficult to express explicitly and cannot be calculated using conventional maximum likelihood estimation methods. Therefore, missing data should be considered. Given the influence of the parameter log-likelihood function, finding the optimal value of the parameter log-likelihood function is a natural choice.

[0024] The basic principle of the EM algorithm is to approximate the maximum likelihood estimate of parameters by maximizing the joint likelihood function, providing a good framework for handling parameter estimation problems involving latent variables. A fundamental prerequisite for applying the EM algorithm is that the latent variables can be estimated from monitoring data.

[0025] In this application, given hour, The conditional probability is For example: (4) It is a conditional probability density function, representing the conditional probability density function given historical degradation data. At that time, drift parameters The probability density of taking a certain value It is the drift coefficient of the m-th actuator at time k. The set of degenerate observation data of the m-th actuator from the initial time 0 to the current time k. Based on historical data ,right The optimal estimate is the most likely value obtained by recursive algorithms such as Kalman filtering. Based on historical data ,right The measure of uncertainty in the estimate It is a Gaussian distribution kernel. It is the normalization coefficient.

[0026] In the formula and The dependencies between them are included in the estimated values. and As can be seen, the conditions for applying the EM algorithm are met.

[0027] In the EM algorithm, according to and The relationship between these parameters may generate a series of maximum likelihood parameter estimates that converge to the parameters. Therefore, the log-likelihood function of the complete data is defined as follows: .

[0028] It is the log-likelihood function of the complete data. It is the joint likelihood probability of the complete data. It is a degraded observation data sequence. It is a sequence of drift parameters (latent variables).

[0029] In one embodiment, the EM algorithm includes the following two steps: Step E: Calculate the corresponding conditional expectation based on the mean and variance of the drift parameters using the RTS smoothing algorithm, which is used to obtain the conditional expectation of the log-likelihood function of the complete dataset. The conditional expectation of the log-likelihood function of the complete dataset calculated in the E-step is: (5) in: , indicating that This is the parameter estimate for the nth iteration of the condition.

[0030] To obtain Compared to nth iteration and degenerate data The expected value, then (6) According to the Markov property of the standard BM, the total log-likelihood function can be expressed as:

[0031] (7) Based on the state-space model in equation (2), we have

[0032]

[0033]

[0034] Ignoring the constant term in equation (7), the complete log-likelihood function can be further expressed as: (8) Equation (5) gives the conditional expectation It can be written as: (9) Clearly, to obtain the expectation in equation (9), the degenerate state needs to be determined. Conditional expectations , and These quantities are all based on observation data. Conditional expectation of the estimated drift parameters.

[0035] In this embodiment, a smoothing filtering (Rauch-Tung-Striebel, RTS) algorithm is used to calculate the above expectation, wherein... The specific steps are as follows: Step 1: Use the forward iteration of the aforementioned Kalman filtering algorithm to obtain the optimal state estimate.

[0036] Step 2: Reverse iteration

[0037]

[0038]

[0039] It is the posterior variance of the drift coefficient at time j (the result of forward filtering). It is the prediction variance at time j+1 based on time j (one-step prediction of the forward filter). It is the gain coefficient of the reverse iteration, used to fuse forward and reverse information to correct the state estimate; Step 3: Initialization

[0040] Step 4: Reverse iteration of smoothing covariance

[0041] Based on the RTS smoothing algorithm, conditional expectation , and The following can be calculated:

[0042]

[0043]

[0044] Furthermore, combining equation (9). It can be written as:

[0045] (10) It is a smoothed estimate of the drift coefficient at time j (a better estimate after fusing the full 0-k data). Log-likelihood function (a measure of parameters) (the degree of matching with the observed data), It is during the conditional period (EM algorithm, step E, for latent variables) (Calculate the expectation of the drift coefficient sequence, bypassing the difficult problem of direct calculation of hidden variables); This completes the calculation of step E. Next, we will process step M.

[0046] M-step, fixed Based on the conditional expectation of the log-likelihood function of the complete dataset, the profile log-likelihood function is obtained. A search algorithm is then used to maximize the profile log-likelihood function. Make an estimate, and based on Update parameters; In obtaining Afterwards, through the analysis of maximize You can get the first Step unknown parameter estimation However, due to the parameters The effects of the nonlinear components involved are difficult to obtain by maximizing equation (10). The explicit form. Because the degenerate nonlinear part of the actuator in this application is represented as... ,therefore In the following estimation of unknown parameters, we directly use replace To address the aforementioned problem, the cross-sectional number likelihood function technique is used to calculate... The specific process is as follows: First, given In the In-step parameter estimation ;

[0047]

[0048]

[0049] (11) in: , , It is about The function, given ,but , and None of them are subject to Influence, These are the estimated model parameters of the m-th actuator after the nth iteration. It is the estimated value of the initial mean of the drift coefficients after the (n+1)th iteration. This is the estimated value of the initial variance of the drift coefficients after the (n+1)th iteration. This is the variance estimate of the coupling effect term between actuators after the (n+1)th iteration. This is the estimated value of the Brownian motion diffusion coefficient after the (n+1)th iteration. It is the degradation trend function 'fire parameter' after the (n+1)th iteration.

[0050] Secondly, substituting equation (10) into equation (11) yields the cross-sectional number likelihood function as follows: (12) Next, the cross-sectional likelihood function in equation (12) is maximized using a search algorithm to obtain the approximation of the cross-sectional likelihood function. The estimate can be obtained by optimizing equation (12) using the "fminsearch" function in MATLAB. Then, Substituting into equation (11), we obtain the following results: , , , and The estimated value is obtained by directly maximizing (10). Unlike other methods, the profile-based number-likelihood function technique reduces the number of unknown parameters in the search algorithm. Then, iterate through the E-step and M-step until the convergence criterion is met. During system operation, the model parameters are continuously updated using newly acquired monitoring data.

[0051] Third, life expectancy prediction: Based on the updated model parameters Based on the system health index, the probability density function of the remaining lifespan of the quadcopter drone is:

[0052] In the formula: This enables the prediction of remaining lifespan.

[0053] It can be seen that, due to The step-by-step updates allow the PDF of the quadcopter drone to be obtained under the new model parameters, thus enabling adaptive predictions of the drone's lifespan under the time-varying model. The adaptive updates of the degradation model and the integrated degradation variables will undoubtedly improve the granularity of the quadcopter drone's RUL prediction.

[0054] Example 2: This embodiment discloses an autonomous maintenance strategy for a quadcopter unmanned aerial vehicle (UAV), the process of which is as follows: Figure 1 ; First, the remaining lifespan of the quadrotor UAV under multi-source uncertainty is predicted using any of the above-described adaptive lifespan prediction methods for multi-source uncertainty. When the predicted remaining lifespan needs to be extended, autonomous maintenance control is implemented by adjusting relevant parameters based on the LF-MPC life extension mechanism (such as adjusting the weight matrices Q and R to extend the working time of the UAV).

[0055] Example 3: An adaptive lifetime prediction system for quadrotor unmanned aerial vehicles (UAVs) under multi-source uncertainty, employing any of the adaptive lifetime prediction methods for quadrotor UAVs under multi-source uncertainty described above, includes: The model building module is used to construct the degenerate model of each actuator and determine the state space model. The parameter update module is used to estimate the mean and variance of the drift parameters in the state-space model based on Kalman filtering, and to update the parameters of the degenerate model using the EM algorithm. The lifetime prediction module calculates the remaining lifetime probability density function based on the updated parameters.

[0056] Simulation experiment setup: Actuator monitoring data can be obtained using TSKF. Furthermore, the EM algorithm is used to adaptively estimate and update the parameters of the degradation model using real-time monitoring data. Let the initial parameters of the model be... , specifically, , , , , The parameter estimation results of the degradation data model for each actuator are as follows: Figure 2 As shown; from Figure 2 It can be seen that when the four actuators degenerate, the model parameters of each actuator converge rapidly with the accumulation of monitoring data, achieving adaptive parameter updates. The model parameters of the first actuator include... , , , and The model parameters for the second, third, and fourth actuators follow the same pattern.

[0057] The overall degradation of the quadcopter UAV obtained after fusion using the entropy weight method is as follows: Figure 3 As shown, the changes are more gradual compared to the case without adaptive methods.

[0058] The parameters of the comprehensive degradation model are also constantly updated and adjusted according to the degradation of each actuator, such as... Figure 4 As shown. The parameters of the comprehensive degradation model include... , and , respectively corresponding Figure 4 (a), (b), and (c) are in the middle.

[0059] Adaptive lifetime prediction results analysis: Figure 5 The health curves of the actuator degradation model before and after adaptive parameter maintenance are presented. The health of the UAV is slightly improved at the same time after the parameter adaptive maintenance, indicating that the state of the UAV has been improved.

[0060] Take the health failure threshold Compare the RUL distributions before and after using the parameter adaptation method ( Figure 6 Based on parameter adaptation, RUL can better fit the actual RUL results.

[0061] Adaptive maintenance effect analysis: Figure 7 The diagram illustrates the position and attitude changes of a quadcopter UAV with adaptive actuator parameter updates. The actuator parameters include x, y, z, phi, theta, and psi, which correspond to... Figure 7 (a)-(f). By Figure 7 It is evident that the state changes relatively slowly after the degradation parameters are adaptively updated.

[0062] After adopting the autonomous maintenance method of this application, the remaining lifespan distribution of the UAV is as follows: Figure 8 As shown in Table 1, compared with the non-parameter adaptive method, the lifespan of the UAV was extended after maintenance using the parameter adaptive method.

[0063] Table 1

[0064] The improved maintenance performance is mainly due to the inclusion of drift parameters in the RUL prediction process. The uncertainty in the estimation improves the targeting of maintenance, making the maintenance function based on RUL predictive feedback adjustment more in line with the actual health status of the UAV.

[0065] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0066] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An adaptive lifetime prediction method for quadrotor unmanned aerial vehicles under multi-source uncertainty, characterized in that, Construct degradation models for each actuator; Update the parameters of the degenerate model using the EM algorithm, including performing the following iterations: Step E: Determine the state-space model and estimate the mean and variance of the drift parameters based on Kalman filtering; the state-space model is: In the formula, and Let K and K-1 represent the drift coefficients of the m-th actuator at times k and k-1, respectively. This represents the effect of other actuators at time k on the degradation of the m-th actuator. and Let k and k-1 represent the degradation amounts at time k and k-1 of the m-th actuator, respectively. Indicates the diffusion coefficient. This represents the random error term of the Brown motion at time k after discretization for the m-th actuator. It concerns the m-th actuator at time k with respect to unknown parameters. The difference function, where: Used to describe the Brownian motion corresponding to the m-th actuator at time t. Similarly, the state value, Is the Brownian motion at time... The state value, It is a trend term used to accumulate the degradation process of the m-th actuator. It is the core trend function of the m-th actuator degradation, describing the degradation process over time. Unknown parameters Nonlinear laws governing change; The RTS smoothing algorithm is used to calculate the corresponding conditional expectation based on the mean and variance of the drift parameters, which is then used to obtain the conditional expectation of the log-likelihood function of the complete dataset. M-step: Updating parameters by maximizing the facet number likelihood function using a search algorithm; including: By fixing the degenerate shape parameters, the profile surface likelihood function is obtained based on the conditional expectation of the log-likelihood function of the complete dataset. Maximize the facet number likelihood function of the cross section using the search algorithm fminsearch; Iterate through E-steps and M-steps until the convergence criterion is met, and calculate the remaining lifetime probability density function based on the updated parameters.

2. The UAV adaptive lifespan prediction method according to claim 1, characterized in that, The steps for estimating the mean and variance of the drift parameters based on Kalman filtering include: Let the initial drift parameters and initialize , Where m represents the m-th executor, This represents the initial prior mean of the drift coefficient. This represents the initial prior variance of the drift coefficient. The state at time step is estimated as follows: Variance updated as follows: In the formula, k represents the current time step. This represents the time point of the k-th monitoring. This represents the prior state estimate. This represents the posterior state estimate. This indicates a priori estimation of the covariance. This indicates the posterior estimate of the covariance. Represents the process noise variance. Represents the Kalman gain matrix. This represents the updated posterior state estimate. This represents the updated posterior covariance.

3. The UAV adaptive lifespan prediction method according to claim 1, characterized in that, The conditional expectation is calculated using the RTS smoothing algorithm based on the mean and variance of the drift parameters, including: In the formula, Indicates that in the given observation data and parameter estimation Below, for latent variables The conditional expectation is given by m, where m represents the m-th executor, j represents the time step, k represents the current time, and n represents the n-th iteration. Represents the set of latent variables. Indicates observed degradation data, Represents the parameter estimation vector. Indicates the drift coefficient. This represents a smoothed estimate based on all data; This represents the mean square error of the smoothing estimate. This represents the posterior variance of the forward filter. Indicates smoothing gain. This represents the covariance of the smoothed state estimate. This represents a smoothed estimate of the drift coefficient.

4. The UAV adaptive lifespan prediction method according to claim 1, characterized in that, The remaining lifetime probability density function is: In the formula, , indicating from the current moment To the future The cumulative nonlinear degradation trend, Indicates from the current moment The length of time calculated Indicates at time Updated shape parameters, Indicates at time Updated drift coefficients This is the difference between the current degradation state and the failure threshold. Indicates at time Updated diffusion coefficient.

5. An autonomous maintenance strategy for a quadcopter unmanned aerial vehicle (UAV), characterized in that, The remaining lifespan of a quadrotor UAV under multi-source uncertainty is predicted using the adaptive lifespan prediction method for a quadrotor UAV as described in any one of claims 1-4. When the predicted remaining lifespan needs to be extended, relevant parameters are adjusted based on the LF-MPC life extension mechanism to implement autonomous maintenance control.

6. An adaptive lifetime prediction system for a quadrotor unmanned aerial vehicle under multi-source uncertainty, characterized in that, The adaptive lifetime prediction method for quadrotor UAVs under multi-source uncertainty as described in any one of claims 1-4 includes: The model building module is used to construct the degenerate model of each actuator and determine the state space model. The parameter update module is used to estimate the mean and variance of the drift parameters in the state-space model based on Kalman filtering, and to update the parameters of the degenerate model using the EM algorithm. The lifetime prediction module calculates the remaining lifetime probability density function based on the updated parameters.

Images