CN-121979285-A - Unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning

CN121979285ACN 121979285 ACN121979285 ACN 121979285ACN-121979285-A

Abstract

The invention discloses a digital twin-based unmanned aerial vehicle non-parameter data fault detection and recovery method which comprises the steps of collecting multi-source data, preprocessing to generate a standard data set, constructing a digital twin, mapping the data to the twin, obtaining a prediction state and a difference between the prediction state and an actual measurement state, dividing multi-scale residual errors, constructing a nested manifold, outputting an abnormal geometric position, constructing a reversible enveloping body, locating a layer, determining a target recovery state, calculating a local response relation, constructing an inversion operator, solving a recovery control quantity, synchronizing the control quantity to the twin, executing verification and issuing to unmanned aerial vehicle control. According to the invention, through constructing a digital twin body and fusing multi-scale residual error geometric identification and partial derivative inversion control, the precise detection, reversible recovery and closed-loop control of the unmanned aerial vehicle under the non-parametric data abnormal working condition are realized.

Inventors

TIAN HONGLIN
LI JUAN
Tian Tianhao

Assignees

重庆添博睿科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260202

Claims (9)

1. The digital twinning-based unmanned aerial vehicle non-parametric data fault detection and recovery method is characterized by comprising the following steps of: The method comprises the steps of collecting multi-source operation data generated by an unmanned aerial vehicle in the flight process, preprocessing the multi-source operation data to form a standardized non-parametric operation data set, and constructing a digital twin body; Mapping the standardized non-parametric operation data set to a digital twin body, obtaining twin prediction state data at corresponding time and performing differential processing with the actually measured operation state data of the unmanned aerial vehicle to generate a residual error data sequence; performing multi-scale division according to a preset time scale based on a residual data sequence to form a multi-scale residual fragment set, performing non-parameter manifold embedding processing, constructing a multi-fractal nested twin residual manifold, and outputting an abnormal state identifier and the geometric position of the abnormal state in a digital twin space; Constructing a reversible steady-state envelope body comprising a steady-state core, a reversible absorption band and a degradation envelope layer in a digital twin space, and determining a steady-state envelope level and a target recovery state corresponding to an abnormal state according to the geometric position of the abnormal state in the digital twin space; Based on the digital twin body, calculating a local partial derivative response relation of control input change to residual data change, constructing a partial derivative response inversion operator by combining control constraint conditions of an unmanned aerial vehicle physical execution mechanism, and calculating a corresponding recovery control quantity according to a target recovery state; Synchronizing the recovery control quantity to the digital twin body for virtual execution verification, transmitting the recovery control quantity to the unmanned aerial vehicle for executing recovery control after the verification is passed, and feeding back the operation data acquired in the recovery execution process to the digital twin body.
2. The digital twinning-based unmanned aerial vehicle non-parametric data fault detection and restoration method of claim 1, wherein the multi-source operational data comprises attitude and motion data, power system operational data, control command data, and environmental disturbance data.
3. The unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning according to claim 1, wherein the preprocessing of the multi-source operation data comprises time synchronization, noise filtering, data alignment and non-parametric normalization of the multi-source operation data.
4. The unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning according to claim 1, wherein the constructing the digital twinning comprises: Digitally mapping body structure parameters, power system parameters, sensor configuration parameters and actuator connection parameters of the unmanned aerial vehicle based on a standardized non-parametric operation data set to form a structure mapping relation; Based on a standardized non-parametric operation data set, carrying out state association processing on attitude data, motion data and power data of the unmanned aerial vehicle under different control instructions and environmental disturbance conditions to form a state evolution mapping relation; based on a standardized non-parametric operation data set and acquired control instruction data, carrying out joint mapping on the corresponding relation between the control input change and the unmanned aerial vehicle state change to form a control response mapping relation; and carrying out consistency constraint fusion on the structure mapping relation, the state evolution mapping relation and the control response mapping relation, and synchronously updating the mapping relation according to the multi-source operation data acquired in real time to form a digital twin body corresponding to the operation state of the unmanned aerial vehicle in real time.
5. The digital twinning-based unmanned aerial vehicle non-parametric data fault detection and recovery method of claim 1, wherein the generating a residual data sequence comprises: Loading the standardized non-parametric operation data set into the digital twin body according to the data time mark and characteristic corresponding relation, so that each operation state variable in the digital twin body and the corresponding data in the standardized non-parametric operation data set are in one-to-one mapping relation; In a digital twin body, carrying out state synchronization processing on standardized non-parametric operation data based on a mapping relation to generate a twin operation state corresponding to the current moment, and forming twin prediction state data by the twin operation state; Under the same time mark, carrying out characteristic alignment processing on the twin prediction state data and the actually measured running state data acquired by the unmanned aerial vehicle in a digital twin body; Performing differential processing on the twin prediction state data and the actual measurement running state data with the aligned features according to the corresponding features item by item to obtain residual data representing the deviation relation between the twin prediction state and the actual measurement running state; and integrating the residual data obtained under a plurality of continuous time marks according to the time sequence to form a residual data sequence.
6. The unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning according to claim 1, wherein the constructing multi-fractal nested twinning residual manifolds, outputting abnormal state identifiers and geometric positions of abnormal states in a digital twinning space, comprises: Acquiring a residual data sequence, sequentially sorting the residual data sequence according to the time marks, and binding the residual data at each moment with a digital twin space state mark corresponding to the moment to form a residual sample sequence with a state mark; Carrying out multi-scale division on the residual sample sequence with the state marks based on a preset time scale, intercepting continuous residual fragments in a sliding window mode under each time scale to form a multi-scale residual fragment set, and carrying out feature dimension unification on the multi-scale residual fragment set; under each time scale, carrying out non-parameter manifold embedding processing on the corresponding residual fragment set to obtain residual manifold representation under the time scale, and introducing a twin space state mark as a neighborhood constraint condition in the embedding processing process to form a state constraint residual manifold; Performing cross-scale nesting construction processing on state constraint residual manifolds formed under different time scales, wherein the cross-scale nesting construction processing comprises the steps of determining a corresponding relation among all scale residual manifolds, establishing cross-scale alignment mapping, generating a cross-scale nesting boundary, and combining all scale residual manifolds into multi-fractal nesting twin residual manifolds based on the cross-scale nesting boundary; Embedding and positioning the current residual fragment in the multi-fractal nested twin residual manifold, calculating the position relation of the current embedded point relative to the trans-scale nested boundary, outputting an abnormal state identification according to the position relation, and outputting the geometric position of the abnormal state in the digital twin space.
7. The method for detecting and recovering non-parametric data faults of an unmanned aerial vehicle based on digital twinning according to claim 1, wherein the determining a steady-state envelope level and a target recovery state corresponding to an abnormal state according to the geometric position of the abnormal state in a digital twinning space comprises: Selecting residual embedded samples corresponding to normal operation time periods in a digital twin space based on multi-fractal nested twin residual manifold, sorting according to time marks and characteristic dimensions, determining a sample set which simultaneously meets the consistency of residual stability and state evolution under each scale, and taking the minimum surrounding domain of the sample set as the initial region boundary of a steady-state kernel; Performing bidirectional rolling simulation around the boundary of the initial area of the steady-state core in the digital twin body, applying control disturbance and environmental disturbance with limited amplitude to the neighborhood state of the boundary, recording a state set which can return to the steady-state core without external reconfiguration in a limited step, and determining a continuous area between the state set and the steady-state core as a reversible attraction belt; Executing a local reversible geometric transformation generating process in a digital twin body for the residual states which do not meet the reversible attraction belt condition, generating a plurality of groups of candidate geometric transformations through local state re-parameterization, control law fine tuning and constraint consistency, carrying out unified aggregation on the states which can enable the states to enter the reversible attraction belt in a limited step, determining the aggregated state set as a degradation envelope layer, and recording corresponding local reversible geometric transformation marks for each type of state; Self-consistent calibration is carried out on the boundaries of the steady-state kernel, the reversible absorption band and the degradation envelope layer in the digital twin body, a counter fact disturbance test and time symmetry playback verification are carried out on the neighborhood of the boundary, and the boundaries of the three types of regions are synchronously updated according to the consistency of the trans-scale residual error structure and the state evolution consistency to form a reversible steady-state envelope body; According to the geometric position of the abnormal state in the digital twin space, finishing the judgment of the belonging hierarchy: when the position is in the steady-state core, determining the target recovery state as the current corresponding steady-state point; when the position is in the reversible attraction zone, determining that the target recovery state reaches a steady-state point with the shortest steady-state nuclear evolution time; when the position is within the degradation envelope, an intermediate target entering the reversible attraction belt is determined according to the corresponding local reversible geometric transformation mark and a final target recovery state is determined.
8. The unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning according to claim 1, wherein the constructing a partial derivative response inversion operator by combining control constraint conditions of a physical execution mechanism of the unmanned aerial vehicle, calculating a corresponding recovery control amount according to a target recovery state, comprises: obtaining residual data corresponding to the current abnormal state in the digital twin body, and obtaining current control input corresponding to the current abnormal state, wherein the residual data is derived from a residual data sequence, and the current control input is derived from control instruction data; Generating a plurality of groups of limited control disturbance sequences in the digital twin body by taking the current control input as a reference, synchronizing the limited control disturbance sequences to the digital twin body for virtual execution, and recording residual variation data corresponding to each group of limited control disturbance sequences to form a corresponding sample set of control disturbance and residual variation; Based on the corresponding sample set, constructing a local partial derivative response relation of control input change to residual data change, and keeping control disturbance results which keep stable change trend in a multi-scale residual structure and an abnormal state geometric neighborhood by carrying out consistency screening on residual change results corresponding to different amplitude control disturbance so as to form a local partial derivative response relation of structural constraint; constructing a partial guide response inversion operator based on a structure constraint local partial guide response relation, wherein the partial guide response inversion operator consists of control constraint projection processing, residual target alignment processing and inversion stabilization processing, and the method comprises the following steps of: The control constraint projection process maps candidate control input changes to feasible control domains meeting the constraints of the physical execution mechanism; the residual target alignment process is used for conforming the current residual change direction with the residual change direction corresponding to the target recovery state; the inversion stabilization process suppresses the uncomfortable direction in the inversion process and preferentially reserves inversion solutions of the reversible attraction zone neighborhood; and taking the target residual error change corresponding to the target recovery state as inversion input, taking the current control input as a control reference, and calculating through a partial derivative response inversion operator to obtain recovery control quantity.
9. The method for detecting and recovering non-parametric data faults of an unmanned aerial vehicle based on digital twinning according to claim 1, wherein synchronizing the recovery control amount to the digital twinning body for virtual execution verification, and issuing the recovery control amount to the unmanned aerial vehicle for executing recovery control after the verification is passed, comprises: acquiring a recovery control quantity, synchronizing the recovery control quantity to a digital twin body, and enabling the digital twin body to enter a virtual execution state taking the recovery control quantity as a control input; performing virtual execution on the unmanned aerial vehicle operation process in the digital twin body according to the recovery control quantity, and generating a virtual operation state sequence corresponding to the recovery control process, wherein the virtual operation state sequence is consistent with the actual operation state of the unmanned aerial vehicle in time sequence and state dimension; Generating virtual residual data based on the virtual running state sequence in the digital twin body, and comparing the virtual residual data with the residual data sequence to complete virtual execution verification judgment; under the condition that virtual execution verification is passed, the recovery control quantity is issued to the unmanned aerial vehicle flight control system, so that the unmanned aerial vehicle executes recovery control operation according to the recovery control quantity; And in the process of the unmanned aerial vehicle executing the recovery control operation, acquiring real-time operation data of the unmanned aerial vehicle, and synchronously feeding back the real-time operation data to the digital twin body for updating.

Description

Unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning Technical Field The invention relates to the technical field of unmanned aerial vehicle intelligent control and operation safety guarantee, in particular to a digital twinning-based unmanned aerial vehicle non-parameter data fault detection and recovery method. Background With the wide application of unmanned aerial vehicles in the fields of inspection, mapping, logistics transportation, emergency rescue and the like, the operation environment of the unmanned aerial vehicle is increasingly complex, the duration of a flight task is continuously prolonged, and the fault risk of the unmanned aerial vehicle in the actual operation process is obviously increased. The existing unmanned aerial vehicle operation safety guarantee technology mainly relies on sensor threshold monitoring, state estimation based on a physical model or abnormal judgment mode based on experience rules to monitor and diagnose flight attitude, a power system and control signals. The method can play a certain role under the conditions of clear structure and stable working condition, but lacks effective adaptability to complex environmental disturbance, system aging and abnormal response caused by multifactor coupling. On the other hand, some of the prior art introduce data-driven or model-predictive methods to analyze and predict the operation state of the unmanned aerial vehicle, but still rely highly on preset parametric models or specific fault samples. Because the unmanned aerial vehicle system has strong nonlinearity, strong coupling and time-varying characteristics, non-parameterized anomalies or unknown faults which are difficult to describe by using a fixed model often occur in actual operation, and misjudgment or missed judgment is easy to generate in an anomaly detection stage in the existing method. In the traditional method, after abnormality is detected, a conservation strategy such as degradation control, returning or stopping is mostly adopted, the recovery process lacks pertinence, and reversible recovery of the state is difficult to realize on the premise of ensuring safety. In recent years, the digital twin technology is started to be applied to operation monitoring and simulation analysis of an unmanned aerial vehicle system, but the existing related scheme focuses on state mapping or off-line simulation verification, the coupling degree between a digital twin body and an actual control decision is limited, and a closed loop fault detection and recovery mechanism based on the digital twin is not formed. The existing digital twin scheme generally adopts a simple error comparison or single-scale analysis mode, is difficult to describe a complex residual structure and evolution characteristics thereof, and cannot support accurate positioning and recovery control of unknown faults. Therefore, how to provide a digital twin-based unmanned aerial vehicle non-parametric data fault detection and recovery method is a problem that needs to be solved by those skilled in the art. Disclosure of Invention One object of the invention is to provide a digital twin-based unmanned aerial vehicle non-parametric data fault detection and recovery method, which constructs a digital twin corresponding to the operation state of an unmanned aerial vehicle in real time, and (3) fusing technologies such as multi-fractal residual structure analysis, reversible steady-state envelope construction, partial derivative response inversion control and the like, and detecting, positioning and recovering non-parameterized anomalies and unknown faults generated by the unmanned aerial vehicle in a complex environment. The invention utilizes the digital twin body to map and virtually execute verification on the multi-source operation data, describes an abnormal evolution rule through the geometric characteristics of the multi-scale residual error, plans a reversible recovery state in the digital twin space, and further generates verified recovery control quantity to be issued and executed. The method can effectively improve the accuracy and stability of unmanned aerial vehicle fault detection, enhance the adaptability to unknown faults, realize the controllability and reversibility of the recovery process, and have the advantages of strong robustness, high safety and high recovery efficiency. According to the embodiment of the invention, the unmanned aerial vehicle non-parametric data fault detection and recovery method based on digital twinning comprises the following steps: The method comprises the steps of collecting multi-source operation data generated by an unmanned aerial vehicle in the flight process, preprocessing the multi-source operation data to form a standardized non-parametric operation data set, and constructing a digital twin body; Mapping the standardized non-parametric operation data set to a digital twin body, obtaining twin prediction stat