Digital twinning-based unmanned aerial vehicle sensor data fusion fault diagnosis method and system

Abstract

The invention relates to the field of unmanned aerial vehicle fault diagnosis and unmanned aerial vehicle digital twinning, in particular to an unmanned aerial vehicle sensor multi-modal data fusion fault diagnosis method and system based on digital twinning, and the method comprises the steps: 1), carrying out the collection and preprocessing of unmanned aerial vehicle multi-modal data; 2) constructing a digital twin body of the unmanned aerial vehicle; 3) extracting and fusing multi-modal data features; and 4) diagnosing and predicting the fault of the unmanned aerial vehicle. According to the method, the digital twin model highly consistent with the physical entity of the unmanned aerial vehicle is constructed, and the multi-modal data and the deep learning algorithm are fused, so that intelligent upgrading of fault diagnosis is realized, and the problems of low utilization rate of the multi-modal data, insufficient model precision, lack of prediction capability and the like in a traditional method are effectively solved. According to the technology, the limitation of single data source diagnosis is broken through by integrating sensor, vision, thermal imaging and other multi-source heterogeneous data, and the fault identification accuracy and prediction capability are remarkably improved. Based on a dynamic updating mechanism of the digital twin model, the system response time is greatly shortened, and real-time monitoring and autonomous decision making of the health state of the unmanned aerial vehicle can be realized in a complex environment, so that the flight safety and the operation and maintenance efficiency are remarkably enhanced.

Description

Technical Field

[0001] This invention relates to the fields of UAV fault diagnosis and UAV digital twins, and particularly to a fault diagnosis method and system for UAV sensor multimodal data fusion based on digital twins. Background Technology

[0002] With the rapid development of drone technology in the aviation field, its application in geological surveying, logistics distribution, emergency rescue and other fields is becoming increasingly widespread, and it has become an indispensable key equipment in modern social production and life. Against this backdrop, the operational reliability and flight safety of drones are becoming increasingly prominent – ​​once a malfunction occurs, it may not only cause equipment damage and mission interruption, but may even lead to safety accidents such as crashes and collisions, resulting in casualties and property losses.

[0003] In complex and ever-changing operating environments, drones rely on various sensors to continuously collect massive amounts of operational data: flight attitude sensors (such as IMUs) output real-time dynamic parameters reflecting the aircraft's attitude, such as pitch, roll, and yaw angles; power system sensors precisely capture key indicators such as motor speed, winding temperature, and operating current; and environmental perception sensors simultaneously record external conditions such as air pressure, humidity, wind speed, and wind direction in the flight airspace. This data comes from diverse sources and has a heterogeneous structure, encompassing different modes such as time-series signals, image information, and vibration waveforms. Data from a single mode can only reflect a localized aspect of the drone's operational status and cannot comprehensively depict its overall health.

[0004] Traditional drone fault diagnosis methods have significant limitations: threshold-based methods only preset fixed thresholds for a single type of data (such as judging an anomaly if the motor temperature exceeds 80°C), ignoring the correlation between multiple parameters (such as the motor temperature rise may be related to factors such as low wind speed or excessive load), which can easily lead to missed diagnoses; experience-based diagnosis models rely entirely on the professional accumulation of maintenance personnel, are significantly affected by subjective judgment differences, and have extremely low accuracy in identifying hidden faults and complex faults.

[0005] While existing multimodal data fusion methods attempt to integrate information from multiple sources, they still face technical bottlenecks in UAV fault diagnosis scenarios. Firstly, they fail to deeply integrate digital twin technology, making it impossible to construct a virtual model that fully maps to the physical UAV and thus difficult to reproduce the dynamic operating state under complex conditions. Secondly, insufficient mining of the spatiotemporal correlation features of multimodal data leads to the loss of key information during the fusion process, resulting in delayed fault prediction and a high rate of misdiagnosis. These shortcomings directly cause low efficiency in UAV fault diagnosis, making it difficult to meet the high-efficiency operation and maintenance requirements of scenarios such as high-density task scheduling and long-endurance operations. A novel technical solution is needed to overcome these existing challenges. Summary of the Invention

[0006] The purpose of this invention is to provide a fault diagnosis method and system for UAV sensors based on digital twins, which solves the problems of insufficient utilization of multimodal data, insufficient model accuracy, and lack of predictive ability in traditional UAV fault diagnosis.

[0007] The technical solution of the present invention: A fault diagnosis method for UAV sensor data fusion based on digital twins includes the following steps: 1) Unmanned aerial vehicle (UAV) multimodal data acquisition and preprocessing; 2) Constructing a digital twin of the drone; 3) Extract and fuse features from multimodal data; 4) Diagnose and predict drone malfunctions.

[0008] Furthermore, the UAV multimodal data acquisition and preprocessing process in step 1) includes: 1-1) Selection and layout of drone fuselage sensors; 1-2) Multimodal data acquisition from UAV sensors; 1-3) Multimodal data transmission from UAVs; 1-4) Preprocessing of UAV multimodal data.

[0009] Furthermore, in step 1), during the data acquisition phase, multimodal data from the temperature sensor, rotation speed sensor, and environmental perception camera of the UAV monitoring system are collected. In the preprocessing stage, wavelet transform is first used to eliminate noise from temperature and speed sensors. Then, a dual long short-term memory network is used to fill in missing values ​​in the time series data. Subsequently, a normalization algorithm is used to achieve scale unification of multi-source data. Finally, joint calibration technology is used to complete the spatiotemporal alignment and system-level calibration of temperature and speed sensor data, ensuring accurate synchronization of multimodal data in the temporal and spatial dimensions.

[0010] Furthermore, step 2) the process of constructing a digital twin of the drone includes: 2-1) Construction of digital twin geometric model of UAV; 2-2) Construction of a digital twin dynamics model for unmanned aerial vehicles (UAVs); 2-3) Physical property modeling of UAV digital twin models; 2-4) Development and real-time mapping of digital twin data interfaces; 2-5) Digital twin model verification and calibration.

[0011] Further, step 2) specifically involves building a geometric model of the UAV using laser scanning and CAD modeling software; then building a dynamic model of the UAV based on its dynamic parameters; simultaneously constructing a physical property model for the materials and characteristics of each component of the UAV; and finally developing a standardized data interface to achieve real-time mapping between multimodal data and the digital twin.

[0012] Furthermore, step 3) of the UAV's multimodal data feature extraction and fusion process includes: 3-1) Feature extraction of time series data, vibration data, and other data; 3-2) Multimodal data fusion of time series data, vibration data, and time series data is performed using a feature fusion network based on the attention mechanism.

[0013] Further, step 3) specifically involves: in the multimodal data feature extraction stage, using an improved YOLOv8 combined with an attention mechanism to extract target features from UAV image data, and using FFT and autoencoder to process vibration data; in the multimodal data fusion stage, a network is constructed based on the attention mechanism to dynamically allocate the weights of each modality of data and splice them to form a comprehensive feature vector, and the weights are adjusted through reinforcement learning and multimodal cross-validation to achieve complementary data advantages to support accurate fault diagnosis of UAVs.

[0014] Furthermore, the fault diagnosis and prediction process for the UAV in step 4) includes: 4-1) Training a UAV fault diagnosis model based on a deep learning model; 4-2) Real-time fault diagnosis of UAVs based on multimodal data mapped in real time from UAV digital twins; 4-3) Predict short-term drone failures based on time series models, and predict long-term drone failures based on machine learning and digital twins.

[0015] Further, step 4) specifically involves: constructing a high-precision fault diagnosis model based on deep learning and ensemble learning algorithms; when diagnosing UAV faults in real time, the collected multimodal data is preprocessed, feature extracted and fused, and then input into the fault diagnosis model; if the threshold is exceeded, a fault is determined and a detailed alarm is output; in the prediction stage, short-term parameter trends and long-term potential fault risks are predicted by combining time series models and digital twins with machine learning.

[0016] A fault diagnosis system for UAV sensor data fusion, applied to the method, includes: Data acquisition and processing module: used for multimodal data acquisition and preprocessing of UAVs; Drone Digital Twin Building Module: Used to build drone digital twins; Data feature extraction and fusion module: used to extract and fuse features from multimodal data; Fault diagnosis module: Used to diagnose and predict drone faults.

[0017] The advantages of this invention: By constructing a digital twin model highly consistent with the physical entity of the UAV and integrating multimodal data with deep learning algorithms, this invention achieves an intelligent upgrade in fault diagnosis, effectively solving problems such as low utilization of multimodal data, insufficient model accuracy, and lack of predictive ability in traditional methods. This technology, by integrating heterogeneous data from multiple sources such as sensors, vision, and thermal imaging, overcomes the limitations of single-source data diagnosis, significantly improving the accuracy and predictive ability of fault identification. Relying on the dynamic update mechanism of the digital twin model, the system response time is greatly shortened, enabling real-time monitoring and autonomous decision-making of the UAV's health status in complex environments, thereby significantly enhancing flight safety and operational efficiency. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The invention will now be further described with reference to the accompanying drawings: One embodiment of the present invention is as follows: Figure 1 As shown, a fault diagnosis method for UAV sensors based on digital twin multimodal data fusion is provided, including the following steps: 1) Multimodal data acquisition and preprocessing of UAVs: In the acquisition stage, multimodal data from the UAV monitoring system, including temperature sensors, speed sensors, and environmental perception cameras, are collected to achieve high-precision synchronous acquisition of time-series signals, environmental images, and spatial environmental parameters. In the preprocessing stage, wavelet transform is first used to eliminate sensor noise, followed by the use of a dual long short-term memory network (Bi-LSTM) to fill in missing values ​​in the time-series data. Then, a normalization algorithm is used to achieve scale unification of multi-source data. Finally, joint calibration technology is used to complete the spatiotemporal alignment of sensor data and system-level calibration to ensure accurate synchronization of multimodal data in the temporal and spatial dimensions. This processing step effectively solves the technical challenges of heterogeneous data sources in terms of sampling frequency differences, data integrity, and spatial alignment, providing data preprocessing for subsequent digital twin modeling, data fusion analysis, and fault diagnosis.

[0020] 2) Construction of a digital twin of a drone: A geometric model of the drone is built using laser scanning and professional CAD modeling software; then, a dynamic model of the drone is built based on its dynamic parameters; at the same time, a detailed physical property model is built for the materials and characteristics of each component; and finally, a standardized data interface is developed to achieve real-time mapping between multimodal data and the digital twin.

[0021] 3) Multimodal data feature extraction and fusion. In the multimodal data feature extraction stage, an improved YOLOv8 algorithm combined with an attention mechanism is used to extract target features from UAV image data, and FFT and autoencoder are used to process vibration data. In the multimodal data fusion stage, a network is built based on the attention mechanism to dynamically allocate the weights of each modality and concatenate them to form a comprehensive feature vector. Through reinforcement learning to adjust the weights and multimodal cross-validation, the data advantages are complemented to support accurate fault diagnosis of UAVs.

[0022] 4) UAV Fault Diagnosis and Prediction. A high-precision fault diagnosis model is constructed based on deep learning and ensemble learning algorithms. When diagnosing UAV faults in real time, the collected multimodal data is preprocessed, features are extracted and fused, and then input into the model. If a threshold is exceeded, a fault is determined and a detailed alarm is output. In the prediction stage, time series models and digital twins are combined with machine learning to predict short-term parameter trends and long-term potential fault risks, respectively, providing a scientific basis for UAV operation and maintenance decisions.

[0023] In this embodiment, the UAV multimodal data acquisition and preprocessing process described in step 1) is as follows: 1-1) Selection and layout of drone fuselage sensors; 1-2) Multimodal data acquisition from UAV sensors; 1-3) Multimodal data transmission from UAVs; 1-4) Preprocessing of UAV multimodal data; In this embodiment, the process of constructing the digital twin of the UAV in step 2) is as follows: 2-1) Construction of digital twin geometric model of UAV; 2-2) Construction of a digital twin dynamics model for unmanned aerial vehicles (UAVs); 2-3) Physical property modeling of UAV digital twin models; 2-4) Development and real-time mapping of digital twin data interfaces; 2-5) Digital twin model verification and calibration; In this embodiment, the multimodal data feature extraction and fusion process of the UAV in step 3) is as follows: 3-1) Feature extraction of multimodal data such as time series data, vibration data, etc. 3-2) Feature fusion network based on attention mechanism for multimodal data fusion such as time series data, vibration data, etc.; In this embodiment, the fault diagnosis and prediction process of the UAV in step 4) is as follows: 4-1) Training a UAV fault diagnosis model based on a deep learning model; 4-2) Real-time fault diagnosis of UAVs based on multimodal data mapped in real time from UAV digital twins; 4-3) Predict short-term drone failures based on time series models, and predict long-term drone failures based on machine learning and digital twins.

[0024] The second embodiment of the present invention provides a method for fault diagnosis of unmanned aerial vehicle (UAV) sensor data fusion based on digital twins, comprising the following steps: Step 1: UAV multimodal data acquisition and preprocessing Step 1.1, Sensor Selection and Layout: Based on the UAV application scenario and fault diagnosis requirements, accurately deploy a variety of sensors such as high-precision IMU, temperature sensor, speed sensor, vibration sensor, and high-definition camera to collect multimodal data such as flight attitude, power system status, and environmental information.

[0025] Step 1.2, Multimodal Data Acquisition: Multimodal data acquisition includes time-series data acquisition of acceleration and angular velocity at a fixed high frequency, image data acquisition of environmental images at a fixed frame rate, and environmental data acquisition of environmental parameters through sensors such as barometers, anemometers, and wind vanes.

[0026] Step 1.3, Data Transmission: Using low-latency, high-reliability communication protocols, such as the MQTT protocol, the raw data collected by the sensors is transmitted in real time to the UAV's data processing module or ground control station, ensuring the stability and efficiency of data transmission.

[0027] Step 1.4, Data Preprocessing: In the data preprocessing stage, wavelet transform algorithm with wavelet basis function for 3-5 level decomposition is used to reduce noise in time series data; a model is built using bidirectional long short-term memory network (Bi-LSTM) to fill missing values ​​in time series data; and the Min-Max normalization method is used to map numerical time series data to the [0, 1] interval and normalize the pixel values ​​of image data to the [0, 1] range. General form of wavelet basis functions, wavelet functions Generated by translation and scaling:

[0028] In the formula: t: time Scale parameters (control frequency) Translation parameters (controlling time and position) Normalization: coefficients Ensure energy conservation (tb) / a: Controls the translation and scaling of the time scale. Min-Max normalization is a commonly used data standardization method that maps the original data to a specified interval, typically [0,1), through a linear transformation. Its mathematical formula is as follows: Mapping to the [0,1] interval:

[0029] Mapping to any interval [a, b]:

[0030] In the formula : Raw data; The minimum value in the dataset; The maximum value in the dataset; : The result value after normalization; [a,b]: Target range (default [0,1]) Step 2, Construction of the drone digital twin; Step 2.1, Geometric Model Construction: Use a 3D laser scanner to acquire point cloud data of the UAV's outline and industrial CT scan to acquire internal structural information. Import the data into professional modeling software such as SolidWorks, and construct a 1:1 accurate geometric model through surface fitting and mesh generation.

[0031] Step 2.2, Dynamic Model Construction: By combining theoretical calculations with actual experiments and other tests, the key parameters of the dynamic model, such as the mass and lift coefficient of the UAV, are accurately determined. Based on Newton's laws of motion and rigid body dynamics theory, a six-degree-of-freedom dynamic equation is established that considers the influence of gravity, aerodynamic force, torque and environmental factors, and a dynamic model that truly reflects the flight behavior of the UAV is constructed.

[0032] Step 2.3, Physical property modeling: By combining theoretical calculations with experimental tests such as wind tunnel experiments, the key parameters of the dynamic model such as the mass and lift coefficient of the UAV are accurately determined. Based on Newton's laws of motion and rigid body dynamics theory, a six-degree-of-freedom dynamic equation is established that considers the influence of gravity, aerodynamic force, torque and environmental factors, and a dynamic model that truly reflects the flight behavior of the UAV is constructed.

[0033] Step 2.4, Data Interface Development and Real-time Mapping: Based on the TCP / IP protocol, and with the help of data caching and preprocessing mechanisms, stable and efficient transmission and processing of multimodal data from UAV sensors are achieved. Data-driven technology is used to realize real-time transmission of multimodal data and synchronize the digital twin with the actual UAV's operating status.

[0034] Step 2.5, Model Validation and Calibration: Simultaneously collect real and digital twin data, and evaluate the accuracy of the UAV digital twin model by calculating error indicators such as RMSE and MAE.

[0035] Step 3: Multimodal data feature extraction and fusion; Step 3.1, Multimodal Data Feature Extraction: Temporal data feature extraction: For temporal data such as IMU and motor speed, a model combining CNN and LSTM is constructed. The former extracts local features, while the latter learns long-term dependencies, and together they extract feature vectors that reflect the changes in temporal data. Image data feature extraction: For UAV image data, the improved YOLOv8 algorithm is used to extract features, and CBAM is introduced to enhance the focus on key regions. The target feature vector is output through the backbone network, attention module and detection head. Vibration data feature extraction: The time-domain vibration signal of the UAV is converted into the frequency domain using FFT, an autoencoder is constructed to extract the frequency domain features by dimensionality reduction, representative features are learned, and a low-dimensional feature vector representing the vibration state is obtained.

[0036] Step 3.2, Multimodal Data Fusion: Construction of a feature fusion network based on attention mechanism: Design a fusion network containing multiple attention sub-modules, input the feature vectors of each modality into the corresponding sub-modules, measure the importance by calculating the attention weight matrix based on dot product, and then concatenate the weighted values ​​to form a comprehensive feature vector; Dynamic weight adjustment: Introducing reinforcement learning algorithms (such as DQN) to dynamically adjust the feature fusion weights of each modality based on real-time multimodal data and fault diagnosis results, adapting to different flight stages and fault types of UAVs; Feature verification and correction: Statistical analysis is used to compare the integrated feature vector with the normal model. If anomalies are found, cross-validation and correction are performed using other modal data to ensure that the feature vector can truly reflect the operating status of the UAV.

[0037] Step 4, UAV fault diagnosis and prediction; Step 4.1, Fault Diagnosis Model Training: Construct a fault diagnosis model based on deep learning, such as deep neural networks (DNN), convolutional neural networks (CNN), recurrent neural networks (RNN), etc., or use an ensemble learning model combined with a deep learning model. Take multimodal fusion features as input and labeled fault information as output, train the model using the corresponding loss function and optimization algorithm, adjust the hyperparameters, and enhance the diagnostic accuracy and stability. Step 4.2, Real-time Fault Diagnosis: After obtaining real-time flight data of the UAV, features are extracted and fused in real time, and then fault diagnosis decisions are made. The real-time generated comprehensive feature vector is input into the trained fault diagnosis model. The model quickly calculates and outputs the probability distribution of the UAV's current fault state for various fault conditions. If the probability value of a certain fault type exceeds a pre-set diagnostic threshold, the UAV is determined to have experienced that type of fault, and the fault location and severity are determined based on the model's output. Step 4.3, Fault Prediction: Short-term fault prediction based on time series models: ARIMA, LSTM and other time series models are used to model the time series data of the operating parameters of key components of the UAV, predict parameter changes in the next few minutes to tens of minutes, and issue an early warning when the threshold is exceeded, so as to facilitate timely response; Long-term fault prediction based on machine learning and digital twins: Machine learning algorithms and digital twin models are integrated, and the model is trained by combining virtual and real data. It comprehensively uses information from multiple aspects to predict the probability of failure and remaining life in the next few hours to several days, which helps proactive maintenance.

[0038] This invention achieves an intelligent upgrade in fault diagnosis by constructing a digital twin model highly consistent with the physical entity of the UAV and integrating multimodal data with deep learning algorithms. This effectively solves the problems of low multimodal data utilization, insufficient model accuracy, and lack of predictive ability inherent in traditional methods. By integrating heterogeneous data from multiple sources such as sensors, vision, and thermal imaging, this technology overcomes the limitations of single-source diagnosis, significantly improving the accuracy and predictive ability of fault identification. Relying on the dynamic update mechanism of the digital twin model, the system response time is greatly shortened, enabling real-time monitoring and autonomous decision-making of UAV health status in complex environments, thereby significantly enhancing flight safety and operational efficiency.

[0039] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A fault diagnosis method for unmanned aerial vehicles (UAVs) based on digital twin sensor data fusion, characterized in that, Includes the following steps: 1) Unmanned aerial vehicle (UAV) multimodal data acquisition and preprocessing; 2) Constructing a digital twin of the drone; 3) Extract and fuse features from multimodal data; 4) Diagnose and predict drone malfunctions.

2. The method for fault diagnosis of UAV sensor data fusion according to claim 1, characterized in that, Step 1) of the UAV multimodal data acquisition and preprocessing process includes: 1-1) Selection and layout of drone fuselage sensors; 1-2) Multimodal data acquisition from UAV sensors; 1-3) Multimodal data transmission from UAVs; 1-4) Preprocessing of UAV multimodal data.

3. The method for fault diagnosis of UAV sensor data fusion according to claim 2, characterized in that, In step 1), during the data acquisition phase, multimodal data from the temperature sensor, rotation speed sensor, and environmental perception camera of the UAV monitoring system are collected. In the preprocessing stage, wavelet transform is first used to eliminate noise from temperature and speed sensors. Then, a dual long short-term memory network is used to fill in missing values ​​in the time series data. Subsequently, a normalization algorithm is used to achieve scale unification of multi-source data. Finally, joint calibration technology is used to complete the spatiotemporal alignment and system-level calibration of temperature and speed sensor data, ensuring accurate synchronization of multimodal data in the temporal and spatial dimensions.

4. The method for fault diagnosis of UAV sensor data fusion according to claim 3, characterized in that, Step 2) The process of constructing a digital twin of the drone includes: 2-1) Construction of digital twin geometric model of UAV; 2-2) Construction of a digital twin dynamics model for unmanned aerial vehicles (UAVs); 2-3) Physical attribute modeling of UAV digital twin models; 2-4) Development and real-time mapping of digital twin data interfaces; 2-5) Digital twin model verification and calibration.

5. The method for fault diagnosis of UAV sensor data fusion according to claim 4, characterized in that, Step 2) Specifically, a geometric model of the UAV is built using laser scanning and CAD modeling software; then, a dynamic model of the UAV is built based on its dynamic parameters; and a physical property model is constructed for each component of the UAV based on its materials and characteristics. Then, a standardized data interface was developed to achieve real-time mapping between multimodal data and digital twins.

6. The method for fault diagnosis of UAV sensor data fusion according to claim 5, characterized in that, Step 3) The multimodal data feature extraction and fusion process of the UAV includes: 3-1) Feature extraction of time series data, vibration data, and other data; 3-2) Multimodal data fusion of time series data, vibration data, and time series data is performed using a feature fusion network based on the attention mechanism.

7. The method for fault diagnosis of UAV sensor data fusion according to claim 6, characterized in that, Step 3) Specifically, in the multimodal data feature extraction stage, the improved YOLOv8 combined with the attention mechanism is used to extract target features from the UAV image data, and FFT and autoencoder are used to process vibration data; in the multimodal data fusion stage, a network is built based on the attention mechanism to dynamically allocate the weights of each modality data and splice them to form a comprehensive feature vector. Through reinforcement learning, the weights are adjusted and multimodal cross-validation is performed to achieve complementary data advantages to support accurate fault diagnosis of UAVs.

8. The method for fault diagnosis of UAV sensor data fusion according to claim 7, characterized in that, Step 4) of the UAV fault diagnosis and prediction process includes: 4-1) Training a UAV fault diagnosis model based on a deep learning model; 4-2) Real-time fault diagnosis of UAVs based on multimodal data mapped in real time from UAV digital twins; 4-3) Predict short-term drone failures based on time series models, and predict long-term drone failures based on machine learning and digital twins.

9. The method for fault diagnosis of UAV sensor data fusion according to claim 8, characterized in that, Step 4) Specifically, a high-precision fault diagnosis model is constructed based on deep learning and ensemble learning algorithms. When diagnosing UAV faults in real time, the collected multimodal data is preprocessed, feature extracted and fused, and then input into the fault diagnosis model. If the threshold is exceeded, the fault is determined and a detailed alarm is output. In the prediction stage, the short-term parameter trend and long-term potential fault risk prediction are realized by combining time series model and digital twin with machine learning.

10. A fault diagnosis system for UAV sensor data fusion, applied to the method of claim 9, characterized in that, include; Data acquisition and processing module: used for multimodal data acquisition and preprocessing of UAVs; Drone Digital Twin Building Module: Used to build drone digital twins; Data feature extraction and fusion module: used to extract and fuse features from multimodal data; Fault diagnosis module: Used to diagnose and predict drone faults.

Images