Research On Intelligent Fault Diagnosis Method Of UAV Flight Control System Driven By Small Sample Dat

Unmanned Aerial Vehicle(UAV)is a type of high-end complex equipment that integrates advanced technologies such as optics,mechanics,electronics and artificial intelligence,and has been widely used in both civil and military fields.Once the flight control system(FCS),which is a core component of the UAV,malfunctions,it will seriously affect the safety and stability of UAV flight,and may lead to significant economic losses and military impacts.Fault diagnosis techniques can significantly improve the autonomous protection capability of UAVs through timely detection and identification of FCS faults.Currently,intelligent diagnosis methods represented by deep learning have become a research hotspot in the field of fault diagnosis by virtue of their advantages in feature extraction and pattern recognition.However,due to the complex composition of UAV FCS,the harsh service environment,and the difficulty in acquiring fault samples,the fault data exhibits characteristics such as small sample size,imbalanced class distribution,and inconsistent data distribution.This has led to traditional deep learning diagnostic models facing challenges such as insufficient samples,difficulty in extracting fault features,and poor model generalization performance,which seriously restrict the wide application of intelligent fault diagnosis methods in the UAV field.To this end,this paper takes the FCS of fixed-wing UAV as the research object,and conducts in-depth analysis of the typical faults of three key components,namely sensors,control surfaces,and servos.Considering the difficulty of obtaining failure samples of different components under different stages and working conditions,this paper explores intelligent fault diagnosis methods for scenarios involving zero sample,unbalanced small samples,and complex working conditions with small samples.Additionally,the fault diagnosis results driven by small sample data are mapped and fused with the fault instances in the ontology knowledge base,achieving the whole process of diagnosis from fault state recognition to fault knowledge inference analysis.The main research of the paper includes:(1)To address the problems of scarce fault samples and poor robustness of models to different faults in early detection of sensors,a Convolutional Variational AutoencoderGenerative Adversarial Network(CVAE-GAN)model is proposed.The model integrates the advantages of CVAE and GAN in data generation based on the characteristics of the zerosample fault detection task.By adversarial training,the generator learns the potential distribution of normal status data of multiple sensors without fault samples.During testing,the generator cannot reconstruct unseen fault data properly and produces large reconstruction errors,which are then compared with adaptive thresholds to perform finegrained and real-time detection of different types of faults in multiple sensors of the FCS.The proposed method is validated using real UAV sensor data and various injected faults,and threshold-dependent and threshold-independent evaluation metrics show that the proposed method has superior detection performance compared to other mainstream models.(2)To address the problems of unbalanced fault classes and incomplete fault features of control surfaces in UAV FCS,a Siamese Hybrid Neural Network(SHNN)model is proposed.The model has two structurally identical One-dimensional Convolutional Neural Network and Gated Recurrent Unit(1D CNN-GRU)sub-networks,which extract mixed fault features that fuse spatial and temporal information from each of the input sample pairs,respectively,and calculate the similarity between the two feature vectors using the L1 distance function.In addition,a weighted binary cross-entropy loss function is introduced to optimize the classification performance of the model on majority and minority fault samples.Finally,the fine-tuning strategy is employed to enhance the model’s adaptability to fault data of control surfaces under environmental disturbances.The proposed method is validated using a real flight dataset of a fixed-wing UAV,and experimental results show that the SHNN model outperforms existing fault diagnosis methods in terms of both local and global performance.(3)A Meta-learning Fault Diagnosis(MLFD)method is proposed to address the problems of frequent bearing failures,inconsistent fault data distribution and poor model generalization caused by the servos in the UAV flight control system under complex operating conditions.First,the raw vibration signals of the servo bearings collected under different working conditions are transformed into time-frequency images,which are randomly sampled according to the meta-learning framework protocol to form meta-learning tasks.Then,during the meta-training process,the MLFD model optimizes the initialization parameters of the model by using multiple fault classification tasks of known working conditions to obtain prior knowledge.Finally,the prior fault knowledge is applied to achieve fast and accurate diagnosis of servo bearing faults under unseen working conditions.Comprehensive validation analysis is conducted based on a public bearing dataset,and the experimental results show that the proposed model has advantages over existing methods in diagnosing servo bearing faults with limited samples under complex working conditions.(4)Aiming at the deficiency of small sample data-driven intelligent fault diagnosis models that can only output quantitative identification results without exploring root causes,an intelligent FCS fault diagnosis framework incorporating ontology knowledge is proposed.Firstly,the fault knowledge of UAV FCS is analyzed and summarized,and the fault diagnosis ontology knowledge base is established.Then,the fault states identification results of the small sample data-driven models are associated with fault instances in the ontology knowledge base by a semantic mapping method,and the fault diagnosis knowledge inference is realized by an inference engine to improve the accuracy and intelligence level of the FCS fault diagnosis.Finally,integrating the above research results,a set of system for intelligent diagnosis of UAV FCS faults is designed and developed.The system includes four core functions: data acquisition and storage,system operation management,fault diagnosis based on small sample data,and inference analysis based on the fault ontology knowledge base.