Research On Active Fault Tolerant Control Method Of UAV Based On Multi Model

With the development of quadrotor unmanned aerial vehicle(Quadrotor UAV)technology,UAV can replace people in some harsh environments and has become an indispensable role in some special jobs.However,due to the failure of its actuators,it not only reduces the reliability of the drone during operation,but also brings unpredictable negative effects.In view of this,this article aims at the problem of quadrotor UAV single actuator failure and two actuators failure at the same time.According to the multi-model active fault-tolerant control idea,the system performs a systematic study on the single or concurrent failure of the actuator.As follows:1)Starting from the problem of quadrotor UAV single actuator failure,a multi-model active fault tolerance method based on fault interval is studied.First,according to the actual fault of the actuator,the system performance is analyzed,and the system performance tolerance is processed in a segmented manner.The problem of fault tolerance failure due to the failure of a single performance index to fully account for all faults is reasonably solved.Second,according to different fault intervals,all actuator fault models and corresponding control laws are divided into model subsets.When the actuator fails during system operation,only the fault subset corresponding to the fault section matches the corresponding fault model according to the model mismatch degree.The fault-tolerant control law solves the time-consuming problem of matching fault models one by one.Finally,simulation experiments are performed to verify the effectiveness and superiority of the method.2)Aiming at the problem of insufficient fault-tolerant timeliness in establishing model subsets by fault region of actuator,a multi-model active fault-tolerance method based on stack self-coded fault classification is proposed.This method divides all fault models and corresponding control laws into three model subsets according to the three types of faults that the actuator is prone to send;using fault data offline to train deep classification networks,and online fault classification,only within the corresponding fault subsets.Match the fault model and call the control law.Simulation results show that using deep learning to classify networks not only does not require too much prior knowledge,but also has high accuracy of fault classification.At the same time,it can effectively reduce the time consumed by fault-to-matching,and it can quickly fault-tolerate faults.3)Aiming at the concurrency failure of two actuators of quadrotor UAV and the classification accuracy of classification network due to noise interference from outside the drone,a multi-model active fault-tolerant control method based on SDA two-stage classification concurrency failure was proposed.First,a single actuator fault model sub-base 1 and an actuator concurrent fault model sub-library 2 are established,and two model sub-libraries are divided into model subsets according to the fault types;Secondly,the classification network is trained using data with certain statistical noise is used to reduce the sensitivity to the interference when the fault types are identified online;Furthermore,SDA single-fault/concurrent faults and two-fault classification of different types of faults accurately match the corresponding models of faults in the corresponding model subsets.Finally,the effectiveness and superiority of this method are verified by experimental simulation.