Research And Application Of UAV Gyro Fault Diagnosis Method Based On Extreme Learning Machine

With the wide application of drones in life,the safety of drones is becoming more and more important.The core content is how to classify the gyro sensor faults of the drone.At present,it is a good method to identify the fault type by using extreme learning machine after the data has supervised learning input signals.However,in most cases,time domain data does not adequately characterize data failure characteristics,resulting in poor classifier recognition.Moreover,the input weight and the hidden layer threshold of the extreme learning machine are completely randomly initialized,and the parameters of the random initialization have a great influence on the accuracy of the classifier.Moreover,there are data in real-time learning in engineering applications,and it takes more time to use offline learning machine training for these classifiers that need to update data frequently.In order to solve the above problems,this paper extracts the data features from the wavelet packet and optimizes the completely random initialization parameters of extreme learning machine and realizes the online learning of the extreme learning machine classifier to realize the fault diagnosis of the drone gyroscope.This paper proposes a method of online constrained extreme learning machine combined with wavelet packet to realize the fault diagnosis of drone gyroscope.Firstly,using wavelet packet algorithm to extract the energy characteristics of time domain data can better distinguish faults and improve the effect of classifier training.Secondly,the constrained extreme learning machine replaces the parameter completely randomized initial selection scheme used by the extreme learning machine by randomly initiating parameters from the differential vector set of the output category,improving the generalization ability of the classifier while ensuring fast performance.Thirdly,in order to improve the online training speed of the constrained extreme learning machine,this paper proposes an online constrained extreme learning machine.The online constraint extreme learning machine ensures the generalization performance of the classifier while reducing the online training time.Finally,the online constrained extreme learning machine is used to realize the fault diagnosis simulation of the drone gyroscope,and the effectiveness of the UAV fault diagnostic device realized by the online constrained extreme learning machine is verified.The simulation results show that compared with the classical extreme learning machine algorithm,the constraint extreme learning function can improve the accuracy of fault diagnosis while improving the accuracy of 4%.Compared with the constrained extreme learning machine,the online sequence constrained extreme learning function can greatly reduce the time spent on the fault diagnosis online training while ensuring the fault accuracy.It shows the excellent performance of the online sequence constrained extreme learning machine.