Study On Fault Diagnoses Method Based On Rudder System

As a servo control system,rudder system is widely used in some areas of sophisticated aerospace like military,ship and autopilot etc.It plays a leading role in the process of equipment operation,and its performance directly affects the operation status of the whole equipment.Therefore,the research on fault diagnosis and identification technology of rudder system is particularly important.The existing rudder system testing equipment is portable,easy to operate and automatic,but the later data analysis is not a test link,the two steps are completely disconnected.In order to realize efficient fault diagnosis based on the tested data,the machine learning algorithms are introduced to realize system analysis.The analog circuit fault diagnosis method is studied at first,in order to eliminate the complex and changeable analog circuit fault.The fault types of analogy circuit are complex and the collected features are a large data bulk.For that reason,we have to introduced Laplace score algorithm preprocess the original data sets and sort the samples on the basis of the importance of features.After that,the sorted samples are fed into the optimized support vector machine by inputting iteratively and incrementally.The training process can tradeoff between sample dimension,training accuracy and training efficiency,and the optimal diagnosis model is finally found among its iteration.Among them,the perturbed particle swarm optimization algorithm is selected to optimize the Support Vector Machine classifier to avoid the dilemma of local optimization.Then,we start to study the data analysis method for system evaluation.Aiming at the unbalanced characteristics of the testing data,resampling technology is introduced to solve the performance degradation problem of traditional classification algorithm in unbalanced learning.Firstly,a weighted oversampling algorithm based on clustering is proposed to realize the synthesis of minority class samples.During the generation mechanism,support vector is mainly considered and been weighted hardly.Then,the Synthetic Minority Oversampling Technique is used to generate minority class samples on the basis of Hierarchical Clustering algorithm.This algorithm effectively avoids the generation of a large number of noises,however,the information carried by the samples is not comprehensive enough,which is only suitable for the realization of the system troubleshooting.In order to generate high quality minority class samples of which contain much more sufficient information,a more reasonable adaptive oversampling technology based on information samples is proposed to further improve the performance of classifier and realize fault diagnosis and location.This algorithm is better than before,in the selection of information samples and the weight distribution mechanism.In the technology,we take the decision hyperplane as a reference to seek informative samples and assign the distribution of weights.The larger the Euclidean distance between sample points and hyperplane is the more importance the point is,and the higher the weight is.In addition,the proportion of informatic examples and resampling rate among minority class samples as well as the key parameters in the Support Vector Machine classifier are not constants but optimized by the Whale Optimization Algorithm during the model training process.Finally,the new generation samples is effective enough to finish fault isolation and the final model has higher performance in fault diagnosis and location.The developed diagnosis model in analogy circuit is verified in two common analog filter circuits.The experimental results show that the fault diagnosis accuracy is high and the performance of the classifier is superior.In the experiment of fault diagnosis and identification of rudder system based on test data,the comparison results among the proposed method and the classical common technologies reflect the efficiency and performance superiority of the developed model.