Predicting research of mechanical gyroscope life based on wavelet support vector

Mechanical gyroscope has characters of high cost and few quantity. In order not to take 1:1 experiment to evaluate its performance and life, we propose a life prediction method that combined wavelet analysis and support vector machine (SVM). First, we use wavelet analysis to do pretreatment on life data to reduce some interference information to improve the data smoothness and weaken data randomness. Then we use SVM to model those preprocessed data. The choosing of model parameters is based on genetic algorithm to search optimal value globally and get prediction data. In order to prove the superiority of this model, we choose the life data of dynamically tuned gyroscope in literature. SVM model and WA-SVM model were used to predict gyroscope´s life and their results were compared. We give root-mean-square error of different model to make the comparison more obviously. The results show that better prediction effect and its root-mean-square error is just 3.47%.