Research On Relevance Vector Machine For Small Samples Fault Diagnosis And Prediction

At present some commonly used fault diagnosis methods and fault prediction methods are based on the large samples data, however we often employ the small samples data in the practical problems. The traditional fault diagnosis and fault prediction methods are not fit into to solve the problem of small samples. Relevance Vector Machine(RVM) is a newly proposed on the basis of Support Vector Machine(SVM)model, this model is more suitable for small samples problems. RVM has been applied to the voice and image processing, medical diagnosis, pattern classification etc. This paper has mainly done following jobs according to the application of small samples fault diagnosis and prediction:(1) Analysis the research status of the fault diagnosis and fault prediction.(2) Research the basic RVM and analysis the RVM algorithm’s feasibility in the fault diagnosis and fault prediction.(3) According to the kernel function selection’ blindness and the parameters in the kernel function in RVM, combination of Gauss kernel function and Cauchy kernel function structure mixed kernel function, the parameters of the mixed kernel function use the cuckoo algorithm optimized choice, then proposed based on the cuckoo algorithm of mixed kernel relevance vector machine model(CS-RVM), then through the simulation experiment proved the validity of the CS-RVM.(4) Applied the CS-RVM in aero-engine gas path diagnostic and diesel engine oil,through the comparison with RVM and the Differential Evolution of mixed kernel relevance vector machine model(DE-RVM), proved the validity of the CS-RVM in the small fault diagnosis and prediction.