ABSTRACT
Avoiding unexpected and costly repairs for mechanical systems, specially wind turbines, brought up the science of prediction named Prognostics and Health management. This technique calculates the remaining useful life, and consequently, gives the opportunity to preserve normal performances of rotary machines. In this paper, an intelligent online data driven PHM method has introduced and applied to the experimental run-to-failure data. Firstly, a reduction gearbox test rig run series of experiments under specific speed and load levels to create a set of vibrating degradation data. Extracted Cumulative indicators are then concatenated to form a health index. Afterwards, a hybrid algorithm of Adaptive Neuro Fuzzy Inference System and Particle Filtering is built to evaluate the health index signal. ANFIS based on the auto-encoder hypothesis can model the degradation trend for prediction. Then, particle filtering is carried out for a-step-ahead behavior prediction. The performance of the result demonstrate that ANFIS-PF can successfully predict the degradation behavior with 95% confidence boundary.
