Crack detection methodology in rotor bearing system by DWT based adaptive neuro-fuzzy inference systems

Abstract

Data-based machine learning methods in condition monitoring of high-speed rotors are gaining more importance due to the increase in computational speeds of computers in recent years. Machine learning methods can analyze and predict the behaviour of the system more accurately based on the methodology and truthfulness of trained data. The proposed research is a data-based crack detection technique in rotors based on Discrete Wavelet Transforms (DWT) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). DWT based ANFIS is used to de-noise the response of rotor with open cracks. A methodology is proposed to effectively train and validate ANFIS to de-noise the signal with DWT noise information. Detail-1 of DWT on ANFIS de-noised response is used to identify the location and severity of the crack in the overhung rotor. Two separate ANFIS are proposed to identify crack location and crack depth from DWT detail-1 coefficients of ANFIS de-noised response. It is found that proposed DWT Based ANFIS de-noise the signal with Root Mean Square Error (RMSE) of 3.96 × 10−8, detects the crack location with RMSE of 0.1656 and depth of crack with the RMSE of 1.321.