Adaptive artificial intelligence for automatic identification of defect in the angular contact bearing

A reliable process making use of adaptive filtering, adaptive wavelet transform (AWT) and adaptive neuro-fuzzy inference system (ANFIS) is proposed for automatic identification of defect in the bearing. The process starts with denoising the signal using a filter designed from the noise part in the signal itself using the least mean square algorithm. Thereafter, AWT is applied to denoised signal to produce AWT scalogram. AWT scalogram clearly gives time location information corresponding to burst produced due to defect. For carrying out AWT, a wavelet is designed from the burst produced to defect using the least square fitting method. Then instantaneous energy (IE) of AWT scalogram is computed. For modeling of ANFIS, features are extracted from signal denoised by adaptive filtering, AWT scalogram and IE graph. The dimension of features is reduced by kernel principal component analysis. ANFIS modeling is carried out using low-dimensional kernel principal components. Finally, defect identification is carried out by applying test features kernel principal components to ANFIS. The proposed method has a recognition rate of 100%. Major benefit of the proposed method is that it requires very less training data and also does not require any optimization algorithm, usually required to find optimum parameters of a pattern recognition model. The method proposed in this work can fulfill the industrial requirement of having sophisticated and accurate automatic defect identification model.