Thermal image enhancement using bi-dimensional empirical mode decomposition in combination with relevance vector machine for rotating machinery fault diagnosis

In this study, a novel fault diagnosis system for rotating machinery using thermal imaging is proposed. This system consists of bi-dimensional empirical mode decomposition (BEMD) for image enhancement, a generalized discriminant analysis (GDA) for feature reduction, and a relevance vector machine (RVM) for fault classification. Firstly, the thermal image obtained from machine conditions is decomposed into intrinsic mode functions (IMFs) by using BEMD. At each decomposed level, the IMF is expanded and fused with the residue by gray-scale transformation and principal component analysis fusion technique, respectively. The enhanced image is then formed by the improved IMFs in reconstruction process. Subsequently, feature extraction is applied for the enhanced images to obtain histogram features which characterize the thermal image and contain useful information for diagnosis. The high dimensionality of the achieved feature set can be reduced by GDA implementation. Moreover, GDA also assists in the increase of the feature cluster separation. Finally, the diagnostic results are performed by RVM. The proposed system is applied and validated with the thermal images of a fault simulator. A comparative study of the classification results obtained from RVM, support vector machines, and adaptive neuro-fuzzy inference system is also performed to appraise the accuracy of these models. The results show that the proposed diagnosis system is capable of improving the classification accuracy and efficiently assisting in rotating machinery fault diagnosis.