A Neural Networks and Rejection Options Approach for Intelligent Novelty Motor Anomaly Detection

The main interest of this article is to present a new approach for induction machine fault diagnosis system. This technique combines a discriminate analysis and novelty detection based on the rejection notion in statistical neural networks in order to recognize undefined fault types and reject confused samples. The effectiveness of this method is extremely related to the experimental database application which has been worked out in STA laboratory in collaboration with industrial company SITEX KSAR HELLAL. Indeed, we have trained both the radial basis function neural networks and probabilistic neural networks for induction machine fault recognition with rejection option. Then, the experimental results demonstrate that the proposed intelligent fault diagnosis system automatically adapts to unknown situation based on its classification performance and novelty detection thanks to rejection integration.