Fault Diagnosis for Three-phase Asynchronous Machine using a Modified SVDD

In this paper, we considered the issue of faults diagnosis for induction motors with many possible functioning modes. The proposed approach is based on data classification using an improved version of the Support Vector Domain Description well known as SVDD. Basically, The SVDD is an efficient technique dedicated for one-class problems (called also novelty detection problems). Though, since we are dealing with systems able to operate in various modes (multi-modes), a multiclass scheme remains crucial to ensure a proper diagnosis for such systems. In this paradigm, we extended the standard SVDD algorithm in order to support the multi-classification problems. We proposed for this purpose a new decision criterion called Maximal Kernelized Similarity (MKS). To prove its effectiveness, the proposed approach is applied for fault diagnosis of three-phase asynchronous machine, and employed to perform a monitoring procedure over the real time. Moreover, a comparison against other conventional algorithms such as RBF, SVM and standard SVDD was made.