Non intrusive operational health identification of in-service electrical machine

Reliable and fault free operation of electrical machines is presently an area of extensive research. Significant progress has been made in identification of faults in electrical machines employing intrusive and non intrusive methods. In most of the published work, supervised learning technique is form the core approach for identification and classification of faults. This technique is more suitable for new commissioned machines. For the development of diagnostic algorithm, complete set of fault signatures over the machine's life span are required. However, in case of machines which are already in-service, complete set of signatures reflecting different fault states can't be made available for development of diagnostic algorithm. This imposes a sever limitation on diagnosis algorithms. In this paper, a proposed method is presented which can be employed for the identification of operational health state of in-service machines. The proposed method is based on the comparison of principal components. Fault diagnosis is based on features extracted from time frequency representation of fault current.