Wavelet analysis and auto-associative neural network based fault detection and diagnosis in an industrial gas turbine

This work presents a Fault Detection and Diagnosis (FDD) system that uses a combination of discrete wavelet transform and auto-associative neural network. The neural network is trained by Levenberg-Marquardt (LM) algorithm. As a case study, it considers a 5.2MW Siemens Taurus 60S industrial gas turbine. The work is unique in the sense that it addresses the signals in the gas path, generator coils, lubrication system, and vibration sensors. Real data are used to train and corroborate the models. In order to test validity of the FDD system, we used abrupt and incipient faults generated by implanting controlled bias to the normal signals. Results show that the proposed method could detect a 10% bias with an average true detection higher than 95% while the diagnoses performance is in the range of 96 to 100%. Since it is designed and tested based on real data, it can be considered competent for practical use.