Palash Pal¹, Rituparna Datta², Aviv Segev², and Alec Yasinsac², ¹Burdwan University, India and ²University of South Alabama, USA

System and sub-system maintenance is a significant task for every dynamic system. A plethora of approaches, both quantitative and qualitative, have been proposed to ensure the system safety and to minimize the system downtime. The rapid progress of computing technologies and different machine learning approaches makes it possible to integrate complex machine learning techniques with maintenance strategies to predict system maintenance in advance. The present work analyzes different methods of integrating an Artificial Neural Network (ANN) and ANN with Principle Component Analysis (PCA) to model and predict compressor decay state coefficient and turbine decay state coefficient of a Gas Turbine (GT) mounted on a frigate characterized by a Combined Diesel-Electric and Gas (CODLAG) propulsion plant used in naval vessels. The input parameters are GT parameters and the outputs are GT compressor and turbine decay state coefficients. Due to the presence of a large number of inputs, more hidden layers are required, and as a result a deep neural network is found appropriate. The simulation results confirm that most of the proposed models accomplish the prediction of the decay state coefficients of the gas turbine of the naval propulsion. The results show that a consistently declining hidden layers size which is proportional to the input and to the output outperforms the other neural network architectures. In addition, the results of ANN outperforms hybrid PCA-ANN in most cases. The ANN architecture design might be relevant to other predictive maintenance systems.

Condition based maintenance, Neural Network, Deep neural network, Principle Component Analysis(PCA), Naval propulsion