Comparison between Multi-Layer Perceptron and Radial Basis Function Networks for Predicting Reliability and Availability: A Case Study
Conference: Creative Trends in Engineering and Technology
The rapidly growing field of Artificial Neural Network (ANN) applications has witnessed several admirable\ncontributions. This paper presents application of ANN for predicting the improved values of reliability and availability after\nsuccessful implementation of reliability centered maintenance (RCM) policy in a thermal power plant. In this study, the\npredictive performance of two Artificial Neural Networks, viz-a-viz Radial Basis Function (RBF) and Multi-Layer\nPerceptron (MLP) were compared. The reliability and availability of any component and/or sub-system can be calculated\nmathematically (using traditional approach) by knowing the two parameters i.e. outage hours and number of faults. However\nin the said problem, after implementation of new maintenance policy, outage hours decrease and this is the only known\nparameter. The second parameter (number of faults) is unknown. The importance of MLP based and RBF based ANN model\nis to predict reliability and availability on the basis of only one parameter known. The test results showed that outcome of\nproposed ANN model is in good agreement with desired or actual results. The MLP network produced a more fitted output to\nthe cross validation data set than the RBF network. This application of ANN helped knowing foreseen indices and\nconvincing maintenance department about benefits of implementation of RCM.
CTET - 2016