Fault prediction/diagnosis and sensor validation technique for a steam power plant

Abstract

An intelligent sensor validation and fault prediction/diagnosis technique for a typical steam power plant is proposed and studied. An auto-associative Artificial Neural Network (ANN) is trained to examine the consistency of the overall simulated data and allocate a confidence level to each signal.The same set is used to replace the missing or faulty data with a close approximation. For fault prediction and diagnostic system a feed-forward ANN with extra linear connections is trained to recognise faulty and healthy behaviour of the steam cycle for a wide range of operating conditions. Both ANNs are tested with unseen data sets, including combined scenarios of the partially failed system to assess fault prediction capability of the proposed ANN. It is concluded that a significantly more reliable sensor reading and a highly accurate fault prediction/diagnosis system is achieved. © 2014 2005 Taylor and Francis Group LLC.