Detection of sensor failure in a palm oil fractionation plant using artificial neural network

Abstract

Artificial neural network by virtue of its pattern recognition capabilities has been explored to systematically detect failures in process plants. In this paper, a two-stage approach integrating a neural network dynamic estimator and a neural network fault classifier is proposed to overcome the problem of malfunction in sensors. The process estimator was designed to predict the dynamic behavior of the normal or unfaulty operating process even in the presence of sensor failures. As such, any variables that are related or influenced by the failures under investigation cannot be used as model inputs. The difference between this estimated “normal� and the actual process measurements, termed the residuals are fed to the classifier for fault detection purposes. The classifier that was founded on feedforward network architecture then identifies the source of faults. The estimator was constructed using externally recurrent network where the estimated values are fed back to the input neurons as delayed signals. The scheme was implemented to detect sensor failure in a palm oil fractionation process. To generate the required simulation data, HYSYS.Plant dynamic process simulator was employed. The proposed scheme was successful in detecting pressure and temperature sensor failures introduced within the system.