APPLICATION OF NEURAL NETWORKS FOR GROSS ERROR-DETECTION

The problem of detecting gross errors in measurements arising from faulty sensors is an important one in operating chemical plants. The problem has applications in modeling, control, optimization and maintenance. Traditionally statistical methods have been used for this purpose. In this study, we explore the use of artificial neural networks (ANN) for solving this problem. Using Monte Carlo simulation, we address the following issues in applying ANNs for gross error detection: (a) type of input/output patterns used for training and their preprocessing; (b) parameters that affect performance; (c) strategy used for detecting multiple gross errors. We compare the performance of the ANN with that of statistical methods on a practical example and show that ANNs offer a competing alternative method for gross error detection.