Fault Detection by Using a Reliable Neural Model: Application to a Chemical Reactor

In this paper a real-time system for detecting changes in dynamic systems is designed. The cumulative sum (CUSUM) or the Page-Hinkley test is intended to reveal any drift from the normal behavior of the process. The process behavior under its normal operating conditions is established by a reliable model. In order to obtain this reliable model for the process dynamics, the black-box identification by means of a NARMAX (Nonlinear Auto-Regressive Moving Average with eXogenous input) model has been chosen in this study. It is based on the neural network approach. This paper will show the choice and the performance of this neural network in the training and the test phases. A study is related to the inputs number, and of hidden neurons used and their influence on the behavior of the neural predictor. Three statistical criterions are used for the validation of the experimental data. After describing the system architecture and the proposed methodology of the fault detection, we present a realistic application like a reactor in order to show the technique's potential. The purpose is to detect the change presence, and pinpoint the moment it occurred.

Keywords: fault detection, process safety, neural network, reactor, narmax model, Page-Hinkley.