Detecting Changes in a Distillation Column by Using a Sequential Probability Ratio Test

In chemical plants, a reliable detection of anomalies is important for a safe operation. To this end, a fault detection (FD) method of abnormal operations applicable to a chemical process is presented in this paper. This method couples an Artificial Neural Network-Multi-Layer Perceptron (ANN-MLP) with a statistical module based on the sequential probability ratio test (SPRT) of Wald, for the analysis of the process residuals. To detect a change, this combination uses the mean and the standard deviation of the residual noise obtained from applying a NARX (Nonlinear Auto-Regressive with eXogenous input) model. The FD effectiveness is tested under real abnormal circumstances on a real plant as a distillation column. The experimental results obtained show the relevance of this method for the fast detection and the monitoring of this chemical process.