Distillation Process Fault Detection by CUSUM Test

Abstract

In this work, The CUSUM (Cumulative Sum) test is employed to determine the real operating conditions of nonlinear processes as a separation unit. To do this, the normal behavior of the system is first predicted by means of a Nonlinear Auto-Regressive Moving Average with eXogenous input (NARMAX) model and then the residual between the real and estimated output of system is evaluated at each time by taking new observations of the process output under consideration. Finally, the thresholds on the CUSUM test are derived such that when the updated residual exceeds the thresholds an alarm is triggered.