Fault Identification in Chemical Processes Through a Modified Mahalanobis-Taguchi Strategy

摘要

On the chemical process industries an appropriate identification of faults has become at present one of the most critical and challenging PSE activities. Recently, Taguchi and Jugulum (2002) described a new data based approach for diagnosis and forecasting using multivariate data, the so-called Mahalanobis-Taguchi System (MTS). MTS is considered to be a non-parametric approach, and has been applied with success in a number of different fields. In the present work, we apply a modified version of MTS (referred as MMTS) to perform fault identification in chemical processes. The performance of our MMTS approach is evaluated by its application to a continuous-stirred-tank-reactor (CSTR) simulated by Kano et al. (2002) and comparing the results obtained here with those provided by these authors. The results thus obtained show that indeed MMTS seems to be a quite promising data based approach for on-line fault detection in chemical processes.