A Bayesian approach to gross error detection in chemical process data

Abstract

Tamhane, A.C., Iordache, C. and Mah, R.S.H., 1988. A Bayesian approach to gross error detection in chemical process data. Part II: Simulation results. Chemometrics and Intelligent Laboratory Systems , 4: 131–146. The performance of the gross error detection scheme based on the Bayesian test is evaluated using Monte Carlo simulation methods. Effects of selected control factors (implementation options) and noise factors (e.g., violation of assumptions and misspecification of priors) are studied. A comparison is made with the gross error detection scheme based on the non-Bayesian measurement test of Mah and Tamhane. The Bayesian scheme is found to be relatively robust. It performs better than the measurement test scheme when gross error occurrences are not infrequent. However, its performance characteristics converge rather slowly and hence accurate prior estimates of the various unknown parameters are necessary before the method can be put to practical use. In conclusion, the Bayesian approach offers the promise of improving gross error detection and identification capabilities by using past failure data. Its technical feasibility is demonstrated by this investigation, but much remains to be done to make it a practical method.