On-line process monitoring using a robust statistics based methodology

Abstract

Robust Data Reconciliation strategies provide unbiasedvariable estimates in the presence of a moderate quantity of measurement grosserrors. Systematic errors which persist in time, as biases or drifts, overcome thisquantity causing the deterioration of the estimates. This also occurs due tothe presence of process leaks. The fast detection of those faults avoids theuse of biased solutions of the data reconciliation procedure, and allows toperform quick corrective actions. In this work, a methodology for leakdetection is incorporated into a robust data reconciliation procedure thatdetects and classifies systematic observation errors. The strategy makes use ofthe Robust Measurement Test, to detect outliers and leaks, and the RobustLinear Regression of the data contained in a moving window to distinguish betweenbiases and drifts. The methodology is applied for two benchmarks extracted fromthe literature. Results highlight the performance of the proposed strategy.