ON-LINE FAULT SENSOR DETECTION AND IDENTIFICATION USING A ROBUST STATISTICS BASED METHODOLOGY

SANCHÉZ, MABEL C.; CHÁVEZ GALLETI ROBERTO J.; LLANOS, CLAUDIA E.; MARONNA, RICARDO A.

Bahía Blanca

Congreso; IX Congreso Argentino de Ingeniería Química; 2017

PLAPIQUI (UNS-CONICET)

Robust Data Reconciliation strategies provide unbiased variable estimates in the presence of a moderate quantity of gross errors. This quantity is known as the break down point of the robust methodology. Systematic errors which persist in time, as biases or drifts, overcome this bound and the quality of the estimates gets worse. In this cases the fast detection of the faulty sensor is relevant in order to apply corrective actions and avoid biases on the solution of the data reconciliation procedure. At this work a new methodology for robust variable estimation and systematic measurement errors detection and classification is applied to model equation that involve measurements with delay. The methodology makes use of the Robust Measurement Test, which was recently proposed, to detect outliers, and the Robust Linear Regression of the data contained in a moving window to distinguish between biases and drifts. This strategy has been devised with the aim of being part of the real-time optimization loop of an industrial plant, therefore the Tennessee Eastman process operating under steady state is used as benchmark. Results highlight the performance of the proposed methodology to detect and identify outliers, biases and drifts for linear and non-linear benchmarks.