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

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

Bahía Blanca

Congreso; IX Congreso Argentino de Ingeniería Química; 2017

Robust DataReconciliation strategies provide unbiased variable estimates in the presenceof a moderate quantity of gross errors.  This quantity is known as the break down pointof the robust methodology. Systematic errors which persist in time, as biasesor drifts, overcome this bound and the quality of the estimates gets worse. Inthis cases the fast detection of the faulty sensor is relevant in order toapply corrective actions and avoid biases on the solution of the datareconciliation procedure. At this work a new methodology for robust variableestimation and systematic measurement errors detection and classification is appliedto model equation that involve measurements with delay. The methodology makesuse of the Robust Measurement Test, which was recently proposed, to detectoutliers, and the Robust Linear Regression of the data contained in a movingwindow to distinguish between biases and drifts. This strategy has been devisedwith the aim of being part of the real-time optimization loop of an industrialplant, therefore the Tennessee Eastman process operating under steady state isused as benchmark. Results highlight the performance of the proposedmethodology to detect and identify outliers, biases and drifts for linear andnon-linear benchmarks.