A Robust Methodology for the Sensor Fault Detection and Classification of Systematic Observation Errors

Abstract

Robust Data Reconciliation enhances the quality of variable estimates when the data set contains a moderate proportion of atypical observations. But if systematic errors that persist in time, i.e. biases and drifts, are not detected, the break down point of the estimates is exceeded and results get worse. In this work, a new methodology based on the concepts of Robust Statistics is presented to deal with this problem. The strategy computes robust variable estimates, classifies the systematic measurement errors, and provides corrective actions to avoid the detrimental effect of biases and drifts until the sensor is repaired. The performance of the methodology is evaluated for the steady state operation of linear and non-linear benchmarks. Results demonstrate that its use significantly improves the estimates accuracy