Detection of abrupt changes of total least squares models and application in fault detection

This paper deals with detection of parameter changes of total least squares and generalized total least squares models and its application in fault detection and isolation. Total least squares and generalized total least squares are frequently used to model processes when all measured process variables are corrupted by disturbances. It is therefore of practical interest to monitor processes and detect faults using the total least squares and generalized total least squares as well. The local approach for detection of abrupt changes is adopted as a computational engine for the change detection. The effectiveness and robustness of the proposed algorithm in fault detection and isolation are demonstrated through Monte Carlo simulations: a pilot-scale experiment and sensor validation of an industrial distillation column.