E. Tafazzoli, M. Alavi, and M. Saif (Canada)
Least Square Support Vector Regression, Fault Detection, Fault Tolerant
Many fault detection and fault tolerant systems are de signed for processes in which there is an analytical model for the system. If a model is not available then data-driven approaches are considered as an alternative method. In this paper we propose a data driven approach for sensor fault detection and accommodation in dynamic systems. Least square support vector machine (LSSVM) is implemented and the system output is predicted and used in control loop to accommodate sensor fault. Using LSSVM regression, a function can be approximated by training the model with available training data. LSSVM is used in the structure of a recurrent predictive model of a sensor output which is used in fault detection and fault tolerance (FDT). A fault toler ant approach is proposed and tested on a three tank system in which the output of the system is controlled in the de sired operating region in presence of a sensor fault. The proposed approach is successfully applied and tested on a three tank system model in case of abrupt sensor fault.
Important Links:
Go Back
