SVM Approach for Sensor Fault Detection in Multi-Variable Processes

An approach for sensor fault detection in case of multi-variable processes with time-delays is presented. Combining support vector machines (SVM) regression algorithm with data driven information fusion technique, a system scheme of sensor fault detection, isolation and data recovery based on support vector machines generalized observer (SGO) is proposed. Each key sensor is dedicated with an observer driven by process inputs and outputs, except for the sensor to be monitored. Data is validated by comparing sensor outputs with the observer outputs. Experiments of multi components distillation column (MCDC) showed that the above method can detect sensor fault in processes effectively and has preferable robustness.