Comparison of Centralized Multi-Sensor Measurement and State Fusion Methods with an Adaptive Unscented Kalman Filter for Process Fault diagnosis

This paper investigates the application of centralized multi-sensor data fusion (CMSDF) technique to enhance the process fault detection. Adaptive Unscented Kalman Filter (AUKF) is used to estimate the process faults of the simulated continuous stirred tank reactor (CSTR) benchmark. Currently there exist two commonly used centralized multi-sensor data fusion methods for Kalman filter including centralized measurement fusion and centralized state-vector fusion. The measurement fusion methods directly fuse observations or sensor measurements to obtain a weighted or combined measurement and then use a single Kalman filter to obtain the final state estimate based upon the fused measurement. Whereas state-vector fusion methods use a group of local Kalman filters to obtain individual sensor based state estimates which are then fused to obtain an improved joint state estimate. The simulation results are shown for single, double, triple and quadruple faults detection and diagnosis.