Fault-tolerant nonlinear MPC using particle filtering

Abstract

A fault-tolerant nonlinear model predictive controller (FT-NMPC) is presented in this paper. State estimates, required by the NMPC, are generated with the use of a particle filter. Faults are identiced with the nonlinear generalized likelihood ratio method (NL-GLR), for which a bank of particle filters is used to generate the required fault innovations and covariance matrices. A simulated grinding mill circuit serves as the platform for illustrating the use of this fault detection and isolation (FDI) scheme along with the NMPC. The results indicate that faults can be correctly identiced and compensated for in the NMPC framework to achieve optimal performance in the presence of faults.