Fail-bounded implementations of the numerical model predictive control algorithms

Methods of fault-hardening software implementations of the numerical Model Predictive Control (MPC) algorithms are discussed in the paper. In particular, Generalized Predictive Control (GPC) algorithms are considered. The robustness of these algorithms with respect to faults is crucial for process safety and economic efficiency, as faults may result in major control performance degradation or even destabilization. Therefore, fault-hardening of GPC algorithms is an important issue. The fault sensitivity of the non-fault-hardened algorithms implementations and the effectiveness of the fault hardening procedures are verified in experiments with a software implemented fault injector. These experiments refer to the control system of a chemical plant. Experience with fault simulations resulted in some methods of fault-hardening which are described in detail. Improvement of the dependability of the GPC algorithms is commented for each of the proposed fault-hardening mechanism.