On improving dependability of the numerical GPC algorithm

The paper studies the dependability of software implementation of the numerical Generalized Predictive Control (GPC) Model Predictive Control (MPC) algorithm. The algorithm is implemented for a control system of a multivariable chemical reactor - a process with strong cross-couplings. Fault sensitivity of the proposed implementations is verified in experiments with a software implemented fault injector. Experimental methodology of disturbing the software implementation of the control algorithm is presented. The impact of faults on the quality of the controller performance is analysed and techniques improving the dependability of the algorithm implementation are proposed. The experimental results prove the efficiency of the software improvements.