Safety-critical multi-core software architecture for model predictive control

Model Predictive Control (MPC) is a process control algorithm based on optimization, in which long, variable optimization time is a major drawback when used in fast dynamic safety-critical systems. In safety-critical systems, timing is vital for the operation of MPC. In this paper, timing, safety and plant performance aspects of a proposed multi-core MPC architecture are investigated in a case study combining simulation and experiment. The proposed architecture performs 'safe' and '(high) performance' optimization in parallel, preferring high performance results, but falling back on safe results. We show that safe optimization, based on Hildreth's algorithm and a minimal set of safety-critical constraints, always meets a deadline, but almost always its result can be subsumed by a high performance result.