Simulation-Based Design of Industry-Size Control Systems With Formal Quality Guarantees

Realistic industrial systems typically need to be modeled as hybrid systems consisting of hundreds (easily thousands) of nonlinear differential algebraic equations (DAEs). The size of such models is one of the major obstacles to overcome when developing automated design methods for industrial control systems. In this article, we present a scenario-based approach that, by exploiting the synergies among simulation, black-box optimization, and statistical model checking, allows us to automate the design of quality-guaranteed industry-size control systems, i.e., control systems for which a user-specified statistical guarantee on correctness holds over the possible operational scenarios. We show the effectiveness of our approach through a Modelica model consisting of a hybrid nonlinear DAE system with 1276 equations, 492 of which are nontrivial, containing 152 continuous state variables and 38 discrete ones, plus 7 algorithm blocks. Our experiments show that within a few hours of computation on an off-the-shelf workstation, we can find quality-guaranteed solutions (with very tight quality guarantees) to our design problem. We also compute an entire discretized Pareto front for such a large system over two conflicting key performance indicators.