Optimization of Dimensioning Problems in PlantLib Models Using FMU-Based Workflows

Abstract

Optimation AB performs optimization of design parameters in industrial plants primarily through experience-based methods and manual trial-and-error testing in simulation models. The development of metaheuristic optimization algorithms has introduced new approaches for optimizing complex, nonlinear systems without requiring knowledge of the internal structure of the problem. To assess the potential of these algorithms for industrial dimensioning problems, a pulp plant model developed in Dymola was exported as a Functional Mock-up Unit (FMU) and simulated externally using Python. An optimization problem was formulated in which storage tank volumes were defined as decision variables. The objective function was based on net present value (NPV), calculated by combining capital expenditure (CAPEX) and operational expenditure (OPEX) under fixed price assumptions. The optimization framework was implemented in Python and applied to two baseline methods—Coordinate descent and Latin hypercube sampling—as well as two metaheuristic algorithms: Differential evolution and Simulated annealing. Before being applied to the real plant, the algorithms were tested and tuned on a simpler model designed to be nonsmooth and noisy. While all algorithms were able to identify tank sizes that improved economic performance compared to the baseline design, the metaheuristic methods demonstrated limited practical potential for the tested problem. This was primarily due to high computational cost and sensitivity to noise in the simulation results.