A distributed multi-disciplinary design optimization benchmark test suite with constraints and multiple conflicting objectives

Collaborative optimization (CO) is an architecture within the multi-disciplinary design optimization (MDO) paradigm that partitions a constrained optimization problem into system and subsystem problems, with couplings between them. Multi-objective CO has multiple objectives at the system level and inequality constraints at the subsystem level. Whilst CO is an established technique, there are currently no scalable, constrained benchmark problems for multi-objective CO. In this study, we extend recent methods for generating scalable MDO benchmarks to propose a new benchmark test suite for multi-objective CO that is scalable in disciplines and variables, called `CO-ZDT'. We show that overly-constraining the number of generations in each iteration of the system-level optimizer leads to poor consistency constraint satisfaction. Increasing the number of subsystems in each of the problems leads to increasing system-level constraint violation. In problems with two subsystems, we find that convergence to the global Pareto front is very sensitive to the complexity of the landscape of the original non-decomposed problem. As the number of subsystems increases, convergence issues are encountered even for the simpler problem landscapes.