Multidisciplinary Optimization

The increasing necessity to better model complex real world problems is drawing attention on the development of new methods and approaches that can cope with the strong interactions among different disciplines that are inevitably present in engineering problems. Therefore the multiobjective optimization, raised from the multidisciplinary aspects, is gaining more and more interest both in the scientific and in the industrial community.

In multidisciplinary optimization, the strong link among different disciplines poses several problems in terms of finding the global optimum, dealing with the constraints and having an efficient architecture regarding the model management and the convergence speed.

Our research aims at efficiently solve these problems both in terms of convergence speed and quality of the obtained solutions. Focus is given on the preliminary design of complex systems trough a concurrent approach that allows getting solutions not reached trough traditional, non simultaneous, study on the disciplines involved.

Those kind of problems can be hardly solved by a single optimizer because of the computational demand and the wildness of the search and cost function spaces. More specifically, two main issues must be addressed whenever dealing with multiple disciplines in optimization:

• the optimization architecture settlement – it can be also dynamically defined depending on the optimization process evolution;
• the meta-modelling – lighter but effective models for each problem area should be defined to keep the process fast and well-balanced among different complexity loads.

A promising solution to the first issue is the process distribution on different optimization modules devoted either to a subset of cost function vector optimization or to a sub area of the search space visiting; however, distribution asks for  the definition  of the coordination rules among the units.

Our group applied different protocols coming from the Game Theory area to settle problem dependent architectures.  Complex space system design and space vehicle optimization during atmospheric manoeuvring (launchers, Entry-Descent-Landing vehicles, aerocapture and aerogravity assisted capsules) have been already solved successfully.

On going research activities focus on:

• defining a general software architecture to dynamically distribute the problem on different parallel processes, even on different physical machines;
• identifying rules to guide the user in efficiently distributing the problem;
• building global optimization tools to deal with highly non-linear and constrained problems in multidisciplinary environments;
• developing  approaches to describe the considered problems, both trough preliminary models and trough existing tools connected in a distributed architecture;
• critically evaluating different meta-modelling techniques according to classes of models to be simplified.