Towards the theoretical limit of optimal requirement decomposition using solution spaces for complex systems design
Mechanics
Final Report Abstract
The increasing complexity of modern technical systems poses significant challenges for engineers. In situations where many components interact and uncertainty plays a critical role, for example, due to manufacturing tolerances or incomplete information, traditional pointbased optimization approaches often fail to provide robust solutions. These solutions may be impractical or unable to meet overall system goals reliably. Solution spaces can be used to deal with these issues in the early phase of the product development process. Instead of looking for one single optimal point design, solution spaces keep whole sets of designs that meet all system requirements. The traditional approach of computing solution spaces involves permissible intervals for each design variable, which can be used independently of each other, producing so-called box-shaped solution spaces. Having said intervals enable independent work on each variable, with flexibility and robustness as large as the intervals themselves. This has been used in several applications in academia and in industry, including vehicle crash design, product family development for electric vehicles, and improvements to vibratory rammers, among others. In this project, the concept of component solution spaces was substantially advanced. With component solution spaces, for the first time, admissible regions for arbitrarily many design variables of a component can be computed in a coupled manner, even for highly non-linear systems, while still allowing for independent choice of values for those design variables that belong to different components. Component solution spaces are computed with a new algorithm, which was adapted from the previously defined one for box-shaped solution spaces. The most important change lies in how the component regions are trimmed to remove bad designs, where two new methods were introduced: planar trimming and corner box removal. Planar trimming produces convex regions and shows great numerical scaling; corner box removal can produce non-convex regions, at the cost of potentially requiring many more computational resources. Throughout the project, new numerical algorithms were developed and successfully tested on structural engineering problems, such as truss designs, and on robot design problems. Through these different applications, we have shown how component solution spaces can generate spaces which are up to eleven orders of magnitude larger than those you could obtain with the previous box-shaped solution spaces. The outcomes are relevant for any domain dealing with complex systems, including automotive, robotics, and aerospace engineering. The methods developed can be used to enable concurrent engineering with maximum design freedom, speeding up product development, enabling better choice of materials, and increasing reliability. This can benefit high-tech sectors as well as everyday products – for example, by enabling electric vehicles that are lighter, sturdier and more affordable.
Publications
-
OPTIMIZING REQUIREMENTS FOR MAXIMUM DESIGN FREEDOM CONSIDERING PHYSICAL FEASIBILITY. Proceedings of the Design Society, 3, 2865-2874.
Rodrigues Della Noce, Eduardo & Zimmermann, Markus
-
Stochastic optimization of convex component solution spaces for arbitrary performance functions. Structural and Multidisciplinary Optimization, 68(8).
Rodrigues Della Noce, Eduardo & Zimmermann, Markus
