The competitive nature of industry demands continuous innovation to enable reductions in design cycle time while satisfying ever increasing design functionality demands. Optimisation technology provides a scientific approach to automatically determine more efficient designs within competitive time constraints.
Over the last decade, optimisation technology has evolved to be an established discipline within many industries. This has been facilitated by software developments, both to algorithm performance and tighter integration within the design environment and hardware developments, which have delivered year on year increased computational speed.
Such developments have removed the barriers to performing optimisation which can be seen as an extension to a standard finite element study. Any optimisation problem, irrespective of the optimisation technology used, requires only three design functions to be defined:
Design objective (e.g. minimise mass) which is achieved by the variation of design variables (e.g. structural shape) subjected to design constraints (e.g. displacements, stresses).
Topology Optimisation differs from other technologies since it is applied at the commencement of the design process even before the development of CAD surfaces. In addition, the design variables are automatically defined by the software (i.e. void dimensions in every element in the design space). Traditionally the intuition and experience of the designer has played the central role in defining the initial geometric layout of the design. The benefit of this revision to the process is that the design cycle is significantly reduced, since the iteration process of structural analysis providing feedback for design changes is removed.
A conventional entry point for optimisation technology commences on the completion of a baseline finite element analysis and is referred to as size and shape optimisation. The size refers to changes in thickness while shape refers to changes to the external boundaries of the structure. This technology is extremely flexible since any quantity computed by the analysis code can be selected as a design objective or a design constraint. While any value specified in a finite element input deck can be specified as a design variable.
An optimisation problem can be completely specified with pre-processing tools such as Hypermesh. In addition Hypermesh contains facilities to specify shape variables, which allow optimum structural configurations to be determined.
The application of topology optimisation benefits from early input of a multi disciplinary team consisting of the analyst, designer and manufacturing engineer.
Gradient-based optimisation techniques have been used in the industry for well over two decades. They are frequently applied to full body car models to automatically down-gauge the thicknesses of various panels. This technique fits well into the framework of a conventional finite element program since the stiffness matrix is used to construct tangents, which provides sensitivities to locate the optimum solution. An example product with gradient-based optimisation is Optistruct, which is routinely applied to cast or moulded components to minimise mass while retaining stiffness.
Determining an optimum solution using a response surface technique consists of two stages. The first stage is to select a space filling design or test plan (e.g. Design of Experiments) to evaluate the design objective and constraints in the design space. These test plans consist of pre-defined locations in the design space. Once the response surface has been evaluated at these locations an analytical fit is obtained to obtain a continuous evaluation of the response surface.
An important deliverable of the response surface optimisation process is the data produced which aids the understanding of the system by assessing the influence of design variables, predicting trends and comprehending how various design variables interact. Altair’s product StudyWizard contains extensive response surface optimisation functions.
A genetic algorithm is a machine learning technique modelled upon the natural process of evolution. The genetic algorithm (GA) attempts to find the best solution to a problem by genetically breeding a population of individuals over a number of generations through the use of operators like reproduction, crossover, and mutation. Every individual is assigned a fitness value, which is a measure of how well each solution solves the problems objective.
Following Darwin’s principle of survival of the fittest, individuals with higher fitness values have a higher probability of being selected for mating purposes to produce the next generation of candidate solutions.
A disadvantage is that typically 100-1000 evaluations of the design objective and constraint are required. The technique is formulated to consider the design variables as discrete, as opposed to other technologies which consider the design variables as continuous. This has advantages since component parts are often only available in discrete sizes.
Optimisation technology is applied in the design process of a number of industries. The technology has refined the conventional design process resulting in a CAE focussed procedure. This reduces design cycle time, by reducing the number of design iterations. The technology can produce innovative non-intuitive designs, which can often be the only way to efficiently achieve rigorous design functionality targets.Optimisation technology provides a significant competitive advantage.
A single baseline assessment is no longer adequate, increasingly it is required to demonstrate that the design is optimised. And the increase in computer power has removed the barrier to performing multiple analyses to achieve this objective.