This paper was produced for the 2019 NAFEMS World Congress in Quebec Canada
The early design phase of technical products is marked by many inherent uncertainties such as loads, geometric parameters and material properties. For a better classification of the influence of the input parameters on the system response behaviour, numerical experiments can be used to find a more convenient design. This is particularly important in the case of a non-linear relationship between input and output variables.
As the number of input variables increases, the number of possible parameter combinations increases exponentially. A prominent example are laminates in which the layer thicknesses and ply angles can be varied. In practice, there are often requirements, such as symmetry, discrete values for angles and layer thicknesses and a balanced layer structure. Although these limitations reduce the dimension of the design space, there are still too many variation possibilities that require the use of special methods such as Latin Hypercube.
This article aims to show applications in which the use of DOE methods can be useful. These can be tasks in which optimization methods cannot be used because target functions or constraints are not available or derivatives according to design variables do not exist, e.g. non-smooth behaviour in contact problems, eigenvalue and eigenvector derivatives in case of multiple eigenvalues. The collected result data of the sampling procedure can be used to generate response surfaces that allow subsequent optimization. Moreover, the importance of the input variables can be analyzed and judged. At the end of a DOE, a deeper insight in the behaviour of the underlying system is usually available.
Various applications from the fields of acoustics (transmission loss), contact and dynamics (stability of a brake system), buckling of structures and composite materials (laminates) are intended to underpin the meaningful use of DOEs. The structure of the DOE model with regard to the necessary data input is supported by its own wizard within VisPER. All computations are carried out in PERMAS, whereas postprocessing is done in VisPER. Additional evaluations such as response surfaces are conducted using Python scripts. PERMAS specific keywords are denoted by capital letters and a preceding dollar sign in the subsequent text.
|Date||18th June 2019|