Utilizing artificial neural networks and design solution spaces to cope with the complexity in subframe design

Vehicle subframes are architecturally significant and highly integrated structural components of the chassis. They have a major influence on functional chains at the overall vehicle level and are characterized by a large number of interfaces. The conflicting objectives of increased component complexity versus cross-platform design modularization require new methods and concepts aimed at more efficient development and evaluation of subframe designs across multiple stages of product maturity. This paper presents a method to cope with the complexity and uncertainties in the early development phase of structural components and to carry out a multidisciplinary as well as functional and geometric design at the detail level. First, the complexity is reduced by using automatically generated knowledge-based CAD models and corresponding FEM models, which can cover different degrees of maturity in the design process within the early development phase. Secondly, automated processes are used for all relevant requirement disciplines in order to generate data in a fast and efficient manner to train neural networks, which can map the complex and often non-linear relationships between the design parameters and the characteristic values. Thirdly, epistemic uncertainties are considered with the help of the solution space engineering framework, in which maximized solution intervals for the individual design parameters are determined, to increase robustness and flexibility for the following design process. The effectiveness of the approach is demonstrated using a rear subframe as an example problem.