Improved Multidisciplinary Optimization and its Practical Usage at Porsche AG

In the automotive industry, the simulation-based design and optimization of vehicle structures needs to combine requirements from a number of different engineering disciplines. Multidisciplinary optimization (MDO) is a term that is used to summarize approaches that aim at integrating all relevant disciplines and finding optimal compromises between them. Due to its complexity, however, the practical use of MDO in the vehicle design process is often limited.

A recent publication on “Making multidisciplinary optimization fit for practical usage in car body development” (Authors: Jana Büttner, Axel Schumacher, Thomas Bäck, Stefan Schwarz, Peter Krause; Structural and Multidisciplinary Optimization 66:62, 2023) addresses this complexity both concerning technical improvements of the underlying methods and concerning the resulting organizational advantages for the product development process. To illustrate the approach, a FE full vehicle example with six load cases (crash and frequency analysis) is optimized using the proposed approach.

The key technical improvements of the MDO process introduced in the paper can be briefly summarized as follows:

1. A global sensitivity matrix approach is introduced, which allows for structuring the design variables and disciplines in a way that effectively supports the coordination between disciplines in the design process.
2. All metamodels trained on simulation results are enhanced by a local uncertainty measure for prediction accuracy. This uncertainty measure is then used to support the use of the metamodels in the optimization process.
3. An adaptive complexity control mechanism is introduced to progressively reduce the complexity of the optimization problem.

In the paper, the state-of-the-art in MDO is presented first, explaining the requirements and challenges for MDO that motivate the proposed extensions. In order to reduce numerical resource requirements and improve the quality of optimization results, the global sensitivity matrix, efficient optimization algorithms using the local uncertainty measure, approaches for reducing the FE computation effort, and the adaptive complexity control are explained in detail.

The full vehicle design example reported in the article shows significant benefits from the proposed approach. For example, one of the new MDO variants reduces the car body weight by 17.75kg, finding a practically feasible solution with a reduction of the computational effort (in terms of CPU hours) by 64.55%, in comparison to the standard MDO process.

For the metamodels, the generalized uncertainty measure, and the global optimization algorithms, divis’ ClearVu Analytics software tool was used, and divis’ ClearVu Solution Spaces software for identifying the largest feasible subspace of the design space.