Deb, A., Srinivas, G., and Chou, C., "Development Of A Practical Multi-disciplinary Design Optimization (MDO) Algorithm For Vehicle Body Design," SAE Technical Paper 2016-01-1537, 2016, doi:10.4271/2016-01-1537.
The present work is concerned with the objective of developing a process for practical multi-disciplinary design optimization (MDO). The main goal adopted here is to minimize the weight of a vehicle body structure meeting NVH (Noise, Vibration and Harshness), durability, and crash safety targets. Initially, for simplicity a square tube is taken for the study. The design variables considered in the study are width, thickness and yield strength of the tube. Using the Response Surface Method (RSM) and the Design Of Experiments (DOE) technique, second order polynomial response surfaces are generated for prediction of the structural performance parameters such as lowest modal frequency, fatigue life, and peak deceleration value. The optimum solution is then obtained by using traditional gradient-based search algorithm functionality “fmincon” in commercial Matlab package. The stated goal of optimization can also be achieved using a practical MDO methodology in which a substantially reduced set of cases need to be considered leading to a computationally efficient solution. The results of both the RSM-based and practical MDO methods are compared and it has been found that the current practical MDO methodology is substantially more efficient when compared to RSM-based optimization and predicts nearly the same solution as the latter. The practical MDO process is finally implemented on a real-world problem taking a full vehicle design problem as an example and the efficacy of the practical approach is demonstrated. It has been found that, using both the methods, a weight reduction of 16% with respect to the baseline design can be obtained and values of design variables yielded by the practical MDO methodology are nearly same as that of the RSM-based weight optimization technique. However, the practical MDO approach does not rely on response surfaces and has significantly higher throughput with a reduction in computation time of 76% as compared to the RSM-based method which requires further analysis for convergence.