Improving Multiobjective Multidisciplinary Optimization With a Data Mining-Based Hybrid Method

Multiobjective, multidisciplinary design optimization (MDO) of complex system is challenging due to the long computational time needed for evaluating new designs’ performances. Heuristic optimization algorithms are widely employed to overcome the local optimums, but the inherent randomness of such algorithms leads to another disadvantage: the need for a large number of design evaluations. To accelerate the product design process, a data mining-based hybrid strategy is developed to improve the search efficiency. Based on the historical information of the optimization search, clustering and classification techniques are employed to detect low quality designs and repetitive designs, and which are then replaced by promising designs. In addition, the metamodel with bias correction is integrated into the proposed strategy to further increase the probability of finding promising design regions within a limited number of design evaluations. Two case studies, one mathematical benchmark problem and one vehicle side impact design problem, are conducted to demonstrate the effectiveness of the proposed method in improving the searching efficiency.