In decomposition-based design optimization strategies such as analytical target cascading (ATC), it is sometimes necessary to use reduced representations of highly discretized functional data exchanged among subproblems to enable efficient design optimization. However, the variables used by such reduced representation methods are often abstract, making it difficult to constrain them directly beyond simple bounds. This problem is usually addressed by implementing a penalty value-based heuristic that indirectly constrains the reduced representation variables. Although this approach is effective, it leads to many ATC iterations, which in turn yields an ill-conditioned optimization problem and an extensive runtime. To address these issues, this paper introduces a direct constraint management technique that augments the penalty value-based heuristic with constraints generated by support vector domain description (SVDD). A comparative ATC study between the existing and proposed constraint management methods involving electric vehicle design indicates that the SVDD augmentation is the most appropriate within decomposition-based design optimization.
Improving Support Vector Domain Description by Maximizing the Distance Between Negative Examples and The Minimal Sphere Center's
