A Machine Learning Based Framework for Model Approximation Followed by Design Optimization for Expensive Numerical Simulation-based Optimization Problems
Numerical simulation is becoming crucial factor in all sectors to survive in the current competitive economy as the process reduce cost and product delivery time. With the advances in computer-based tools, however, the modeling and simulation tools are becoming expensive particul arly in searching for optimum solutions for complex design problems. For this and similar reasons, utilizing metamodels to replace the expensive numerical models is in turn becoming vital. In this article, a complex implicit-numerical simulation-based optimization problem is approximated using a machine learning tool. An optimization framework based on Artificial Neural Network tool has been employed to train and replace an implicit computational fluid dynamic analysis followed by design optimization. Genetic algorithm is used to conduct design optimization directly on the approximated model. The case study on micro cross-flow turbine conducted using the approach indicates that the optimized design model from the approximated model approach showed better output than the original.
展开▼
