2021/01/07 by Opeoluwa Owoyele, Pinaki Pal, Owoyele, Opeoluwa +11
Computer Science · Engineering · #Advanced Multi-Objective Optimization Algorithms #FOS: Computer and information sciences #Machine Learning (cs.LG) #Refrigeration and Air Conditioning Technologies #Turbomachinery Performance and Optimization
paper · pdf · doi:10.48550/arxiv.2101.02653
openalex publication_date 2021/01/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In recent years, the use of machine learning-based surrogate models for\ncomputational fluid dynamics (CFD) simulations has emerged as a promising\ntechnique for reducing the computational cost associated with engine design\noptimization. However, such methods still suffer from drawbacks. One main\ndisadvantage of is that the default machine learning (ML) hyperparameters are\noften severely suboptimal for a given problem. This has often been addressed by\nmanually trying out different hyperparameter settings, but this solution is\nineffective in a high-dimensional hyperparameter space. Besides this problem,\nthe amount of data needed for training is also not known a priori. In response\nto these issues that need to be addressed, the present work describes and\nvalidates an automated active learning approach, AutoML-GA, for surrogate-based\noptimization of internal combustion engines. In this approach, a Bayesian\noptimization technique is used to find the best machine learning\nhyperparameters based on an initial dataset obtained from a small number of CFD\nsimulations. Subsequently, a genetic algorithm is employed to locate the design\noptimum on the ML surrogate surface. In the vicinity of the design optimum, the\nsolution is refined by repeatedly running CFD simulations at the projected\noptimum and adding the newly obtained data to the training dataset. It is\ndemonstrated that AutoML-GA leads to a better optimum with a lower number of\nCFD simulations, compared to the use of default hyperparameters. The proposed\nframework offers the advantage of being a more hands-off approach that can be\nreadily utilized by researchers and engineers in industry who do not have\nextensive machine learning expertise.\n