A machine learning-based comparative analysis of surrogate models for design optimisation in computational fluid dynamics

• DOI: 10.1016/j.heliyon.2023.e18674
• Publication Date: 2023-07-26T03:27:36Z
• Authors: Azfarizal Mukhtar; Ahmad Shah Hizam ...; Mohamad Fariz Moh...
• Keywords: Surrogate model; Design of experiments

Complex computer codes are frequently used in engineering to generate outputs based on inputs, which can make it difficult for designers understand the relationship between inputs and determine best input values. One solution this issue is use design of experiments (DOE) combination with surrogate models. However, there a lack guidance how select appropriate model given data set. This study compares two modelling techniques, polynomial regression (PR) kriging-based models, analyses critical issues optimisation, such as DOE selection, sensitivity, adequacy. The concludes that PR more efficient generation, while models better assessing max-min search results due their ability predict broader range objective number location points affect performance model, error lower than PR. Furthermore, sensitivity information important improving efficiency, suited determining variable greatest impact response. findings will be valuable simulation practitioners researchers by providing insight into selection All all, demonstrates techniques solve complex problems effectively.