Evaluating Diverse Meta-Modeling Approaches for Predicting Performance Characteristics of a Twin Air Intake Based on Experimental Data

Abstract Air intakes are critical components in maximizing the efficiency of jet-powered engines. Their diverse designs, ranging from conventional shapes to innovative configurations, coupled with the intricate interplay of fluid dynamics, boundary layer effects, and structural considerations, render the determination of their performance characteristics a time-consuming task. However, a meticulous and confident evaluation of these characteristics is the key to achieving optimal air intake design and, consequently, significant enhancement of overall engine performance. This article assesses various meta-modeling approaches for predicting the performance characteristics of a twin air intake system. A comprehensive exploration of meta-modeling methods, particularly those specifically tailored for data derived from experiments, is presented. A database of 4000 experimentally obtained runs is utilized to construct train and test data for diverse models, including polynomials, decision trees, random forest regression, multivariate adaptive regression splines, and neural networks. The performance of each model is rigorously evaluated based on goodness of fit, precision, accuracy, monotonicity, and interpretability. This study provides a cost-effective and time-efficient alternative for predicting crucial flow parameters associated with the air intake of jet engines. The results reveal that the Random Forest Regression (RFR) model outperforms all other models across all evaluated metrics, demonstrating its superior effectiveness in predicting the performance characteristics of the twin air intake system.