Optimizing Propeller Efficiency Using AI / Machine Learning Techniques

Abstract

Abstract In recent years, the aerospace industry has placed increasing importance on enhancing propeller efficiency. However, traditional methods like computational fluid dynamics (CFD) simulations and empirical analysis demand substantial expertise and resources. Moreover, the intricate nature of propeller aerodynamics complicates accurate modeling requiring few other techniques that can develop predictive models based on the available data on existing propellers. This paper addresses this challenge by utilizing experimental wind tunnel data from a UIUC repository to predict propeller efficiency using AI/ML models. Key design parameters, including number of blades, diameter, pitch, brand, alongside advanced ratio and RPM rotation inputs, are used. IMB Watson could-based AI platform was used to develop predictive models. Various algorithms including linear regression, decision tree and snap random forest were tested with and without enhancements (HPO-1, FE, HPO-2). Utilizing a diverse dataset, we achieved a notable 0.036 root mean squared error (RMSE), with significant features identified including thrust and power coefficients. A 10:90 test:train split ratio contributes to robust model performance. While the application of advanced AI and machine learning techniques in aerospace engineering is still in its embryonic stage, this study lays a foundation for future advancements in propeller design and efficiency optimization through rigorous analysis of various predictive models that were employed in this work.