Prediction of Volumetric Drag Coefficient of TriLobed Airship Envelopes using ML tools

Hybrid Airships use tri-lobed envelopes, which enable the installation of flatter solar panels on their upper surface to increase their efficiency. The envelope shape can be parameterized in terms of some geometrical parameters. Optimization of the envelope shape to reduce its volumetric drag coefficient (CDV) requires wind-tunnel tests or numerical investigations, both of which are expensive and time consuming. In this paper we describe an approximate model to predict the envelope CDV as a function of nine geometrical parameters. This model is built by applying six Machine Learning models, viz., Linear Regression (LR), K-Neighbors Regression (KNNR), Decision Trees (DT), Gradient Boosting Regressor (GBR), Elastic Net with Polynomial Terms (ENPT) and Neural Network (NN) on a dataset comprising of numerically computed values of CDV for 170 combinations of the shape parameters. Sensitivity of CDV to each of the envelope shape parameters was determined using Accumulated Local Effects (ALE) plots. It was found that ENPT performed the best in this dataset and resulted in predicting the CDV within an error band of ± 3%. Feature Importance Analysis was carried out to reveal that envelope fineness ratio had the largest contribution towards the prediction of CDV, followed by the inter lobe distance and envelope prismatic coefficient. Two additional parameters, viz., Total Surface Area and Cross-Sectional Area of the envelope were introduced, and the former was found to be significantly influencing CDV.