Predicting the Erosion Rate of Uni-Directional Glass Fiber Reinforced Polymer Composites Using Machine-Learning Algorithms

Abstract

Abstract The wind turbine and helicopter rotor blades when exposed to dust borne environment are subjected to leading edge erosion because of the impact of dust particles. These blades are manufactured from fiber reinforced polymer (FRP) composites and therefore, it is important to predict the erosion rate of FRP composites. In this paper, the main aim is to accurately predict the erosion rate of uni-directional FRP composites using machine-learning algorithms like artificial neural networks (ANNs) and extreme gradient boosting (XGB) and compare between the algorithms. The model uses input parameters like erodent impact angle, velocity of erodent particle, fiber orientation, and fiber volume fraction as the input and erosion rate as the output variable. The total dataset considered for training and testing the model is obtained from two parts. The first part of the dataset is obtained from the literature and the other part is collected from performing in-house experiments on uni-directional glass fiber reinforced polymer (GFRP) composites. The crater profiles of the tested specimens are characterized using 3D Alicona imaging microscopy. The machine-learning models considered in this study provide accurate results on the dataset. However, the XGB method is more robust, reliable, and faster to train and more accurate than the ANN model in the case of an unknown dataset (dataset not used for training). The feature importance from the XGB model suggests that impact particle velocity, impact angle, and fiber orientation are the most important input features. The model predictions by taking into account the three input features provide accurate results without affecting the accuracy of the model.