Long-term performance prediction framework based on XGBoost decision tree for pultruded FRP composites exposed to water, humidity and alkaline solution

Fiber reinforced polymer (FRP) composites are susceptible to material degradation when exposed to environmental effects. To predict the residual tensile strength and modulus of pultruded FRP composites, an XGBoost decision tree model was developed in this work. XGBoost decision tree, as a machine learning technique, is able to provide accurate predictions for tabular dataset with a good prediction interpretability. In this work, the methodology of XGBoost decision tree was presented in detail. Datasets for training and testing included a total of 746 data points which were collected from an existing database. XGBoost decision tree model predictions were cross-validated with 149 test data, and an excellent agreement was observed, showing R2 values of 0.93 and 0.85 for tensile strength and modulus, respectively. In addition, attribute importance analysis was conducted to quantitatively evaluate the attributes pertaining to FRP degradations, including exposure time, exposure temperature, pH value of environment, fiber volume fraction, plate thickness, fiber type and matrix type. Exposure time and temperature were observed to have the greatest impacts on residual tensile properties. The proposed XGBoost decision tree model provides a new approach for predicting the long-term degradations of FRP composites subjected to environmental effects.