Machine learning-assisted prediction of durability behavior in pultruded fiber-reinforced polymeric (PFRP) composites

Pultruded fiber-reinforced polymer composites are emerging as construction material in civil engineering infrastructure. These composites are susceptible to degradation when exposed to long-term environmental conditions. This paper employs various machine learning models to predict the mechanical properties under various environmental exposures. A large dataset of experimental results of pultruded composites exposed to moisture, alkaline or acidic solutions, temperature, UV radiation, freeze-thaw and wet-dry cycles, and in-situ environments has been collected from literature and used to train and test the machine learning models. The aim is to provide the best machine learning models that can predict the mechanical properties with no or minimal testing thus reducing time and cost. Moreover, these models are intended to help implement design according to specifications and standards. The results reveal that Decision Trees, Artificial Neural Networks, and Random Forests are the best models to predict behavior of pultruded composites under environmental exposure, which achieved R2 values of 0.9647, 0.9537, and 0.8970, respectively for the case of flexural strength. Moreover, the results emphasize that the mechanical properties are highly dependent on the fiber and matrix types, the aging effect, the fiber volume fraction, the temperature and the exposure period. The models can be exploited to predict long-term effects of environmental conditions on mechanical properties of pultruded composites without extensive experimental testing. The findings presented in the current study can assist codes and standards in the design of structural systems utilizing pultruded composites without relying on the conservative strength reduction factors but rather in machine learning and modelling of experimental data.