COMPRESSIVE STRENGTH MODEL OF CONCRETE WRAPPED WITH CARBON FIBER-REINFORCED POLYMER USING NEURAL NETWORK

Abstract

This study developed a predictive model for compressive strength utilizing a feedforward neural network (FNN) trained on the dataset from cylindrical specimens reinforced with carbon fiber-reinforced polymer (CFRP) wraps. Bridging a critical gap in the literature, the FNN model incorporates simultaneous effects of variables like concrete cylinder diameter, wrap thickness, CFRP laminate modulus of elasticity, and unconfined compressive strength. The optimized 4–10–1 FNN architecture, trained on 101 samples, exhibits exceptional precision and generalization, as evidenced by a high Correlation Coefficient of 0.99795, and low Mean Squared Error of 0.00024. This highlights its effectiveness in predicting compressive strength and identifying influential factors. A comparative analysis underscores the FNN model's superiority over existing approaches. The study’s novelty lies in its focus on unconventional specimens, for which no established codes or standards currently exist. It introduces a new way for developing future guidelines for CFRP-wrapped structural elements in engineering applications, contributing valuable insights into enhancing predictive modeling for advanced composite materials.