Predicting flexural strength in fiber-reinforced UHPC via random forest

Area of Science:

• Materials Science
• Civil Engineering
• Computational Mechanics

Background:

• Ultra-High-Performance Concrete (UHPC) offers superior mechanical properties, particularly flexural strength and ductility, due to fiber reinforcement.
• Predicting the behavior of fiber-reinforced UHPC is challenging due to complex material interactions, especially with supplementary cementitious materials (SCMs) altering microstructure.
• Existing models struggle to capture the nuanced performance of UHPC with diverse SCMs and multiple fiber types.

Purpose of the Study:

• To develop a predictive model for the ultimate flexural strength of fiber-reinforced UHPC.
• To investigate the influence of various SCMs and fiber types on UHPC flexural performance.
• To provide interpretable insights into the key factors governing UHPC flexural behavior.

Main Methods:

• Utilized Random Forest (RF) regression, a machine learning algorithm, for prediction.
• Trained and validated the RF model on an extensive dataset of 550 experimental UHPC mixtures.
• Employed Partial Dependence Plots (PDPs) to visualize and interpret the impact of input variables on flexural strength.

Main Results:

• The RF model successfully predicted the ultimate flexural strength of UHPC with high accuracy.
• Identified fiber type and dosage as critical factors influencing flexural strength.
• Determined the significant impact of matrix parameters including cement content, silica fume, water-to-binder ratio, and aggregate size.

Conclusions:

• Random Forest regression is an effective tool for predicting the flexural strength of complex UHPC formulations.
• Fiber characteristics and matrix composition are primary drivers of UHPC flexural performance.
• The study provides valuable, interpretable data for optimizing UHPC mix designs for structural applications.