Authors: Micah Stark (Texas A&M University) , Cesario Tavares (Texas A&M University) , Pratik Gujar (Texas State University) , Xijun Shi (Texas State University) , Kinsey Skillen (Texas A&M University)
Fracture patterns of concrete test cylinders under uniaxial compression have been associated with many factors. Literature suggests that saturation level, mix constituents, fiber content, curing temperature and specimen end conditions play a crucial role in the failure mode. As such, tools capable of modeling the fracture pattern as a function of mix composition and curing conditions are attractive considering they have the potential to help explain microstructural behavior. In particular, understanding the relationship between the cementitious matrix and fiber reinforcement in dictating crack propagation could help develop more ductile ultra-high-performance concrete (UHPC) compositions. In this study, orthogonal arrays are used to reduce experimental runs required to model fracture patterns (following ASTM C39/C39M) when testing UHPC cylinders for compressive strength at 24 hours. Processed data indicates a strong correlation between fracture type, mix constituents and compressive strength. Therefore, a k-nearest neighbor algorithm is used to model this data with slag, microsilica and fly ash replacement levels and compressive strengths used as features. The model was trained on 85% of the experimental data while validation was performed in the remaining 15%. A 79% model accuracy was estimated using a k-fold cross validation process, suggesting that this method is suitable to model small categorical datasets. The predictions obtained are illustrated through categorical and probabilistic performance density diagrams. This AI-based tool helps evaluate the synergistic relationship between raw constituents and fracture pattern. This method could be used in future work to improve the ductility and cost-efficiency of UHPC through optimizing the fiber content.
Keywords: machine learning, mix design optimization, fracture pattern
How to Cite: Stark, M. , Tavares, C. , Gujar, P. , Shi, X. & Skillen, K. (2023) “Predicting Early-Age Fracture Pattern of UHPC Cylinders Using Categorical Performance Density Diagrams”, International Interactive Symposium on Ultra-High Performance Concrete. 3(1). doi: https://doi.org/10.21838/uhpc.16709