Estimating the initial fracture energy of concrete using various machine learning techniques

Abstract

The assessment of the energy required for crack propagation in concrete structures has been fascinating since fracture mechanics was applied to concrete. In the case of concrete, considered a quasi-brittle material, the fracture energy has proven to be a crucial factor in the reliable design of structures and modeling failure behavior. However, due to the complex, time-consuming, and expensive laboratory tests, there has been ongoing and intense debate regarding the methods to estimate the fracture energy of concrete. The advent of machine learning (ML) methods in this domain can hold great promise for resolving such issues once and for all. This study used a comprehensive analysis of twelve ML algorithms for estimating the initial fracture energy of concrete (IFEC), utilizing a more extensive and diverse database (500 data points) than previous studies. The performance of the ML models was evaluated using several metrics, such as coefficient of determination (R2) and variance accounted for (VAF). The findings revealed that all the ML models employed in this study demonstrate remarkable accuracy in estimating the IFEC value, with R2 and VAF values of more than 0.86 and 93.10 %, respectively. A ranking of the models based on their estimation accuracy was provided, facilitating the selection of the support vector regression (R2 = 0.9897; VAF = 99.50 %) and long-short-term memory (R2 = 0.9804; VAF = 99.00 %) methods as the most reliable models for IFEC estimation. Both the laboratory test and ML models presented the highest IFEC value for a water-to-cement ratio of 0.35. Additionally, by increasing the values of each of the parameter's maximum size of aggregates (from 7 mm to 35 mm) and the specimen’s age (from 3 days to 180 days), the IFEC value was increased by about 100 %. Notably, a user-friendly software based on the ML models was developed, enabling fast and highly accurate estimation of IFEC, thereby eliminating the need for time-consuming and expensive laboratory tests.