Evaluation of fracture toughness properties of polymer concrete composite using deep learning approach

Abstract

Using artificial intelligence-based methods in predicting material properties, in addition to high accuracy, saves time and money. This paper models and predicts the fracture toughness properties of polymer concrete (PC) composites using the deep learning method. After preparing a database consisting of 209 experimental data from 19 relevant studies, the effect of seven important variables on critical stress intensity factor (K Ic ) and crack tip opening displacement (CTOD) is considered. Then, the deep neural network (DNN) model is developed and trained using the prepared database. The accuracy of the DNN model is examined by implementing four statistical criteria, MSE, R 2 , RMSE, and MAE. Finally, the sensitivity of the K Ic and CTOD to each input variable is evaluated using a partial dependence plot (PDP) analysis. While aggregate size, nanofiller content, and a/R ratio have the most positive effect on K Ic , aggregates and nanofiller content have the most positive influence on CTOD.

Physical cellular automata and artificial fish swarm fusion catastrophic failure prediction of brittle-like materials

“…With regrading to the first kind of application strategies, material properties including compressive/tensile strength, fracture toughness indexes, such as stress intensity factor, crack tip opening distance (CTOD), as well as elastic modulus, are predicted by utilizing various artificial intelligence approaches [38][39][40][41], such as artificial neural network (ANN), deep neural network (DNN), particle swarm optimization (PSO) and ant colony optimization (ACO). With regrading to the second kind of application strategies, some artificial intelligence approaches are utilized to predict the local damage and crack path distribution based on the global indexes such as the structural nature frequency, displacement, and stress intensity factor obtained from monitoring or theoretical solution [42][43][44][45].…”

Using deep learning method to predict dimensionless values of stress intensity factors and T‐stress of edge notch disk bend (ENDB) specimen

“…Ly et al 35 constructed a dataset, including 233 experimental data points from past studies, to predict the strength of the cementitious concrete. Similarly, Niaki et al 36,37 constructed the experimental datasets, based on the weight percentage (wt%) and properties of the polymer concrete (PC) [38][39][40][41] from the results of the relevant studies. In these works, 70% of the datasets were randomly allocated for training, and the rest of the datasets were equally allocated for generalization and validation.…”

Predicting geometry factors and normalized T‐stress of centrally cracked Brazilian disk specimens using deep learning method

“…Ly et al 32 prepared a database composed of 233 data points from previous experimental works to predict the strength of the concrete. Similarly, Hassani Niaki et al [33][34][35] prepared the databases, including the weight content (wt%) of the materials from the experimental results of the previous studies to train the DNN. They used 70% of the database for training, 15% for generalization, and 15% for validation.…”