Strength Prediction of Concrete with Cold Joints Using Artificial Neural Network and Sensitivity Analysis

Large amounts of concrete pouring are heavily seen in growing countries like the Philippines. Big projects with better planning and scheduling still face construction delays, and the construction sector, from private to public institutions, experiences it. Concrete cold joints are one of the unavoidable results of construction delays when the initial layer of concrete already hardens before placing the second layer. Different studies regarding the mechanical properties of cold joints and their effect on concrete structural aspects are experimentally proven. This study compared compressive and flexural tests of six ordinary cylindrical and beam samples to 24 models with cold joints; the strength test results confirm the literature data. The study wants to predict the strength of concrete due to cold joint formation, precisely, concrete’s compressive and flexural strength about concrete age, casting delay, joint interface angle, and water-cement ratio as variables. An Artificial Neural Network (ANN)-based model involving input variables was initiated in Levenberg-Marquardt (LM) and hyperbolic tangent sigmoid as training algorithm (TA) and transfer function (TF) respectively. The best model generated was found to have a topology of 4 – 9 – 1 (input – hidden – output). Sensitivity analysis (SA) was employed to assess the relative importance of variables about compressive and flexural strength. This involved utilizing the connection weight values obtained from the trained neural network and subsequently checking them through Garson’s Algorithm (GA). In addition, Finite Element Analysis (FEA) was used in this study to simulate and verify the stress developed in the laboratory.