Deep learning based high strength concrete prediction model

Concrete is widely utilized building material. With the integration of machine learning, particularly deep learning techniques, its performance assessment and mix optimization have become more data-driven and precise. By analyzing vast amounts of data, such as the properties of different materials, curing conditions, and strength test results, deep learning algorithms can identify patterns and make predictions. These models help eliminate guesswork by reducing dependency on trial-and-error methods, thereby lowering both time and cost. In this context, the current study explores the deep learning's application, specifically the Capsule Network (CapsNet), to predict the strength and concrete specimen's behavior such as cubes, cylinders, and beams. The main objective is to estimate the compressive strength of cubes, the tensile strength of cylinders, and the flexural strength of beams produced with various dosages of metakaolin, superplasticizers, and other conventional materials. For this purpose, a total of 75 cubes (150x150x150 mm), 15 beams (100x100x500 mm), and 15 cylinders (150x300 mm) were cast and tested under controlled laboratory conditions. The results demonstrated that the CapsNet model effectively captured the variations in mechanical performance, particularly enhancing the prediction of ultimate load-carrying capacity. This validates the machine learning-based approach's potential in improving concrete performance prediction and supporting intelligent material design.