Machine learning predictions for optimal cement content in sustainable concrete constructions

Abstract

The designated concrete compressive strength at 28 days plays an important role in determining the quantity of cement (water to cement ratio) needed in concrete mix designs. Whilst a 28-day target is common, some structural elements receive deferred loads during construction, allowing a reduced cement content and the possibility to optimize the concrete mix to target full strength at 90 days. Machine learning techniques are used in this study to optimize the concrete mix design of structural elements that commonly receive deferred loads, e.g., foundation and pavements, targeting a 90-day full strength. Specifically, the compressive strength of concrete samples cured for 28 and 90 days are considered to estimate the cement content per m3 of concrete using Artificial Neural Network and Regression algorithms. The proposed machine learning and deep learning methods are proved to be capable of predicting the cement content with 94% and 90% accuracy, respectively. The Elastic Net algorithm shows the best performance in cement content optimization to a target 90-days compressive strength. This algorithm is hence employed to assess the carbon reduction benefits in a real case study: a typical mid-sized reinforced concrete structure is considered as a baseline to quantify the environmental benefit of optimizing the cement content for a 90-days target compressive strength. Results of the case study show that the proposed method may result in a reduction of approximately 10% in cement usage, consequently leading to a parallel reduction of about 10% in carbon emissions.