Concrete Decisions: How XAI is Paving the Way for Future Construction Materials

Abstract

For approximately twenty-five years, machine learning methods have been used to develop predictive models applied to construction materials. Concrete in particular is widely studied as it is the core of this industry, seeking to improve its properties to comply with both safety standards and market demands for more competitive products. There are major challenges in this area, one is the need for reliable data for the correct training of models, and other is understanding the choices made by computational methodologies to achieve such accurate models. To increase confidence in these useful tools, for example, when deciding to change a formulation and estimate its mechanical profile, it is necessary to evaluate the behavior of the model. For this, explainable artificial intelligence methodologies are beginning to be used. In this paper we review problems and advances in the area, hoping to contribute to the decision-making of construction engineers.