A new strategy using intelligent hybrid learning for prediction of water binder ratio of concrete with rice husk ash as a supplementary cementitious material

Abstract

It is important to point out that the precise prediction of water binder ratio “w/b ratio” is indispensable for gaining the desirable characteristics of strength and duration of concrete constructions. This research offers a new method for w/b ratio prediction based on state-of-art machine learning algorithms accompanied with Explainable artificial intelligence (XAI) methods. The main aspect of the research approach is described using 192 database containing different mix design parameters and the environmental conditions. With the help of ensemble learning models such as Random forest (RF), Recurrent neural network (RNN) model, Relevance vector machine (RVM) and Response surface methodology (RSM), the prediction model has performed better than the empirical methods with RVM-M3 surpass all other models with the highest R value equal to 0.9992 in calibration phase and RF-M3 surpass the other model combination in verification phase with R value equal to 0.9984. Moreover, addressing the integration of XAI, the specifics of model prediction and the main influential variables related to w/c ratio as well as their importance are determined, where Cement (Ce) highlight to be the most influence parameter that improved he prediction accuracy of RF-M3 model. The results prove that the proposed method increases the prediction accuracy and provides engineers with a dependable means of augmenting concrete mix designs to enhance concrete’s durability performance and sustainability. This research expands the understanding and principles of concrete technology, hence facilitating the use of AI-based solutions in civil engineering practices and other relevant domains.