Machine learning guided design of microencapsulated phase change materials-incorporated concretes for enhanced freeze-thaw durability

Abstract

Prolonged exposure to freeze-thaw cycles (FTCs) induces adverse effects in concrete structures in terms of a significant reduction in strength and durability. This paper proposes a microencapsulated phase change material-incorporated concrete (MPCMC) layer to mitigate FTCs in the concrete bridge decks. Toward this, the present work implements finite element analysis (FEA)-based multiscale numerical simulation, which shows the effectiveness of the proposed MPCMC protective layer toward FTC reduction. Moreover, the multiscale framework is leveraged to generate a large dataset containing 8096 data points of FTCs considering a uniform variation of five design parameters, namely the thickness of the protective MPCMC layer, volume fraction, transition temperature, size, and the latent heat of the MPCM. Thereafter, the dataset is leveraged to develop an interpretable neural network (NN)-based predictive model as a computationally efficient alternative to the complex multiscale FEA-based design tools. The interpretable machine learning (ML) model provides a reliable and efficient continuous mapping between the design parameter space and the resultant number of FTCs. Moreover, it explains general engineering intuitions and physics. Besides, the interpretable ML characterizes various underlying interplays of the design parameters in dictating the resultant FTCs.