EVALUATING VARIOUS MACHINE LEARNING MODELS FOR PREDICTION OF MICROMECHANICAL PROPERTIES OF ENGINEERED CEMENTITIOUS COMPOSITES

Abstract

Engineered Cementitious Composites (ECC) are high-performing cementitious materials with fiber reinforcement exhibiting outstanding mechanical properties such as ductile behaviour, high tensile strength, and outstanding crack resistance. It is, however, challenging to design and optimize ECC mixtures due to the numerous interactions between the different components and the lack of accurate prediction models. This study develops, validates, and evaluates machine learning (ML) based predictive models for some of the ECC mechanical characteristics i.e., Compressive Strength (CS), Tensile Strain (TS), Flexural Strength (FS) to provide useful information for material design and optimization. In a comprehensive dataset, approximately 400 ECC combinations were included, along with fiber types, volume fractions, binder proportions, and other important components. To enhance the dataset, certain mechanical properties namely CS, FS and ultimate TS were measured. Using this dataset, a variety of ML algorithms, such as deep learning, random forests, SVR, and linear regression, were investigated for predicting mechanical features. To evaluating the performance of the models, cross-validation was employed, and the best model was picked based on its accuracy, precision, and generalizability. To assess the effectiveness of each suggested model, a variety of statistical metrics were computed, including mean absolute error (MAE), mean absolute percentage error (MAPE), root mean squared error (RMSE), and coefficient of determination (R2). The selected model underwent additional analysis to comprehend the intricate connections between ECC elements and their influence on the mechanical properties, revealing important insights for material design and optimization. In this research, engineers and researchers have a strong tool to create customized ECC mixtures for various applications by using ML methods to anticipate mechanical characteristics. The created model and its evaluation methodology promote sustainable and resilient infrastructure through improved resource efficiency and cementitious composite performance.