Generative AI for performance-based design of engineered cementitious composite

Abstract

Engineered cementitious composite (ECC) has been intensively studied due to its excellent tensile performance. However, classical micro-mechanical design theory of ECC is qualitative and fails to give detailed ECC mixtures at specific tensile parameters. This study aims to develop a performance-based mixture design model to generate ECC mixtures using generative AI method. An experimental database consisting of 129 polyethylene fiber reinforced ECC (PE-ECC) records has been built. The database was used to train one invertible neural network model and two artificial neural network models. A series of PE-ECC mixtures were generated by the proposed model based on desired mechanical performance and sustainable requirements. Based on the experimental results, the developed model was proven to compose PE-ECC mixtures that satisfy the target requirements with a maximum deviation of less than 16%. The neural network-based model can be used in various application scenarios (e.g., low-cost ECC and low-carbon ECC), thus promoting the development of ECC materials in the area of research and engineering application.