An AI-based partial explainable prediction of rubber concrete strength on mobile devices

Recently, there has been a growing trend in utilizing waste rubber as a partial replacement for aggregates in concrete. This approach not only promotes the reuse of waste rubber but also addresses the shortage of natural aggregates. An issue arising from various compositions between rubber and the cement matrix is the accurate prediction and control of the mechanical properties of rubber concrete, which impedes the widespread application of rubber concrete because the indispensable on-site mechanical tests are time-consuming and labor-intensive. In response to this challenge, an integrated AI-based approach that enables the real-time prediction of the compressive strength of rubber concrete through mobile devices was proposed. Firstly, a U-Net-based semantic segmentation model is employed to identify different compositions within cross-section photos of rubber concrete. Subsequently, an artificial neural network (ANN) model is adopted to promptly and precisely predict the compressive strength of rubber concrete using the proportions of the semantic segmentation compositions. The proposed approach is validated through a database based on past experimental results. The U-Net-based component recognition model achieves an accuracy of 89.31 %, while the strength prediction model attains an accuracy of 82.08 %. Overall, this method effectively identifies various compositions and establishes a correlation between their proportions and the compressive strength of rubber concrete. This provides a partially explainable and efficient approach for the widespread on-site application of rubber concrete.