The teaching of fundamentals of concrete structures has long faced challenges, including outdated knowledge systems, rigid teaching methods, and a lack of integration between theory and practice. This has resulted in low student learning efficiency and a tendency for students to lose their interest. This study deeply integrates machine learning into concrete teaching, and constructs a new teaching path of four-step fusion: predictive modeling-constructing, a nonlinear mapping between input parameters and mechanical responses, breaking through the shackles of traditional formulas; interpretive analysis, quantifying the contribution of variables through interpretable machine learning models, transforming empirical coefficients into intuitive expressions of physical logic; discovery of laws, using symbolic regression to mine mathematical formulas that are both accurate and interpretable, and restore the exploration process of scientific laws; intelligent design, generating multi-objective equilibrium solutions based on genetic algorithms, achieving a leap from manual trial-and-error to intelligent optimization. The new teaching path reshapes the knowledge learning model with data thinking, realizing the transformation from static standardized formulas to dynamic renewable models. It reconstructs the teaching methodology, and realizes the transformation from one-way teaching to interactive exploration between teachers and students. It upgrades the ability evaluation system, and realizes the transformation from result assessment to process tracking, and ultimately promotes students to transform from formula applicators to science discoverers, providing new ideas for the cultivation of interdisciplinary talents in the context of new engineering.