Study on an aided design method for partially composite precast concrete sandwich panels based on degree of composite action prediction

Abstract

Precast concrete sandwich panels (PCSPs) derive their structural performance primarily from the degree of composite action (DCA) between the two concrete wythes. Due to the coupled influence of multiple design parameters on DCA, traditional experiments and case‐by‐case finite element (FE) analyses are inefficient for large‐scale parametric studies and rapid design iterations. This study proposes a machine‐learning‐based DCA prediction and aided design method for partially composite PCSPs to predict the initial‐stiffness‐based degree of composite action (DCA _s ) and the load‐bearing‐capacity‐based degree of composite action (DCA _c ). A two‐dimensional FE model of PCSPs was first developed and validated. A parametric dataset containing eight design variables and 591 samples was then generated through Latin hypercube sampling. Ten machine‐learning models were developed and evaluated using cross‐validation and an independent test set. The SHapley Additive exPlanations (SHAP) results identified connector initial stiffness and connector capacity as the dominant predictors of DCA _s and DCA _c , respectively. Based on these findings, a SHAP‐guided workflow for rapid DCA evaluation and parameter optimization support was developed, together with an aided design graphical user interface (GUI) for PCSPs. The proposed method provides an efficient data‐driven tool for evaluating composite performance, analyzing parameter sensitivity, and supporting preliminary design optimization of partially composite PCSPs.