Theory‐guided machine learning for optimal autoclave co‐curing of sandwich composite structures

The bonding of a honeycomb core to the thermoset prepreg facesheets by co‐curing them allows one to manufacture composite sandwich structures in a single operation. However, the process is strongly dependent on the prescribed autoclave cure cycle. A previously developed physics‐based simulation can predict the bond quality as a function of the process parameters. The disadvantage of physics‐based simulations is the high computational effort needed to identify the optimal cure cycle to fabricate sandwich structures with desired bond‐line properties. Theory guided machine learning (TGML) methods have demonstrated their capabilities to reduce the computational effort for different applications. In this work, three TGML models are trained on a data set produced from physics‐based simulations to predict the co‐cure process of honeycomb sandwich structures. The accuracy of the TGML models were compared to select the best performing predictive tool. In addition to reduction of computational time by orders of magnitude, we demonstrate how the TGML tools can also quantify the contribution of each process parameter on the properties of the fabricated part. The most accurate model was implemented in an optimization routine to tune the input process parameters to obtain the desired properties such as the bond‐line porosity and facesheet consolidation level. This methodology could be extended to any process simulation of composites manufacturing processes. Theory guided machine learning (TGML) methods have demonstrated their capabilities to reduce the computational effort for different applications. In this work, three TGML models are trained on a data set produced from physics‐based simulations to predict the co‐cure process of honeycomb sandwich structures. The accuracy of the TGML models were compared to select the best performing predictive tool. In addition to reduction of computational time by orders of magnitude, we demonstrate how the TGML tools can also quantify the contribution of each process parameter on the properties of the fabricated part.