Data-driven discovery of the design rules for considering the curing deformation and the application on double-double composites

Abstract

The process-induced deformation (PID) of composite laminates has been one of the critical problems for engineering structures. While lots of design rules has been proposed for standardize the laminate design, there is a lack of specific rule to follow when controlling PID is a necessity due to the numerous affecting parameters. In this regard, a data-driven framework was proposed in this paper to determine the layup rules to follow for minimizing PID. Two specific machine learning (ML) models were built. One is combined model of convolutional neural networks (CNN) and principle component analysis (PCA) technique for connecting the layup sequences and their corresponding PID. Another one is the symbolic regression model, as an explainable ML technique, to quantitatively evaluate this connection. With the training data generated from the robust numerical simulation, it is found that a proper asymmetry is the key intrinsic factor that makes a smaller PID as it will counteract with the contributions of other extrinsic mechanisms. More importantly, a formula for easy evaluation of the asymmetry is provided to assist in guiding the layup design considering PID constraints. The formula is applied on the design problem of double-double (DD) composites. With the proper asymmetry added onto the original DD layup, the DD composites show a clear improvement on controlling the PID.