Rapid prediction of the cure-induced distortion of newly developed composite parts via transfer learning

Accurately predicting Cure-Induced Distortion (CID) is paramount to ensuring manufacturing precision. Existing prediction models heavily rely on material parameters associated with the CID, which presents challenges for newly developed composites due to the time-consuming, expensive, and limited accuracy of parameter characterization processes. With recognizing the shared curing mechanism across various composites, this paper introduces a transfer learning-based approach that leverages a small dataset of practical CID data from the newly developed composite and historical knowledge from existing composites to achieve rapid and accurate CID predictions. This approach was validated for two typical CID prediction cases by implementing a transfer learning framework, and various sampling methods were explored to assess the stability and repeatability of prediction results. The results indicate that the proposed method can predict the CID within a 5% margin of error. This paper offers a promising approach for predicting CID in the design and manufacturing processes.