Physics-Informed Machine Learning Framework for Predicting Joule Heating Activated Shape Recovery in Carbon Fiber Reinforced Shape Memory Epoxy Composites
Year: 2026
Type: article
Abstract: Shape memory epoxy composites recover a programmed shape when Joule heating drives the matrix through its glass transition, and a key challenge for these smart materials is predicting the recovery angle at drive currents outside the tested range. In this work, three models were compared for this task: a Kohlrausch–Williams–Watts (KWW) theory based Physics-KWW constitutive model, a gradient-boosted tree model (XGBoost) using thermomechanical state features, and a support vector regression (SVR) model using raw experimental variables together with thermomechanical features. All models were assessed by a leave-one-current-out cross-validation method over nine current levels from 1.50 to 3.50 A. The physics-informed models generalized well to unseen currents, with the Physics-KWW model and XGBoost achieving R2 values of 0.911 and 0.915, respectively, compared with 0.841 for the SVR model under the same tuning protocol. Shapley analysis of a representative XGBoost fold identifies accumulated activation and estimated temperature as the dominant predictors. The results demonstrate comparable accuracy for the two physics-informed approaches, while their advantages remain conditional on the tested pipelines and calibration assumptions.
Source: Journal of Composites Science
DOI: 10.3390/jcs10100537
Authors: Mehrab Hasan, Carter Thomas, Yingtao Liu
Institutions: University of Oklahoma, Norman Regional Hospital
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,172
OpenAlex ID: W7221507002