Machine Learning Approaches and Multi‐Physics Coupling Model for Predicting Fiber Volume Fraction of Composite Laminate Cured by Autoclave Processing

ABSTRACT

In composite curing, accurate prediction of fiber volume fraction is essential for process optimization and quality evaluation. Autoclave experiments and thermogravimetric analysis were conducted to obtain fiber volume fraction data under different curing pressures. A thermal–flow–solid multi‐physics model considering resin flow and fiber compaction was established, and five machine‐learning models, including SSA‐LSSVM, GA‐LSSVM, SSA‐RBFNN, GA‐RBFNN, and KNN, were developed. Results show that fiber volume fraction increased from 52.28% at 0 MPa to 66.33% at 0.8 MPa. Meanwhile, density increased from 1.415 to 1.618 g/cm ³ , and the maximum thickness deviation decreased from 0.42 to 0.14 mm. The flow‐compaction model yielded prediction errors within 5.2%, while larger deviations occurred under low‐pressure conditions because pore delamination and related defects were not explicitly considered. Given the limited dataset of 17 groups, a unified outer 5‐fold cross‐validation protocol was adopted for all machine‐learning models, and performance was evaluated using aggregated out‐of‐fold predictions. Among all models, SSA‐LSSVM showed the best overall performance, with R ² values of 0.996378, 0.980822, and 0.979296 for fiber volume fraction, density, and thickness, respectively, indicating the highest accuracy and strongest generalization capability. These results demonstrate that the proposed framework provides an effective method for predicting fiber volume fraction and related physical properties of autoclave‐cured composite laminates.