C³M-OMMT twin: A causal process–structure–property digital twin for morphology-aware mechanical optimization of OMMT-Modified PLA in fused-filament fabrications

Abstract

Mechanically reliable fused-filament-fabricated (FFF) PLA/OMMT parts depend not only on filler fraction and nominal printing settings, but also on the coupled evolution of clay dispersion, melt rheology, thermal healing, void topology, and strain localization. This study develops C³M-OMMT Twin, a six-stage causal process-structure-property digital twin comprising ExfoRheo-MAP, ThermoBond-Sense, MorphoFrac-3D, CauPSP-Net, ConBayes-Pareto, and CounterTwinV. Approximately 60 D -optimal material-process combinations were used for model development and 20 Bayesian-selected trials were reserved for experimental closure. The design space covered 0–5 wt% OMMT, 195–225 °C nozzle temperature, 0.10–0.25 mm layer height, 25–55 mm s⁻¹ print speed, raster orientation, and infill. A directly characterized condition at 2 wt% OMMT, 215 °C, 0.15 mm, and 35 mm s⁻¹ produced 62.5 MPa tensile strength and approximately 2.2% void fraction. On held-out data, CauPSP-Net achieved R²= 0.974, RMSE= 1.72 MPa, MAE= 1.29 MPa, and MAPE= 2.53%; 90% conformal intervals achieved 90.7% empirical coverage with a 4.46 MPa mean width. Model-based multi-objective optimization identified the same operating neighborhood, with predicted Young's modulus of 3.82 GPa, flexural strength of 94.1 MPa, and CVIFSI of 0.162, while CounterTwinV reported a 3.1% validation discrepancy. The originality lies in retaining physically interpretable intermediate states from exfoliation and rheology through thermal bonding and three-dimensional defect morphology, then separating direct and morphology-mediated effects under calibrated uncertainty. The resulting manufacturing window is intended for reproducible, morphology-aware optimization of lightweight PLA/OMMT FFF components rather than unrestricted extrapolation beyond the validated material and process domain.