Printed PLA gets a causal chain from clay exfoliation to layer bonding and voids, reaching 62.5 MPa at 2 wt.% clay

The mechanical performance of a printed part is no longer treated as set only by filler percentage and print parameters; clay exfoliation, melt rheology, interlayer thermal bonding and pore morphology enter one prediction chain as intermediate stages. Earlier work on OMMT-modified PLA prints relied mostly on single measurement techniques, and optimization used response surfaces or parameter-to-property machine learning that excluded clay exfoliation, melt flow, heat-induced interlayer bonding, void topology and local fracture sites. At 0, 1, 2, 3 and 5 wt.% clay the pipeline gives tensile strengths of 50.3, 56.4, 62.5, 59.6 and 52.9 MPa, each above the corresponding values of the four comparison models, MOGA-ANN, response surface methodology, ANN and ensemble learning.

The chain runs from material state through thermal history and morphology to mechanical response and uncertainty-aware optimization, then closes with experimental validation. ExfoRheo-MAP combines X-ray diffraction basal spacing, the agglomerate area fraction from microscopy and complex viscosity under oscillatory shear into a material-state vector, and ThermoBond-Sense uses it to reconstruct layer-by-layer heating and cooling and estimate healing between adjacent roads. Compounding used a twin-screw extruder at 185 to 195 °C, 60 rpm and 4 to 6 minutes residence time, and specimens were printed at four nozzle temperatures from 195 to 225 °C, four layer heights from 0.10 to 0.25 mm, six speeds from 25 to 55 mm/s and several raster orientations, with about sixty D-optimal combinations for training and twenty Bayes-selected trials for validation.

The optimization lowers voids and interface fracture susceptibility rather than only raising tensile strength. MorphoFrac-3D builds a defect-clay-interface graph from pore volume, sphericity, connectivity, orientation and nearest-neighbor spacing together with clay agglomeration and strain localization, and defines the Clay Void Interface Fracture Susceptibility Index (CVIFSI). The pipeline gives a void fraction of 2.18%, agglomerate fraction of 4.3%, interlayer bond index of 0.88, exfoliation index of 0.87 and CVIFSI of 0.162, against 2.71%, 6.2%, 0.82, 0.80 and 0.224 for ensemble learning.

The scope covers PLA specimens compounded in a twin-screw extruder, printed as 1.75±0.05 mm filament at 0 to 5 wt.% organo-modified montmorillonite and the stated nozzle temperature, layer height, speed and raster combinations, plus public PLA and carbon-PLA datasets used for transfer evaluation. The authors propose next to add in-situ thermal imaging, acoustic emission, melt pressure sensing and vision-based defect detection for closed-loop operation, along with molecular dynamics descriptors and multi-fidelity Bayesian optimization.

The reported optimum and values such as 62.5 MPa come from the pipeline's predictions and its experimental closure, with a validation error of 3.1%, not from independent replication in another laboratory. The abstract's 59 to 65 MPa tensile range, 3.5 to 4.0 GPa modulus and 35 to 55% void reduction are anticipated ranges, stated differently from the single-point results in the body. Ablation shows that removing any module lowers performance, so the modules' contributions are entangled and single-module attribution needs care.