Autonomous closed-loop optimization for MAX-phase synthesis via image-based quality evaluation and sequence learning

MAX-phase synthesis is highly sensitive to composition and thermal history, making manual exploration inefficient and batch quality difficult to stabilize. In this work, a characterization-driven closed-loop framework is established by coupling a programmable logic controller (PLC)-supervised synthesis robot with image-guided quality assessment and sequence-based process planning. SEM and XRD data are used to classify synthesis outcomes into non-layered, layered-with-impurities, and layered impurity-free regimes, while batch metadata and encoded furnace profiles are integrated to recommend updated synthesis routes. The quality-assessment model yields a micro-averaged receiver operating characteristic curve area (ROC–AUC) of 0.941 on the held-out test set, and the sequence-based planner achieves an out-of-fold ROC–AUC of 0.953 with an average precision of 0.994. Representative Rietveld refinements and EDS mapping further strengthen the experimental interpretation by providing diffraction-based phase analysis and microscale compositional support. Under identical hardware constraints, the closed-loop workflow delivers roughly three times more clean batches within eight experimental cycles and increases the weekly screening rate from 8 to 24 candidate recipes. These results demonstrate an effective route toward autonomous optimization of MAX-phase synthesis under experimentally grounded feedback.