Predicting Quality Issues in Advanced Manufacturing Using Process, Material, and Monitoring Data

Aim: The study aimed to develop and evaluate a data-driven framework integrating material characteristics, process parameters, and in-situ monitoring signals to improve manufacturing failure prediction and identify significant predictive features. Methods: The framework was evaluated using 3,847 manufacturing runs comprising 24 process parameters, 18 material characteristics, and 47 real-time monitoring signals. Z-score standardization and correlation-based feature selection were applied to reduce the number of features from 89 to 73. An ensemble gradient-boosting model with neural feature extraction was used to predict porosity, dimensional changes, and material breaks. Results: The framework achieved a prediction accuracy of 89.1%, an F1-score of 0.868, and an AUC-ROC of 0.937. Adding material characteristics to process parameters improved prediction accuracy from 67.2% to 76.4%. Early prediction accuracies of 74.3% and 82.7% were achieved at 25% and 50% of the process, respectively. Conclusion: Combining process data, material characteristics, and real-time monitoring improves manufacturing defect prediction. The framework has potential to support data-driven process optimization and early quality-risk identification. Recommendation: Further independent and cross-domain validation is required before the framework can be used to modify inspection protocols or deployed in automated factory control.