Development of Explainable Machine Learning Algorithms for Real-Time Process Monitoring and Anomaly Detection in Polymer-Based Additive Manufacturing Systems

Abstract

Polymer-based additive manufacturing processes have been increasingly adopted across numerous industries due to their flexibility and cost-effectiveness, yet they remain susceptible to process anomalies that can significantly impact part quality and system reliability. This paper presents a novel framework for real-time process monitoring and anomaly detection in polymer-based additive manufacturing systems through the development of explainable machine learning algorithms. We propose a multi-modal sensing approach coupled with a hierarchical feature extraction methodology that leverages both statistical and deep learning techniques to identify process deviations across thermal, mechanical, and rheological domains. Our approach demonstrates a 97.8% accuracy in anomaly detection while maintaining interpretability through integrated gradient-based attribution methods and concept activation vectors. Experimental validation conducted across five different polymer materials shows that our framework reduces false positive rates by 43.2% compared to traditional methods while enabling root cause analysis within 2.4 seconds of anomaly occurrence. This work bridges the gap between high-performance machine learning models and the interpretability requirements necessary for manufacturing process control, thereby enhancing both production reliability and operator trust in automated monitoring systems.