CNC Milling Machine Anomaly Classification with Continual Active Learning

Computer Numeric Control (CNC) milling machines on the manufacturing field has significantly impact on product quality. Anomalies during CNC milling can disrupt manufacturing costs. Addressing such failures manually presents substantial challenges. As an alternative, artificial intelligence (AI) has been used manufacturing sites, enabling real-time data analysis and anomaly detection. However, AI’s realization relies on quality data acquisition, a time-intensive process compounded by labeling. We proposes a continual active learning framework to enhance AI model learning from limited datasets in manufacturing. This paper presents three active learning methodologies: Least Confidence (LC), Entropy Sampling (ES), and Active Transfer Learning for Adaptive Sampling (ATLAS). We evaluate these methods using CNC milling machine data against Simple Random Sampling (SRS). Results highlight those methods have improved performance rather than SRS.