Thermal Data-Driven Quality Prediction in Wire Arc Additive Manufacturing via Machine Learning and Anomaly Detection

Abstract

Maintaining consistent quality control is a significant challenge in Wire Arc Additive Manufacturing (WAAM). However, real-time monitoring remains difficult due to the impracticality of manual labelling of complex, continuous thermal data during production. This paper proposes a framework utilizing temperature data and the combination of machine learning and anomaly detection algorithms. The temperature data was collected during the fabrication process of a five-layer wall using six thermocouple sensors attached to the baseplate. DBSCAN clustering algorithm was employed to identify the anomalies in the temperature data to classify them into ‘normal’ and ‘anomaly’ categories. In this study, six supervised machine learning algorithms were applied to predict the weld quality. Across the ten experimental trials, it was demonstrated that Random Forest scored the highest accuracy of 98.62% and AUC of 99.45%. Results proved that the combination of anomaly detection and machine learning algorithms with the thermal data is highly effective in predicting the quality in the WAAM process.