Machine Learning Approach for the Prediction of Defect Characteristics in Wire Arc Additive Manufacturing

Additive manufacturing (AM) technology is becoming one of the demanding manufacturing methods for various applications. Various AM methods have been developed to produce the near net shape components with complex geometry to increase productivity. On the other hand, standards for processing, testing, and quality control are under development. Wire arc additive manufacturing (WAAM) can produce large-size components using different welding processes. Quality is a significant concern for AM components, and mainly the variations in product quality cause a major barrier to the general application of WAAM in a production environment. To solve this issue, welding arc stability, voltage, arc length, and bead geometry formation sensors have been widely used to detect defects of WAAM deposits. The present study introduces an acquiring data system from the welding arc to identify the defects in WAAM and make a new wave of WAAM-related data available. The experimental findings and predicted data provide a valuable resource for achieving unique insight into WAAM processes and decision-making. Machine learning (ML) offers the opportunity to advance this insight by learning fundamental knowledge about WAAM processes and identifying predictive and actionable suggestions to optimize part quality and process design. The analysis of the defects of the first layer and successive layers data has been used to train the ML models and predict the WAAM process defects of further layers. The predictive models successfully developed the new prediction algorithms to detect the defects in WAAM deposited layers. The validation test was suitable for predicting the defects during the deposition process.