Real-time in-situ defect detection via multi-modal classification of waveforms for wire arc additive manufacturing

Metal Additive Manufacturing (AM) is a promising set of new technologies, enabling rapid prototyping and near-net shape fabrication of parts across a wide variety of scales ranging from small complex geometries to large-scale components at reduced production and material costs. However, compared to traditional manufacturing strategies, metal AM is immature and under-developed, leading to difficulties in widespread adoption in industry. One such technology is Wire Arc Additive Manufacturing (WAAM), where a controlled platform is used to synthesize parts layer by layer via the welding process. WAAM parts and processes currently suffer from several defects due to unknown interactions among process variables, all of which significantly impact the overall part quality. Traditionally, human welders have been able to detect welding quality via audio signals emitted from the process while welding, leading to them adapting certain process variable settings on-the-fly for improved part quality. This insight has led many previous researchers to explore using audio waveform signatures measurable via in-situ instrumentation for WAAM weld quality monitoring. However, previous research has failed to address three primary topics covered in this paper: first, it is undetermined which exact defects are measurable via in-situ instrumentation; second, other process variable waveforms (such as welder current and voltage) have not been incorporated for predicting potential defects; and third, deep machine learning methods have not been incorporated for providing solutions to the previous two topics. This paper explores these topics, with the aim of providing a basis for the defect detection of the WAAM process via strict in-situ monitoring.