Advances in machine learning for parameters optimisation and in-situ monitoring of wire arc additive manufacturing

Abstract Wire arc additive manufacturing (WAAM) demands both real-time monitoring of process stabi-supervised learning for real-time anomility and defects and offline optimisation of process parameters to guarantee part quality and production efficiency. This review critically surveys recent machine learning (ML) techniques for in situ monitoring and parameter optimisation in WAAM, with an emphasis on the integration of ML and bio-inspired optimisation algorithms. In relation to in-situ monitoring, this review examines the roles of supervised and unsupervised learning, as well as advanced deep-learning architectures—such as generative AI and frequency-informed neural networks—in processing welding current and welding voltage, as well as vision-based, audible, acoustic-emission, and thermal imaging data. Furthermore, this paper surveys the latest developments in bio-inspired optimisation models applied to WAAM, discussing how ML-enabled frameworks can enhance sustainability and efficiency metrics in the offline selection of optimal process parameters. The synthesis of insights at the end of each section establishes a structured framework for practitioners, highlights existing research gaps, and outlines strategic directions for future advancements in ML-driven WAAM monitoring and optimisation.