Diagnostic of the DED-LB process via multi-sensor monitoring and supervised machine learning

Laser Beam Directed Energy Deposition (DED-LB) is mainly used for coating and remanufacturing applications. Moreover, this technology can also be used for the production of functional components, although its industrial adoption in this context remains limited. Ensuring the quality of manufactured components is a complex task unless destructive tests are performed, which do not allow the certification of 100% of the components. To overcome this limitation, this study introduces a novel approach centred on the development and experimental validation of two Machine Learning (ML) models capable of detecting small fluctuations in processing conditions and predicting clad geometry in the DED-LB process. By leveraging monitoring data, these models enable rapid and reliable assessment of process deviations and build quality, facilitating their integration into online defect-detection frameworks where fast and accurate decision-making is essential. The first model, a classification algorithm, is trained on monitoring data and achieved an accuracy of 90% and a precision of 83% for the reference class, with less than 20% false positives. The second model, a regression algorithm, predicts key geometrical features of the deposited clads: width, height, depth, area, and dilution, with an average 92% R2 value. Both models are validated under real working conditions using clads with intentionally induced process deviations, demonstrating their ability to detect anomalies and assess their impact on build geometry with a maximum relative error below 8.9%. These results highlight the potential of ML-based techniques to enable real-time quality assurance in DED-LB manufacturing. Graphical abstract