Enhancing hybrid manufacturing with AI-driven real-time adaptive process control: integrating machine learning models and robotic systems

The potential of additive methods to integrate with machine learning (ML) models in hybrid manufacturing is significant in developing closed-loop processes for intelligent and adaptive control. This study presents a framework to defect mitigation that leverages a hybrid deep convolutional neural network (CNN) architecture with laser powder bed fusion additive systems. It connects on-site optical vision and sensor streams to identify process abnormalities and report real-time remedial actions. The hybrid CNN automatically connects reported anomalies to the physical actuation where quality issues might occur in a closed-loop fashion. This enables self-learning and autonomous corner-cutting while offering a pathway to new applications in digital manufacturing, such as surface finish optimisation and shape adaptation within a single build. In a case study, we demonstrate that our framework enables inspection and control at a level never seen before, as demonstrated by its precision in product inspection with an F-score of 93.8% and its ability to adjust control at a frequency exceeding 10 Hz. This work not only enables new learning paradigms for the development of real-time control systems for hybrid manufacturing, but also accelerates the process of AI-driven intelligent manufacturing by equipping industrial production systems with intelligence, flexibility and resilience. Our technique for tight coupling of process signals and actuation is a significant technical breakthrough in this critical pathway.