In-situ droplet inspection and closed-loop control system using machine learning for liquid metal jet printing

Abstract Liquid Metal Jet Printing (LMJP) is a revolutionary three-dimensional (3D) printing technique in fast but low-cost additive manufacturing. The driving force is produced by magneto-hydrodynamic property of liquid metal in an alternating magnetic field. Due to its integrated melting and ink-jetting process, it can achieve 10x faster speed at 1/10th of the cost as compared to current metal 3D printing techniques. However, the jetting process is influenced by many uncertain factors, which impose a significant challenge to its process stability and product quality. To address this challenge, we present a closed-loop control framework by seamlessly integrating vision-based technique and neural network tool to inspect droplet behaviours and accordingly stabilize the printing process. This system automatically tunes the drive voltage applied to compensate the uncertain influence based on vision inspection result. To realize this, we first extract multiple features and properties from images to capture the droplet behaviour. Second, we use a neural network together with PID control process to determine how the drive voltage should be adjusted. We test this system on a piezoelectric-based ink-jetting emulator, which has a very similar jetting mechanism to the LMJP. Results show that significantly more stable jetting behavior can be obtained in real-time. This system can also be applied to other droplet related applications owing to its universally applicable characteristics.