<p>Aerosol jet (AJ) printing is a versatile additive manufacturing process capable of producing high-resolution interconnects on 2D and 3D substrates. The process is complex with many unobservable states, including aerosol particle diameter, carrier density, vial level, and ink deposition in the tube and nozzle, that influence performance. Widespread adoption of AJ printing is limited by print inconsistencies that stem from variability in these hidden states. To address this, we develop a digital twin of the AJ process that infers and tracks latent states of the machine, giving operators visibility into otherwise unobservable dynamics affecting part quality and data traceability. The digital twin is built around a physics-based reduced-order model derived from simulation and experimentation. The states and parameters of the digital twin are continuously updated using probabilistic sequential estimation to align with real-time sensor and video data. The resulting digital twin evolves over the machine lifecycle, enabling monitoring of hidden characteristics, anomaly detection and prediction, and forecasting of control adjustments. This work presents a comprehensive framework that integrates computer vision, reduced-order modeling, and probabilistic estimation. While the methodologies are customized for AJ printing, the process for constructing the digital twin is transferable to other advanced manufacturing processes.</p>