Data Supporting "An Efficient and Uncertainty-aware Reinforcement Learning Framework for Extrusion Additive Manufacturing"

This dataset supports the paper "An Efficient and Uncertainty-Aware Reinforcement Learning Framework for Extrusion Additive Manufacturing." It provides zero-shot evaluation data for error correction performance on a Creality CR-20 Pro fused deposition modeling (FDM) 3D printer. The dataset includes endoscope images captured by a nozzle-mounted endoscope during 21 3D printing tests, each simulating a different extrusion error. Images focus on the extrusion region, are time-stamped, and synchronized with print logs recording key process parameters. Errors are introduced by varying the initial flow rate and nozzle temperature, followed by control adjustments to observe defect emergence and resolution. Each test contains: 1. A time series of raw endoscope images collected at 20 Hz. 2. Cropped images (350×350 pixels) centered on the nozzle tip to highlight the surrounding region. 3. A print log file with image filenames, timestamps, nozzle temperature, flow rate, and other process parameters. Print logs link each image to its corresponding control parameters, capturing both initial error states and real-time adjustments. Combined with the images, they offer insights into the printer’s behavior under varying error conditions.