While we didn’t fully achieve our application goals, this project gave me valuable insight into the software aspects of mechanical products and provided me with a more holistic understanding of 3D-printing product development.
This project leverages machine vision techniques in fused filament fabrication to detect and alert users to the common extrusion 3D printing error known as stringing. This project originated from my Advanced Manufacturing class (ME500A2) at Boston University. In another course, we frequently used Creality Ender-3 printers to build assemblies. Unfortunately, many students experienced the frustration of returning to failed prints after hours of waiting. This wasted both time and effort, prompting our group to develop a solution. Given the tight timeline, we chose to focus specifically on detecting stringing, a common printing failure, and aimed to notify users of the error rather than directly controlling the printer.
As mechanical engineers unfamiliar with software engineering, we began by familiarizing ourselves with the necessary software packages. After some research, we identified the key tasks required for a successful product: developing a machine-learning model to generate annotation data when stringing was detected in an image, processing the image from a webcam feed to include these annotations, and sending this image data to the user via email upon print failure. We discovered RoboFlow, an API that allowed us to upload and manually annotate images for our machine-learning model. I then established a reliable connection to a webcam, enabling us to consecutively run inference on images using our computer vision model on RoboFlow. Upon receiving the inference data from RoboFlow, I could create annotations on the image to show where the stringing occurred. Although we could detect stringing with our model reliably, our timeline did not allow the application to reach full functionality. Our biggest challenge was notifying the user of the error via email, encountering issues with access token validation and image attachments. With class projects continuing across multiple semesters, we hope that future students can build upon our work to improve the application's reliability. To access the GitHub for this project, click the button below.
Your interest in my work is truly appreciated. If any of my projects interest you, please don't hesitate to use the contact form below. I look forward to the possibility of speaking with you.
