Session: ASME Undergraduate Student Design Expo

Paper Number: 176043

Application of Machine Vision for In-Situ Process Monitoring for Additive Manufacturing 

With the rise of 3D printer availability and its wide range of applications over the past decade, a notable lack of in-situ and post-manufacturing quality control has been observed in comparison to traditional manufacturing. This lack of quality control hinders the widespread adoption of additive manufacturing (AM) across various industries. To bridge this gap, developers have been looking toward automated machine-vision (MV) algorithms, which can be effective in developing AM technologies for industry-wide adoption. Currently, most research has explored in-situ monitoring methods that aim to detect printing errors during manufacturing. Significant limitations to this approach are the single, fixed camera angle and low resolution, which cause a failure to identify small or hidden defects due to part geometry. In light of this, we are proposing a strategy that utilizes the advantages of MV to address these limitations, specifically, the viability of image recognition algorithms, and how such algorithms can be integrated into the current infrastructure to automatically classify surface faults in printed parts. 

 

In our approach, a machine learning-based vision model, YOLO (You Only Look Once), is adapted and trained on a dataset of AM parts containing prescribed defect categories. We deploy a Cognex vision system (camera plus proprietary software) to acquire images of parts produced by a binder jet 3D printer using sand as the powder and a vegetable oil–based binder. Through the study, a difficulty in detecting “hills” or “valleys” (any uneven spreading of sand) was noticed, and to remedy this issue, a laser grid projecting onto the print bed was implemented. The model is evaluated on the percentage of accurate defect prediction, in comparison to traditional inspection methods. When introduced to a new set of images (errors the model had not seen before), the model obtained an accuracy of 94%. During real-time operation throughout a full print, the model achieved 100% accuracy in identifying all flaws and defects, with no need for manual override of any predictions.

 

This research enhances science and engineering in several key ways. First, it shows that object‑detection algorithms like YOLO can significantly improve defect detection in additive manufacturing, especially for small or subtle defects that tend to go unnoticed by traditional inspection. By training the model on a variety of prescribed defect types and testing it on unseen images, we demonstrate strong generalization and reliability. Second, integrating an MV system in line with the print process allows for continuous monitoring of surface conditions, enabling early detection of defects before they can continue throughout subsequent layers. This reduces waste and improves overall quality control efforts. Third, the high accuracy achieved, 94 % on new images and 100 % detection in real‑time during printing, suggests that automated machine vision is moving past proof‑of‑concept toward reliable deployment. In summary, this work brings additive manufacturing closer to industrial scalability by enhancing inspection fidelity, reducing manual oversight, and improving repeatability and consistency in part quality.

Presenting Author: Haley Romine Georgia Southern University

Presenting Author Biography: Haley Romine is a Sophomore at Georgia Southern University in the Mechanical Engineering department who aspires to be an Aerospace Engineer. She has been involved with the machine vision integration into binder jet printing research project since the first semester of her freshman year (Fall 2024). Last year, Haley and her team received an award at Georgia Southern University's Undergraduate Research Expo. Haley is the Vice President of the Coastal Georgia Section of The American Society of Nondestructive Testing at Georgia Southern University. When not in the lab, Haley enjoys drawing and crocheting.

Authors:

Haley Romine Georgia Southern University
Elsie Lappin Georgia Southern University
Hossein Taheri Georgia Southern University