Abstract
This work describes a novel methodology for the quality assessment of a Fused Filament Fabrication (FFF) 3D printing object during the printing process through AI-based Computer Vision. Specifically, Neural Networks are developed for identifying 3D printing defects during the printing process by analyzing video captured from the process. Defects are likely to occur in 3D printed objects during the printing process, with one of them being stringing; they are mostly correlated to one of the printing parameters or the object’s geometries. The defect stringing can be on a large scale and is usually located in visible parts of the object by a capturing camera. In this case, an AI model (Deep Convolutional Neural Network) was trained on images where the stringing issue is clearly displayed and deployed in a live environment to make detections and predictions on a video camera feed. In this work, we present a methodology for developing and deploying deep neural networks for the recognition of stringing. The trained model can be successfully deployed (with appropriate assembly of required hardware such as microprocessors and a camera) on a live environment. Stringing can be then recognized in line with fast speed and classification accuracy. Furthermore, this approach can be further developed in order to make adjustments to the printing process. Via this, the proposed approach can either terminate the printing process or correct parameters which are related to the identified defect.
Highlights
Computer Vision is considered an interdisciplinary field of informatics, mathematics and image processing, which aims to develop techniques and algorithms so that computers can process, interpret and understand visual information such as video and image
We considered the need to balance between good accuracies and fast detection, as the purpose of the model usage is to be deployed in a live environment
We found that the threshold of 0.4 Intersection over Union (IoU) is often selected in similar custom object detection tasks [29,30,31,32], whereas the most used value is 0.5
Summary
Computer Vision is considered an interdisciplinary field of informatics, mathematics and image processing, which aims to develop techniques and algorithms so that computers can process, interpret and understand visual information such as video and image. Initial research in Computer Vision in the 1960s aimed to develop computer algorithms which could mimic the human visual behavior [1]. A research boost is remarkable, where algorithms do try to copy the human eye perception and improve themselves by constantly dealing with new data. Based on this principle, newly Artificial Intelligence applications on Computer Vision were described.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have