Failure Detection on 3D Printing Images with Color-based Data Augmentation

The advent of three-dimensional (3D) printing has profoundly transformed the realms of design, manufacturing, and prototyping. Despite its revolutionary impact, the technology introduces novel challenges that may lead to operational setbacks. The vast spectrum of filament colors available in 3D printing introduces complexity, necessitating a comprehensive investigation into strategies for enhancing error detection. This research focuses on evaluating the effectiveness of color-based data augmentation applied to neural networks for detecting failures in 3D printing processes. This research uses grayscale and Hue-Saturation-Value augmentation to improve error detection. The effect of this augmentation was tested on data gathered from shared failure from 3D printer user on the internet. The study achieves 9 points improvement on recall, underscoring the potential of color-based data augmentation techniques to enhance the accuracy of defect detection in 3D printing operations. This underscores the significance of exploring diverse augmentation methods to optimize the detection model performance, with grayscale augmentation identified as a particularly promising approach in specific situations.