Examination of Training Data to Improve Accuracy of Detecting Warpage of 3D Printed Objects Using Convolutional Neural Network

This study aims to improve the reliability of CNN-based warping detection system in Fused Deposition Modeling (FDM) 3D printer by identifying effective training data under blurred input. Image blur, often caused by changes in camera-object distance, significantly reduces detection accuracy. To improve the detection accuracy, we plan to train CNN using five types of datasets. The detection accuracy on blurred test images will then be evaluated to determine effective training data. The experimental results demonstrated that incorporating training data augmented with a combination of blur addition, contrast reduction, and noise addition improved the system's detection accuracy.