Keywords: additive manufacturing, artificial intelligence, crimping tools, data augmentation, neural networks, quality control.

In this paper, the use of neural networks is investigated in the course of optical inspection control of crimping tools and the economic benefit of data augmentation of training data. It should be noted that the augmentation of training data can possess a positive effect on the prediction accuracy of neural networks. Using data augmentation, small data sets can be artificially enlarged for the training process of convolutional neural networks. The goal is to increase the prediction accuracy of convolutional neural networks in recognizing real test data. The original images are augmented in such a setting, so that the important features of the real images are still recognizable. In the process of this work, various images of crimping tools produced by a 3D printer are captured. The crimping tools were additively manufactured from both black PLA and filament with wood content. These crimping tools are to be classified as defective or properly manufactured components through visual inspection. Various defects that can occur during additive manufacturing will also be mapped. Based on the conducted experiments and results, it can be stated that the prediction accuracy can achieve high accuracy of the model with lower number of real data per class using the augmentation of the training data. Thus, data augmentation can be evaluated as a suitable method of data augmentation in the field of optical inspection control of additively manufactured crimping tools.

download the full-text of this paper (PDF, 6 pages, 474 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description