Inspecting Method for Defective Casting Products with Convolutional Neural Network (CNN)

It is essential to conduct the quality control for gauranteeing sound products after finishing conventional manufacturing processes. Vision-based inpection system has been extensively applied to various industries linked with concept of the smart factory since it does not only enhance the inspecting accuracy, but also decrease the cost for the human inspection, substantially. This paper mainly concerns the development of the inspecting system for the casting products with supported by the convolutional neural network, which makes it possible to detect various types of defects such as blow hole, chipping, crack, and wash automatically. To obtain high accuracy in inspecting system, it does not only require sub-partitioning of the original images, but also apply multiple labeling according to the order of the sub-images and the existence of the defects. Performance of the proposed inspecting algorithm has been validated with the 400 casting products, in which it exhibits substantially high accuracy more than 98%, experimentally.

Automated summary

This paper focuses on the development of an inspecting system for casting products supported by convolutional neural network, which can automatically detect various types of defects. In order to achieve high accuracy in the inspecting system, sub-partitioning of original images and multiple labeling based on the order of sub-images and defect existence are required.