Development of Corrosion Segmentation Using Deep Learning Double Architecture Method to Assist the Analysis and Evaluation Process of Corrosion Inspection

Abstract

Corrosion of pump unit components often occurs in coal mines and can lead to frequent failures of some components. As a result, a corrosion inspection needs to be performed on each component to minimize the possibility of damage. Currently, manual inspection methods are used for corrosion testing but there are still metal defects in the form of corrosion that are uninspected. Therefore, this study aimed to develop corrosion segmentation using computer vision with deep learning double architecture method for detection and evaluation of metal corrosion in order to reduce the loss due to manual inspections. To produce a faster and more accurate analysis method, deep learning double architecture algorithm, namely VGG16-UNET, can be applied with the help of computer vision technology. Consequently, the use of VGG16-UNET method achieved an accuracy of 98.42%. This is in contrast with the single UNET architecture, which produced an accuracy of 92.6%. Based on these findings, it was concluded that the development of this recommended inspection made the analysis and evaluation of corrosion inspection to be quick and easy.