Personal Image Classifier Based Handy Pipe Defect Recognizer (HPD): Design and Test

Abstract

Pipelines are known as a traditional solution for transporting various media such as gas, oil, and water. But pipes are always combined with the defects. Some of these defects are caused by the manufactory process, while some defects occur after installation due to the type of medium and environmental condition. Currently, various methods are used to detect these defects from industry to the site, but vision-based systems due to huge amount of data that can capture machine learning development algorithms have more demand. In this research paper Personal Image Classifier (PIC) is used as the machine learning method combine with the MIT APP inventor to make the handy system called Handy Pipe Defect assistant recognizer (HPD) to address the defect name and help the user to investigate and somehow correct their process of product in industry (as the production stage) or the possible change in usage place (as the maintenance stage). The HPD designed based on the hypothesis of having the handy, portable and available tool for the operator in industries as the quality control and site engineers. The HPD can classify the pipe defects, especially the welding type by taking the picture of the defect and give the user feedback. The HPD focuses on the welding defect with the manufactory source over 150 image number trained and tested with the real 28 sample image. The design process and training model for HPD described and result based on sample images from manufactory and real situation are shown the 100% defect name detection of the defect along with image quality affected by environment light and similarity between the defects.