A deep CNN for Image Analytics in Automated Manufacturing Process Control

Abstract

Computer vision is widely used in control of manufacturing processes. However, due to the relatively high costs, automated visual inspection is applied only for the final product and not after each production phase. Faulty products are identified at the end of the production pipeline, little information is provided with regard the causes and the production phase that produced the defect products. This paper presents a deep CNN for the enhancement of the computer vision system by providing additional information on the type of the product and consequently on the production phase that generated that product, and fault, respectively. The CNN has been integrated into the pipeline for the automated process control. This information is important for further decision makings in production flow management. The paper presents full description of the deep CNN model design with discussions based on image analytics carried out on 12k database of images of products from the porcelain industry.