Image-Based Defect Detection in Assembly Line with Machine Learning

Giovanni Burresi, Martino Lorusso, Lisa Graziani, Alice Comacchio, Federico Trotta, Antonio Rizzo

Abstract: In this paper we aim to compare different Machine Learning approaches to solve a production problem in embedded devices production factories. In our case, a set of electronics components have to be assembled in a specific order and folded by screws to compose the final product. It could happen that some of the screws may be forgotten. To help the operator we implemented a visual detection system able to detect missing screws. The system is based on two algorithms - YOLO and Binary Classification - with two radically different approaches. We also aimed to verify if the ensemble of these two algorithms will be more robust to the insurgence of false-positives predictions and to prototype a solution that smoothly integrates in the LEAN production line.