Deep Learning Models for Visual Inspection on Automotive Assembling Line

Abstract

In automotive manufacturing, assembly tasks depend on visual inspection to ensure product and process quality, for example, scratches identification on machined surfaces or correct part identification and selection, which may belong to more than one type of vehicle to be produced on the same manufacturing line. Typically, these tasks are essentially human-led, which have recently been supplemented by the artificial perception provided by computer vision systems (CVSs). Despite their relevance, the accuracy of CVSs depends mostly on the environment control, providing appropriate lighting, enclosure, and stops for images to be collected. These problems makes the solution expensive and overrides part of its benefits, mainly when it interferes with the operating cycle time. Thus, this paper proposes the use of deep learning-based methodologies to assist in the visual inspection task, generating little influence on the original manufacturing environment and exploring it as an end-to-end tool to ease CVSs setup. The approach has illustrated by four proofs of concept in a real automotive assembly line based on models for object detection, semantic segmentation, and anomaly detection.