Deep Learning-Based Real-Time Engine Part Inspection With Collaborative Robot Application

Abstract

Vehicle manufacturing requires error-free processes, as modern vehicles are made up of thousands of parts, including around 280 critical components for safe driving. According to the National Highway Traffic Safety Administration (NHTSA), 2% of vehicle accidents will be caused by defective parts. Current inspection systems in manufacturing plants have limitations, with a high risk of defective parts reaching consumers and leading to recalls. This study aims to develop a real-time, deep learning-based engine part inspection system to improve accuracy and efficiency in mass production. The system, implemented in a large automotive manufacturing plant, uses a Fanuc CR-15ia collaborative robot to inspect engine parts. Combining the single-shot detector (SSD) and faster region-based convolutional neural network (R-CNN) algorithms, the system achieves 99.9% accuracy measured after four months of use, with an Average Precision (AP) of 0.994 for Faster R-CNN and 0.955 for SSD. The inspection system addresses cycle time concerns and is integrated with factory systems for real-time data sharing. Ongoing enhancements aim to improve system performance further.

Keywords

• Collaborative robot
• part inspection
• deep learning
• real-time applications