Session: 03-16-01: AI Integration in Mechanical Engineering and Smart Manufacturing

Paper Number: 142813

142813 - Ai Powered Robot-Assisted Automated Assembly Inspection System for Industry 4.0 

The objective of this research project is to develop an advanced AI-powered robotic system for automated assembly inspection within the industry 4.0 framework. This system leverages a Fanuc robot guided by vision-based AI technology, integrating computer vision and AI algorithms within MATLAB, and subsequently evaluating its performance. The research work makes a significant contribution to the integration of AI and robotics within advanced manufacturing by implementing an automated material handling system capable of inspecting materials or assemblies for defects.

The experimental setup entails employing a Fanuc LR Mate 200 iD/4S robot with an R-30iB controller equipped to run vision tools using iRVision software. To augment its capabilities, a specially designed two-jaw gripper, manufactured using additive manufacturing techniques, has been incorporated into the system. However, since the iRVision software available in the research lab lacks AI capabilities, a Logitech C920x HD Pro Webcam with a 1080p resolution is affixed to the robot's wrist using a 3D printed holder. This camera is connected to a laptop running MATLAB for tasks such as image acquisition, processing, and deep learning.

The robot facilitates automatic handling of parts and ensures precise positioning for image data acquisition, minimizing variations in image location and orientation and enhancing image registration efficiency. The system employs tailored robot programming techniques for the FANUC LR Mate 200 iD arm through the iPendent interface. As a proof of concept, images of a toy car assembly are captured using the mounted camera and then subjected to thorough analysis using computer vision tools for preprocessing. The dataset comprises images captured from five views of the car: Top, Front, Back, Left-Side View, and Right-Side View. Each view includes 1000 images of both good and defective cars, which are, later, synthetically augmented using deep learning tools to introduce rotational and positional offsets within reasonable ranges. The dataset is divided into training (70%), validation (10%), and testing (20%) sets.

Transfer learning techniques are employed to minimize training time using pre-trained deep learning models such as Google Net, VGG16, and VGG19. The model with the highest accuracy is selected based on preliminary results and further optimized for hyperparameters. Once trained, the model is deployed using Simulink tools for real-time testing. A Simulink model comprising image blocks, deep learning blocks, and communication blocks facilitates image acquisition and classification of non-defective and defective parts. Inspection decisions are efficiently relayed to the robot using Simulink communication tools, guiding it in the sorting process.

The integrated system forms a closed-loop environment where the robot, camera, and MATLAB tools collaborate seamlessly. By combining computer vision, AI, and robotics, the system aims to expedite assembly inspection, accurately identifying both good and defective assemblies, thereby enhancing inspection speed and product quality while reducing production time and increasing output. Anticipated benefits include lower rework and scrap rates, leading to cost savings and prevention of expensive recalls. The project's outcomes are expected to provide valuable insights into the adaptability of automated inspection systems in diverse manufacturing environments, contributing to advancements in Industry 4.0 technologies and processes.

Presenting Author: Vedang Dilipkumar Chauhan Western New England University

Presenting Author Biography: Dr. Vedang Chauhan serves as an Associate Professor in the Department of Mechanical Engineering at Western New England University since 2018. He is a highly engaged researcher with a focus on the design and development of machine vision, robotics, and machine learning systems applied to industrial automation. Driven by his passion for technology, he holds certifications in deep learning and machine learning from Coursera. Dr. Chauhan has actively contributed to research projects funded by the Department of Energy (DoE) and Toyota North America, applying machine vision and machine learning techniques to these initiatives. His expertise extends to robotic programming, demonstrated by his Certified Robot Programmer status. He earned his Ph.D. in Mechanical Engineering, specializing in machine vision and mechatronics, from Queen’s University, Canada, in 2016. Previously, he worked as a robotics engineer at Bluewrist Robotics Company in Toronto for a year. Dr. Chauhan holds Master's and Bachelor's degrees in Mechanical Engineering and Mechatronics Engineering, respectively, from Sardar Patel University, India, in 2007 and 2003. His dedication to advancing knowledge is evident through his publications in robotics and machine vision journals and conferences. Additionally, he actively contributes to the academic community as a reviewer for journals in the fields of robotics, mechatronics, and machine learning. Dr. Chauhan also serves as the director of the Robotics and Mechatronics Lab at WNE, reflecting his leadership and expertise in these domains.

Authors:

Vedang Dilipkumar Chauhan Western New England University
Grant Maziarz Western New England University
Simone Cardoso Dos Santos Western New England University