Abstract:

The technological changes that the world is going through and the transformation that the manufacturing industry requires due to the fourth industrial revolution (I4.0), are putting pressure on universities to modify their study programs to ensure the knowledge and development of the skills demanded by this I4.0 concept. Additionally, a series of digital technologies are transforming industrial processes to become more reliable, predictable, and robust.

This paper presents the design and implementation of a vision system with Artificial Intelligence (AI) for the classification of defects during the assembly of cabin components in a simulated work environment (SWE).

The SWE was used to define the different stations that allow the assembly/disassembly of the different parts of an automotive cabin model. The cabins are transported through various workstations due to the automatic activation of two conveyor belts. The manufacturing cell has eight stations, six of which are for the assembly/disassembly of the cabin parts and two for quality inspection.

Students participate collaboratively in the process of assembly/disassembly of the parts in the SWE and learn about the problems in the quality inspection station for the identification/classification of defects in the assembled parts.

For the implementation of the vision system, an economical option was sought for the purchase of the equipment (\$1950.00 US dollars) and the programming was based on Matlab for the AI algorithms and the use of LabView for image acquisition.

The vision system allows classifying and identifying defects in the assembly process of the cabin parts. Starting from the images acquired of the cabins by the vision system, it is necessary to investigate, design and select the best AI algorithm that efficiently allows the classification/detection of defects in the assembly of the cabins.

A dataset of images was obtained with the vision system and Labview. The dataset consisted of 230 RGB images that represented 10 different classes (23 images per class). The first class defines the complete assembly cabin, and the other 9 represent a cabin with a missing component.

Two deep learning models were developed: an Artificial Neural Network (ANN) model and a Convolutional Neural Network (CNN) model. A programming environment was designed and implemented in MatLab, where students can propose different architectures, modify important parameters in the network learning process and use different training algorithms to obtain a neural network model with the highest performance. With the proposed platform for designing and evaluating the different architectures, the following performances were achieved: for the ANN model, a performance of 60.7% was obtained, and for the CNN model, a performance of 96% was obtained.

On the other hand, this project applied Active Learning (AL) and Collaborative Learning (CL) to guarantee a theoretical and practical learning experience with great impact on the use of these technologies and their applications, favoring the curricula of Industrial Engineering and Mechatronics students in the concepts of the fourth Industrial Revolution. Different instruments evaluated the impact of this academic proposal. For example, surveys showed that 91% of the students had an active learning experience, and 85.5% agreed that they had received a very useful benefit for their professional life.

Keywords:

Vision Systems, Active Learning, Neural Networks, Collaborative Learning.