Utilizing Deep Learning and Computer Vision to Detect Defects in a 3D Printed Product in a Manufacturing Environment
Open Access
- Author:
- Farkas, Alyson
- Area of Honors:
- Letters, Arts, and Sciences (Abington)
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Robert Louis Avanzato, Thesis Supervisor
Dave Ruth, Thesis Honors Advisor - Keywords:
- transfer learning
deep learning
3D printing
robot arm
manufacturing defects
blemish detection
MATLAB
neural network
quality assurance
robotics
engineering
computer vision
artificial intelligence - Abstract:
- The objective of this project is to implement deep learning and computer vison practices to detect defects within 3D printed plastic parts in a manufacturing environment using a conveyer belt and a robot arm. Deep learning is an important technological development within artificial intelligence, and is successful in different application areas including: manufacturing, medical analysis, and self-driving cars. The study focuses on using transfer learning to identify and classify defects in 3D printed food trays that are designed to hold food in a car. MATLAB is used to develop and train a deep learning neural network based on a pretrained network. Images of trays classified as good and bad quality were used to train and test the network. Several studies were completed by changing training parameters to determine optimal settings for repeatable and accurate classifications of manufacturing defects. The trained network was able to classify good and bad quality trays with a success rate ranging from 90%-100%. A camera positioned above the conveyer belt was used to classify trays based on the trained deep learning network. A robot arm was integrated to sort good and bad quality trays from the conveyer belt to designated containers. The robot integration and accurate classifications demonstrated the potential for increased efficiency in quality assurance and manufacturing procedures. The study suggests that deep learning with transfer learning is a viable option to test 3D printed parts for defects in a manufacturing environment.