Graduation Year

2024

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Mechanical Engineering

Major Professor

Nancy Diaz-Elsayed, Ph.D.

Committee Member

Ashok Kumar, Ph.D.

Committee Member

Ankit Shah, Ph.D.

Committee Member

Susana Lai-Yuen, Ph.D.

Committee Member

Robert Hooker, Ph.D.

Keywords

Additive Manufacturing, CNC Machining, Machine Learning, Surface Roughness, Sustainable Manufacturing

Abstract

Sustainability in manufacturing is emerging to be a key factor to minimize waste and enhance resource efficiency while maintaining product quality. Sustainability and quality are intertwined and have an impact on each other wherein sustainable practices for surface quality assessment play a pivotal role in waste management and process optimization. Advanced technologies such as machine learning have gained prominence in manufacturing and the availability and ubiquity of diverse data streams allows for the enhancement of manufacturing systems across metrics critical to the sectors in which they are employed such as performance and quality. However, many challenges remain in the effective adoption of smart data-driven technologies including ease of use, expertise required, infrastructural costs, and post-production quality assessment. With the aim of mitigating these challenges, the goal of this research is to enhance the quality and sustainability of manufacturing systems by leveraging AI technologies for additive and subtractive processes. Based on the goal, there are three key objectives of this research, with specific outcomes:

  • Objective 1: The first research objective is to automate the quality detection for resource-efficient additive manufacturing, focusing on optimizing process parameters to achieve high-quality parts in terms of surface roughness and dimensional accuracy while minimizing resource usage. The edge and dimensional accuracy is assessed using images acquired from readily available devices, while the surface roughness of the printed parts is evaluated from the printing parameters using supervised machine learning algorithms. The results indicate that while thresholding techniques work well for edge and dimension quality detection and fine tree algorithm works well for surface roughness assessment. The prediction of the surface roughness and enhanced process parameters using regression can support the environmental sustainability of production by reducing the time and resources needed to assess the quality of the part, as well as the rate of part failure and its subsequent energy consumption.
  • Objective 2: The second objective is to investigate the application of Vision Transformers (VIT) and the role of transfer learning for advanced quality inspection of CNC (computer numerical control)-machined brass components. The surface roughness of milled brass parts will be assessed from images acquired by readily available devices. Different model architectures are evaluated along with a combination of data augmentation and transfer learning techniques. It was found that an amalgamation of optimal hyperparameters and a fine-tuning technique can lead to high accuracy even with a limited dataset.
  • Objective 3: The third research objective is to investigate the use of predictive analytics to enhance environmental assessments in additive manufacturing to drive sustainable operations. Two frameworks are proposed wherein the first framework can predict the carbon dioxide and VOC emissions of the process, and the second framework can predict the carbon dioxide emissions from the G-Code. While supervised learning models were used for the first framework, boosting algorithms were used for the latter. Optimal results were achieved for both scenarios, wherein gaussian process regression and extreme gradient boosting showed the best performance.

Download

Share

COinS