Date of Award

4-2025

Degree Name

Doctor of Philosophy

Department

Industrial and Entrepreneurial Engineering and Engineering Management

First Advisor

Lee Wells, Ph.D.

Second Advisor

James Burns, Ph.D.

Third Advisor

Ilgin Acar, Ph.D.

Fourth Advisor

Robert Tuttle, Ph.D.

Keywords

High density dataset, machine learning, neural networks, surface roughness, variogram

Abstract

In the transformative landscape of Industry 4.0, advanced quality control in manufacturing demands innovative methods for processing and analyzing datasets. This dissertation presents a comprehensive study on leveraging high volume data and machine learning to enhance decision making in the foundry industry. The research explores three crucial challenges in the foundry industry: predicting surface roughness, monitoring refractory coating thicknesses, and comparing feature extraction techniques for surface quality classification. The findings underscore the transformative potential of integrating advanced feature engineering techniques and machine learning into manufacturing quality control.

The first of the three challenges introduces a framework that integrates digital images and point clouds to predict surface roughness using only digital images. By applying Multilinear Principal Component Analysis (MPCA) the cross-correlation between digital images and point clouds is learned off-line to train a neural network model that predicts surface roughness online. This online model only requires the use of digital images, allowing an efficient performance in prediction as compared to using both digital images and point cloud for prediction.

Classification of refractory coating thickness helps in monitoring them, which is important to enhance cast surface finish and reduce defects that occur at the mold-metal interface. In the second research thrust, the classification of refractory coating thicknesses is achieved using time-series data from a chemically bonded sand tester. Feature extraction techniques are employed to extract important features, enabling precise, coating thickness classification.

The final research thrust evaluates the surface quality classification accuracy of three different feature types: MPCA, variogram analysis, and Sa surface roughness values—using point cloud data. MPCA emerges as the most effective method, preserving critical spatial features and achieving superior classification accuracy. Although, MPCA possess the drawback of lower physical interpretability when compared with the other two feature types.

This dissertation offers solutions for optimizing the prediction surface roughness, monitoring refractory coating thickness, and identifying the best feature extraction technique for surface quality classification, all of which helps advance the foundry industry's adaptability to modern manufacturing demands of Industry 4.0.

Access Setting

Dissertation-Open Access

Download

Included in

Metallurgy Commons

Share

COinS