Defect Identification and Classification of Cellular Glass
Open Access
- Author:
- Burlovic, Benjamin
- Area of Honors:
- Mechanical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Ryan L Harne, Thesis Supervisor
Margaret Louise Byron, Thesis Honors Advisor - Keywords:
- Non Destructive Testing
Neural Networks
Image Classification
Machine Learning - Abstract:
- Cellular glasses are heavily employed insulation materials due to their high strength-to-weight ratio, low cost in mass production, closed-cell microstructure, and low conductivity. The high strength-to-weight ratio stems from the porous microstructure of cellular glass. The porosity of cellular glass is characterized by millions of thin glass-walled cells filled with pressurized gasses (CO2 and O2). When compressed, these cells absorb high amounts of pressure while maintaining a low density and subsequent weight. The size and distribution of the cells drive the compressive strength of a given cellular glass sample and can range from 1.5 – 5 MPa. The high variation in strength leads to the failure of products when operating under normal conditions. A major challenge faced by manufacturers of cellular glass is the lack of a non-destructive testing (NDT) method. They can't analyze an entire sample’s microstructure without cutting into the block and visually inspecting the cells. This research seeks to undertake a series of exhaustive computational designs and experimentation to investigate how the true material properties of a given cellular glass sample can be identified. Specifically, the use of a neural network to perform defect detection using computerized tomography scans of cellular glass is explored. A logistical regression neural network for image classification was designed and written in the MATLAB platform and tested for its accuracy and efficiency. The results showed the neural network was highly accurate, with a peak rating of 100%, and efficient with a variety of sample data. The results of this research will guide future work to create an independent operating NDT system to characterize cellular glass products during the manufacturing of products and ultimately limit failures in the field.