Non-Visible Defect Detection in Glass using Infrared Thermography and Artificial Neural Networks
Conference Paper
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
Defects in glass are a major problem in the manufacturing industry as these defects cause for recall and rejection of materials and products. Defects such as bubbles need to be detected early on in the manufacturing process in order to avoid loss of revenue for a company. Many of these defects are not immediately visible making them difficult to detect. In the past, methods have included embedding artificial defects within two layers of a different material, and then optical and visual methods have been used in order to classify and locate these defects. In this study an infrared camera was used in order to take images of non-visible artificial defect samples and understand the use of infrared technology for defect detection. This could then be used to locate and classify defects based on diameter and depth. Artificial Neural Networks were used to predict the sample classifications based on temperature variation, cooling rate, and the time at which the image was taken. Artificial Neural Networks showed to be a good prediction method for depth classification as it reached a 76% accuracy rate, whereas this method was not as effective for diameter classification. Results from this study produced a mathematical model for this range of data and showed that temperature variation amongst different depths of samples was higher compared to the temperature variation of diameter size. Artificial Neural Networks can therefore be used to classify a sample as defective or non-defective, and then the mathematical model presented can be used to estimate the depth of the defect. 2013 SPIE.
