Date of Publication


Document Type

Master's Thesis

Degree Name

Master of Science in Mechanical Engineering

Subject Categories

Mechanical Engineering


Gokongwei College of Engineering


Mechanical Engineering

Thesis Advisor

Laurence A. Gan Lim

Defense Panel Chair

Jeremias A. Gonzaga

Defense Panel Member

Conrad Allan Jay R. Pantua
Robert Kerwin D. Billones


With the rise of industrial automation, more and more methods are required to enable more efficient production lines. A common component of the production line often relying on subjective input is quality control. This is especially apparent in plastics manufacturing where defects in the workpieces are inspected manually based on various factors. This study aims to develop a more objective and efficient solution to quality control through detecting cloudy spot defects on PET preforms through machine learning. The detection aims to use techniques from image processing, as well as neural networks, to allow for real time detection of cloudy spot defects along transparent PET preforms commonly manufactured during the production process. The system is trained and tested using preforms already manually sorted as defective or good, with its accuracy observed through the resulting confusion matrix. Adjustments were then made based on the results from the confusion matrix and tested again for accuracy. A custom set up was created to collect image data under near consistent conditions to be separated between training and testing data. This data was then sent to a program developed on MATLAB for this application with the steps of preprocessing the data, segregating between training and testing, testing using a convolutional neural network and the computation of results through a confusion matrix. The resulting program was able to classify defected preforms but was unable to do so on a region-based method of image processing. Results from the study showed a need to investigate both a method to use region-based neural networks in identifying defects, and a method of determining clean preforms which do not have any defects.

Abstract Format







Machine learning; Neural networks (Computer science); Polyethylene terephthalate—Defects

Upload Full Text


Embargo Period