Machine Learning for Pore Detection and Density Determination of Sintered Ceramics

Poon, Justin Hao Yin (2025) Machine Learning for Pore Detection and Density Determination of Sintered Ceramics. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text JUSTIN POON HAO YIN_FULL TEXT.pdf Restricted to Registered users only Download (18MB)

Abstract

Hydroxyapatite is a calcium phosphate-based ceramic that has applications in the biomedical field as dental and prosthetic implants due to its biocompatibility and similarity to the human bone. Recently, research on hydroxyapatite synthesis using biogenic materials, such as chicken bones and eggshells are growing. As sintering is a crucial step in the synthesis of the said material, the sintering parameters influence the porosity and mechanical capabilities of the material. Conventional methods of porosity analysis such as X-ray Computed Tomography and intrusion porosimetry are high-costing and destructive to the sample. Hence, a machine learning model based approach, particularly with the YOLOv8-medium variant from the YOLO family of single stage detector models, is implemented for efficient and effective porosity analysis. In this research, Scanning Electron Microscopy (SEM) images of locally synthesized hydroxyapatite derived from chicken bones, eggshells, fish bones and lobster crab shell, taken using Mini SEM and Field Emission SEM, are obtained as the dataset. Data preparation is performed using the Roboflow computer vision platform, whereas the training and evaluation of the YOLOv8-medium object detection model is performed using the Google Colab platform. By tuning the hyperparameters for the model, including the training batch size and epochs, cleaning data annotation, modifying data augmentation, modifying image dimensions as well as inference confidence thresholds, the final YOLOv8-medium model is able to achieve a performance metrics of 0.5984 accuracy, precision of 0.7422, recall of 0.7554 and F1-score of 0.7487 on unseen data. The model is used to extract visual as well as quantitative porosity information of SEM images of hydroxyapatite.