Investigation of Machine Vision-Based Quality Inspection Methods for Detecting Surface Defects in Ceramic Tiles Produced in Kogi State

Ceramic tile manufacturing in Kogi State, North Central Nigeria, is a growing industry supplying the construction sector, yet quality inspection for surface defects including cracks, chips, colour variation, and pattern misalignment is conducted manually by trained inspectors, introducing subjectivity, fatigue-related inconsistency, and throughput limitations. This study investigates machine vision-based automated quality inspection methods for detecting surface defects on ceramic tiles at production line speeds at a tile manufacturing facility in Obajana. A multi-camera imaging system with structured LED illumination is designed and installed to capture top-surface images of tiles moving on a production conveyor at 0.8 metres per second. Three machine vision algorithms, classical morphological image processing, traditional machine learning with handcrafted feature extraction using support vector machines, and a convolutional neural network trained on a labelled tile defect dataset, are compared for defect detection accuracy and computational throughput. The CNN approach, implementing a ResNet-18 architecture fine-tuned on 12,400 labelled tile images, achieves the highest defect detection sensitivity of 97.3 percent with a false positive rate of 2.8 percent, outperforming the SVM at 89.4 percent sensitivity and morphological methods at 81.2 percent. CNN inference time of 38 milliseconds per tile satisfies the real-time processing requirement at the conveyor speed tested. The study analyses the cost of the machine vision system relative to human inspector employment costs, projecting payback within 11 months. Keywords: machine vision, quality inspection, CNN, ceramic tile defects, Kogi State