Awais, Muhammad

Comparison of Different Machine Learning Models for Quality Control in Biscuit Manufacturing Industry of Pakistan / Muhammad Awais - 113p. Soft Copy 30cm

Ensuring consistent product quality is a critical challenge in the biscuit
manufacturing industry, particularly in developing economies like Pakistan, where
manual inspection remains the norm. This approach, however, is prone to human error,
fatigue, and inconsistency, leading to variable quality control outcomes. This research
proposes an intelligent, machine learning–based automated quality inspection system
tailored for real-world biscuit production lines. A comprehensive dataset of 40,000
high-resolution images was curated, representing four popular biscuit varieties Rio,
Sooper, Candy, and Marie captured under actual factory lighting conditions. The
dataset includes both defective and non-defective samples, covering a wide range of
defect types such as breakage, charring, deformation, and missing pieces. To ensure
high-quality annotations with scalable efficiency, a semi-automated iterative annotation
framework was developed, combining initial manual labeling with model-assisted
annotation and human-in-the-loop refinement across multiple cycles. Multiple state-ofthe-art machine learning models were implemented, fine-tuned, and rigorously
evaluated, including YOLOv8, YOLOv11, YOLOv12, Faster R-CNN (via Detectron2),
and Vision Transformer (ViT-B/16). Models were assessed using key performance
metrics such as mAP@0.5:0.95, precision, recall, F1-score, inference speed (FPS), and
computational efficiency. The results demonstrate that modern deep learning models,
particularly YOLOv12 and Vision Transformers, achieve high detection accuracy
(mAP > 90%) while maintaining feasibility for real-time deployment when optimized.
This study provides a comparative analysis of accuracy-speed trade-offs, offering
actionable insights for manufacturers seeking cost-effective, scalable solutions. The
research concludes with practical recommendations for integrating AI-driven
inspection systems into existing production infrastructure in resource-constrained
environments, balancing performance, hardware requirements, and long-term
maintainability.


MS Design and Manufacturing Engineering

670