|
Original research AI-DRIVEN DEFECT DETECTION AND PROCESS OPTIMISATION IN BREWERY BOTTLING USING COMPUTER VISION AND SIX SIGMAPages 1-12
Abstract
The global brewing industry faces persistent quality assurance challenges, particularly at the bottling stage, where defects such as glass cracks, labelling irregularities, underfilled containers, and improper seals compromise product integrity, trigger regulatory sanctions, and erode consumer confidence. Conventional manual inspection remains susceptible to fatigue-induced error, operator subjectivity, and throughput limitations, while legacy rule-based machine vision systems lack the adaptive intelligence needed to evolve with changing bottle designs and subtle defect patterns. This study presents the design, implementation, and experimental evaluation of an integrated AI-driven defect detection and process optimisation framework for brewery bottling operations, combining deep Convolutional Neural Network (CNN) architectures with the Six Sigma DMAIC methodology. Four CNN models, a Custom CNN, VGG16, ResNet50, and MobileNetV2, were trained on an annotated dataset of broken and intact bottle images using structured hyperparameter configurations. Model performance was assessed using classification accuracy, precision, recall, and F1 score, alongside Six Sigma Defects Per Million Opportunities (DPMO) and sigma level metrics to quantify inspection reliability. The findings confirm that integrating transfer-learned CNN architectures with Six Sigma process control constitutes a technically viable, cost-effective quality assurance paradigm for brewery operators in Nigeria and comparable resource-constrained manufacturing environments.
Keywords:
Convolutional Neural Networks, Defect Detection, Six Sigma, DMAIC, Computer Vision, Brewery Bottling, Edge Computing, Transfer Learning.
|
Egwuagu Onyekachi Marcel, 
PDF 