Enhancing operational performance in textile manufacturing: impact of deep learning-based defect detection

Quality performance in manufacturing has a direct influence on efficiency, generated waste, and costs. In collaboration with a textile manufacturer as a case study, this paper develops an automated defect detection system for a weaving process and evaluates its impact on operational performance. The system identifies defects immediately at their onset and prevents their propagation to subsequent fabric and production stages. A deep learning image classification model is developed, with six well-established network architectures being compared, leveraging a non-invasive image acquisition method that averts machinery disturbances for data collection. Based on the best-performing model, key indicators of operational performance are estimated using Markov Chain modelling, addressing a gap in linking model performance to operational impacts. Notable operational gains are demonstrated, namely a cost reduction of 1.3% and over 90% of waste reduction. A sensitivity analysis guides the definition of the image acquisition frame rate to minimise false alarms and shows that different operational indicators are impacted differently by different predictive performance metrics, affecting model selection. This research not only underscores the potential of integrating deep learning into textile production but also guarantees the effective communication of its impact to industry stakeholders, thus offering valuable practical insights to enhance operational performance.