Comparative Analysis of Machine Learning Algorithms and IoT Integration for Fault Detection in Textile Manufacturing

Abstract

This research presents a comparative analysis of advanced fault detection methodologies in textile manufacturing, specifically focusing on clothes fabrication processes. The study integrates an IoT-enabled smart sensor network for real-time process monitoring and quality control using YOLOv8 as the primary detection model, with YOLOv5 serving as a comparative benchmark. Data collection was conducted across major textile manufacturing hubs in India—Tiruppur, Coimbatore, Surat, Ludhiana, and Bhilwara—encompassing 50 textile production units with continuous sensor data recorded over a six-month period. The multi-modal data acquisition system incorporated optical sensors, thermal imaging, vibration monitoring, and RFID tracking to capture various fabric defect types including weaving inconsistencies, colour variations, and structural anomalies. The research introduces a novel Hybrid IoT-AI Framework (HIAF) that combines real-time sensor networks, edge computing, and deep learning-based predictive analytics to enhance textile manufacturing processes. Performance metrics for each detection algorithm were evaluated based on mean Average Precision (mAP), inference time, and computational resource requirements in edge computing environments. Results demonstrate that the YOLOv8 implementation achieved superior defect detection accuracy (93.7%) compared to YOLOv5 (89.2%), while maintaining acceptable inference speeds for real-time industrial deployment. The IoT framework facilitated seamless integration with manufacturing execution systems, enabling automated parameter adjustments that reduced false detection rates by 27.3% compared to conventional inspection methods. Additionally, a digital twin model was implemented to simulate manufacturing conditions, facilitating predictive maintenance and reducing fault detection errors by 32% through virtual environmental testing.