Implementation of a Convolutional Neural Network (CNN) for Ergonomic Analysis in Garment Manufacturing Operations

Abstract

The garment manufacturing sector operates under high production demands, where repetitive movements and sustained awkward postures increase workers’ exposure to musculoskeletal disorders (MSDs). Conventional ergonomic assessment methods such as Rapid Upper Limb Assessment (RULA) and Rapid Entire Body Assessment (REBA) rely on direct observation and extended evaluation periods, requiring trained personnel and potentially introducing subjectivity. This study proposes the implementation of a Convolutional Neural Network (CNN) to automate ergonomic risk assessment in garment operations. A dataset of 1,000 images capturing arm and forearm postures during sewing tasks was compiled and labeled according to the RULA method. Image preprocessing and annotation were conducted using Kinovea and Roboflow. The CNN model integrates a keypoint detection algorithm based on the COCO-Pose framework for posture estimation and classification. The proposed model achieved a mean Average Precision (mAP) of 96.40%, recall of 93.50%, and precision of 92.30%, demonstrating high reliability in ergonomic posture classification. The results indicate that the automated approach enhances evaluation accuracy, reduces observer subjectivity, and provides a scalable tool for improving workplace ergonomics and preventing MSDs in garment manufacturing environments.