A Smart Logging System with High Accuracy for Industrial Welding Activity Monitoring Based on a High Dimensional Feature Classification System

This study introduces a high-precision intelligent system that utilizes a Support Vector Machine (SVM) algorithm to monitor welding activities in an industrial environment. The system leverages embedded sensors in smartphones, specifically accelerometers, gyroscopes, and magnetometers, to facilitate the Human Activity Recognition (HAR) task. The system is designed to classify six different combinations of welding activities and postures: standing-active, sitting-active, lyingactive, standing-inactive, sitting-inactive, and lying-inactive. Here, “active” indicates engagement in a welding task, while “inactive” indicates a pause in the activity. The developed system supports the acquisition of structured and labeled activity data, suitable for training HAR models, with the primary goal of improving occupational safety and operational efficiency, especially in high-risk environments such as shipyards. Experimental evaluations show that the SVM-based classification model achieves an accuracy of approximately <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">9 8. 6 1 %</tex> in identifying welder activity categories, including standing-active, standing-inactive, sitting-active, and sitting-inactive states. In 20 real-time testing sessions, the average error rate was approximately 3.5 %, with a classification latency of approximately 1 millisecond per event. These results underscore the system's robust performance and practical utility for industrial applications.