Safer and Efficient Factory by Predicting Worker Trajectories Using Spatio-Temporal Graph Attention Networks

Abstract Occupational accidents in manufacturing industries pose a significant risk, necessitating advanced strategies to ensure worker safety and enhance operational productivity. The unpredictable nature of worker movements, influenced by varied tasks such as material transportation, machine operation, and collaborative efforts, highlights the critical need for effective trajectory prediction mechanisms. This paper introduces an innovative approach utilizing Spatio-Temporal Graph Attention Networks (STGAT) and Spatio-Temporal Graph Convolutional Neural Networks (STGCNN) to predict worker trajectories with high accuracy and to analyze worker interactions within the manufacturing environment. Our methodology employs qualitative evaluation techniques to reveal intricate worker dynamics during assembly line processes, offering new perspectives on spatial-temporal interplays in a factory setting. By applying this method to movement data from a detailed case study involving six workers on a tribike assembly line, we demonstrate the effectiveness of our proposed algorithm in real-world scenarios. The utilization of advanced Graph Neural Network technologies allows for the precise modeling of complex spatial-temporal relationships, enabling the accurate prediction of worker paths. This research contributes significantly to the fields of occupational safety and industrial efficiency by providing a comprehensive framework for anticipating worker movements and understanding their interactions in intricate manufacturing landscapes. Moreover, it addresses existing challenges in trajectory prediction and outlines potential directions for future research, aiming to broaden the application of predictive analytics in enhancing safety protocols and operational strategies in the manufacturing sector.