Finite Element-Integrated Deep Learning Model for Early Risk Detection of Work-Related Musculoskeletal Disorders

Abstract

In manufacturing textile industries, Work-Related Musculoskeletal Disorders (WMSDs) are considered as the one of the occupational health issues because the workers usually use repetitive upper-limb movements such as rotation, abduction, and flexion. The traditional machine learning (ML)-based WMSD prediction models and ergonomic assessments are depend on the historical posture data and it does not integrate the biomechanical principles or predicts the upcoming motion risks. So, it has limited accuracy and delayed involvement. To overcome these limitations, the novel Finite Element with Puma Optimized Deep Network was proposed for Proactive WMSD Prediction (FEPO-WMSDNet). Initially, the proposed system utilized Finite Element–Integrated Neural Network (FEINN) for biomechanical motion forecasting, which is then optimized by a bio-inspired Puma Optimizer (PO) for adaptive hyperparameter tuning of FEINN Network. Then, Convolutional Karcı Fractional Artificial Neural Network (CK-FANN) were utilized for intelligent risk classification based on RULA and ISO 11226 standards. By this, the proposed FEPO-WMSDNet method attains Average Matthews Correlation Coefficient (MCC) as 95.37%, an Average Accuracy as 96.5%, an Average F1-Score as 96.1%, an Average Precision as 95.2%, an Average Recall as 96.9%, and an Average Specificity as 98.2% for WMSD risk classification. Likewise, the proposed FEPO-WMSDNet method attains R² (Coefficient of Determination) as 0.981, Average Mean Absolute Percentage Error (MAPE) as 3.60%, an Average Root Mean Squared Error (RMSE) as 0.063°, and an Average Mean Absolute Error (MAE) as 0.043° for the Ahead-of-Time (AoT) motion angle forecasting task. So, the proposed FEPO-WMSDNet method is an effective tool for averting work-related musculoskeletal disorders.