[Paper Review] Automated Workers’ Ergonomic Risk Assessment in Manual Material Handling Using sEMG Wearable Sensors and Machine Learning

Srimantha E. Mudiyanselage|arXiv (Cornell University)|Oct 19, 2021

Ergonomics and Musculoskeletal Disorders27 references133 citations

TL;DR

This study proposes an automated ergonomic risk assessment system for manual material handling using surface electromomyography (sEMG) sensors and machine learning to classify risk levels based on the NIOSH Lifting Equation. The Decision Tree model achieved up to 99.35% accuracy in classifying nominal, increased, and high-risk lifting tasks from sEMG signals collected from the thoracic muscles.

ABSTRACT

Manual material handling tasks have the potential to be highly unsafe from an ergonomic viewpoint. Safety inspections to monitor body postures can help mitigate ergonomic risks of material handling. However, the real effect of awkward muscle movements, strains, and excessive forces that may result in an injury may not be identified by external cues. This paper evaluates the ability of surface electromyogram (EMG)-based systems together with machine learning algorithms to automatically detect body movements that may harm muscles in material handling. The analysis utilized a lifting equation developed by the U.S. National Institute for Occupational Safety and Health (NIOSH). This equation determines a Recommended Weight Limit, which suggests the maximum acceptable weight that a healthy worker can lift and carry, as well as a Lifting Index value to assess the risk extent. Four different machine learning models, namely Decision Tree, Support Vector Machine, K-Nearest Neighbor, and Random Forest are developed to classify the risk assessments calculated based on the NIOSH lifting equation. The sensitivity of the models to various parameters is also evaluated to find the best performance using each algorithm. Results indicate that Decision Tree models have the potential to predict the risk level with close to 99.35% accuracy.

Motivation & Objective

• To develop an automated system for real-time ergonomic risk assessment during manual material handling tasks.
• To evaluate whether sEMG signals from the thoracic muscles can reliably reflect ergonomic risk levels as defined by the NIOSH Lifting Equation.
• To compare the performance of four machine learning models—Decision Tree, SVM, K-Nearest Neighbor, and Random Forest—in classifying risk levels from sEMG data.
• To identify the optimal time-segmentation window size for sEMG signal processing to maximize classification accuracy.
• To provide a data-driven, non-invasive method for detecting hazardous postures and muscle strain before injury occurs.

Proposed method

• sEMG sensors were placed on the thoracic (T1-T3) muscles of a single participant to record electrical activity during 54 controlled lifting trials.
• Lifting tasks were performed with six different weights across three risk levels (nominal, increased, high) defined by the NIOSH Lifting Equation.
• sEMG signals were segmented into 1-second, 0.5-second, and 0.25-second time windows for analysis.
• Four machine learning models—Decision Tree, Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest—were trained to classify risk levels based on sEMG amplitude features.
• The NIOSH Lifting Index (LI) was used as the ground truth for risk classification, with LI < 1.0 indicating nominal risk, 1.0 ≤ LI < 3.0 indicating increased risk, and LI ≥ 3.0 indicating high risk.
• Model performance was evaluated using classification accuracy, confusion matrices, and sensitivity analysis across different window sizes.

Experimental results

Research questions

• RQ1 Can sEMG signals from the thoracic muscles reliably indicate ergonomic risk levels during manual lifting tasks as defined by the NIOSH Lifting Equation?
• RQ2 Which machine learning model—Decision Tree, SVM, KNN, or Random Forest—performs best in classifying risk levels from sEMG data?
• RQ3 How does the choice of time-segmentation window (1s, 0.5s, 0.25s) affect the classification accuracy of the machine learning models?
• RQ4 What is the misclassification pattern across risk classes, particularly for the 'Increased Risk' category?
• RQ5 Is overfitting a concern in the KNN model, especially with K=1, and how does increasing K affect performance?

Key findings

• The Decision Tree model achieved the highest classification accuracy of 99.35% using a 0.5-second time-segmentation window.
• The 0.25-second segmentation yielded a 99.05% accuracy with the Random Forest model, the second-highest performance among all models.
• The 'Increased Risk' class had the highest misclassification rate across all models and segmentation sizes, indicating greater difficulty in distinguishing from other classes.
• KNN models showed the lowest overall accuracy, with a maximum of 96.21% at 0.25-second segmentation, and K=5 did not outperform K=1, suggesting no significant overfitting.
• A clear correlation was observed between sEMG amplitude (both time and frequency domain) and the NIOSH Lifting Index, with higher muscle activity corresponding to higher risk levels.
• The confusion matrix confirmed that the Decision Tree model correctly classified 137 nominal risk, 77 increased risk, and 204 high-risk instances in the 0.25-second segmentation, indicating strong class separation.