Electromyography (EMG) in Ergonomics: EMG Armband to Assess Injury Risks
3 min read
Medium
What do we mean when we describe someone as a “hard worker”? This term often brings the image of feeling drained after a long day, particularly in physically demanding jobs. However, hard work shouldn’t require us to push our bodies to their limits. In the UK, musculoskeletal disorders (MSDs) were responsible for 30% of all work-related health issues in 2019/2020, leading to 8.9 million lost working days. Workplace injuries harm employees’ physical health and impose significant economic costs on businesses. Last year, these costs amounted to £16.2 billion in the UK. The issue of work-related MSDs is also prevalent across Europe. In 2015, over a third (37%) of EU workers reported that their work adversely affected their health, with MSDs making up 70% of work-related health problems in countries such as the Czech Republic, Cyprus, Poland, and Finland.
In ergonomics, there has never been such a timely need for proper and efficient methods for evaluation of lifting tasks and hand loads. In one of the latest studies which appeared in Applied Ergonomics, the researchers explore the use of the MindRove Armband sEMG sensors to achieve this; what they offer is rather thought-provoking. The prominent scholars, S. Taori and S. Lim, have developed a research on different machine learning algorithms helping to identify lifting tasks and categorise hand loads with the view of increasing ergonomics at working environments with the help of such technologies as presented in the paper. This paper gives a brief summary on how the study was carried out, what was observed and established, as well as the importance of the study.
Study Overview
The subjects of the research were nine young men and women from the university. Eight electrodes were placed on their forearm using the MindRove EMG Armband, to be able to capture muscle activity during different types of lifting tasks and then analyze such data with the help of Machine Learning algorithms.
sEMG Sensor Setup
The MindRove Armband was placed on the right forearm of the participants. Muscle activity data was captured at 500Hz through the device. Data collection herein included several lifting trials, which were conducted using different types of hand loads, different heights of tables and different heights of shelves.
Lifting Tasks
This study also involved three types of lifting; symmetrical, asymmetrical and free-dynamic. Each of the scenarios demanded a certain degree of complexity, and the participants were expected to lift several loads and their actions were recorded at the same time.
Machine Learning (ML) Models
The study further proposed and validated several ML methods for identifying lifting tasks and categorizing hand loads. Such models include Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) that made use of the time-domain and frequency-domain features extracted from sEMG signals.
Key Findings
Lifting Detections
Some of the preliminary results from analyzing the moving ML models provided promising results of identifying the initiation and conclusion stages of the lifting process. As expected, the classifiers obtained higher accuracy values when time-domain or time-frequency is used as the features and out of all the classifiers, Random Forest and Support Vector Machine provided the best results. The accuracy of the start detector of the RF classifier was 71.3% and an end detection accuracy was equal to 77.3%. The SVM classifier and just like the previous one, with the most frequent errors made on the start and end time.
Hand Load Classification
While analyzing the classification of hand loads (2.3 kg vs 6.8 kg) RF and SVM models outperformed the LR model. In load classification, RF and SVM classifiers achieved the mean accuracy of about 74%. However, one can observe that the selection of the sEMG features did not have a significant impact on classification, which gives evidence that such models can be built to be invariant to the features selected.
Practical Implications
Improved Ergonomic Assessment
The ability to automatically detect lifting activities and classify hand loads using sEMG sensors represents a significant advancement in ergonomic assessment. It could provide some benefits for the simplification of the approach to the evaluation of ergonomic risks and for the quality of the assessment in real working conditions.
Model Performance and Trade-Offs
While RF and SVM models had relatively higher accuracy, the LR model was slightly less accurate though more efficient in training than the other two models. It is highly important to find balance between the speed of execution and the performance especially when real time analysis is required.
Future Research Directions
As pointed out in the study, the focus for future work is the further refinement of the ML algorithms, and the development of more analysis of the characteristics of sEMG data. More diverse lifting tasks and participants of the study could add more specifics to the models which would make the models even more precise.
The future holds many new possibilities, but one thing remains clear. We must adapt the way in which we approach “hard work”.