Armin Bonakdar

Abstract

Work-related musculoskeletal disorders (WMSDs) pose a global challenge, impacting millions annually. Performance fatigue, a key contributor to WMSDs, is often detected by monitoring muscle activities using electromyography (EMG) sensors. However, the application of EMGs in actual jobsites has drawbacks, as they are expensive and their measurement can be affected by signal noise and external factors such as sweating. Alternatively, variation in body joint angle coordination measured by wearable inertial measurement units (IMU) can be considered to detect fatigue. This study aims to explore performance fatigue levels during long- term activities using IMU readouts.

Methods

In our experimental study, we recruited five able-bodied participants in a material handling experiment structured as a series of repetitions: lifting a 7.2 kg box from a 15 cm table (lifting task), transferring it to a 75 cm table (carrying task), and lowering it to the initial table (lowering task). We placed six IMUs on the sternum, sacrum, right upper arm, forearm, thigh, and shank. The knee, hip, trunk, shoulder, and elbow joint angles were calculated using the IMU readouts. Participants evaluated their perceived exertion using the Borg scale (1 to 10) every two minutes throughout the experiment, and the experiments concluded when a participant reported a score of 10. Then, based on the reported perceived exertion, lifting, carrying, and lowering tasks were categorized into five exertion levels. Joints’ ranges of motion and root mean square (RMS) values were calculated for the abovementioned groups and showed a descending behavior as participants felt more exerted.

Results and Conclusion

During lifting and lowering repetitions, both the mean and peak angles of the knee decreased, alongside a reduction in the knee's range of motion. Conversely, during the carrying activity, mean and peak angles increased for all joints except the shoulder. Additionally, shoulder range of motion decreased specifically during the lifting activity. The root mean square (RMS) values increased for all joint angles except for the knee during lowering, indicating fatigue onset. To model these observations, a long-short term memory (LSTM) network was employed, utilizing a window size of five repetitions. The input data structure comprised dimensions of (2733, 5, 60), and the output labels were based on perceived fatigue reports, categorized into three levels: 0 (RPE 1-3), 1 (RPE 4-7), and 2 (RPE 8-10). The model's evaluation was conducted using subject-wise cross-validation, achieving an accuracy of 66%.