Abstract
Monitoring workers’ status is crucial to prevent work-related musculoskeletal disorders and to enable a safe human-robot interaction. This is typically achieved relying on muscle activation recordings, commonly performed via wearable electromyographic EMG sensors. However, to properly acquire whole-body muscular status, a large number of sensors is needed. This represents a limitation for a real deployment of wearable acquisition systems, due to cost and wearability constraints. To overcome this problem, we propose a solution to provide a reliable muscles estimation from a limited number of EMG recordings. Our method exploits the covariation patterns between muscles activation to complement the recordings coming from a reduced set of optimally placed sensors, minimizing the estimation uncertainty. Using a dataset of EMG data recorded from 10 subjects, we demonstrate that it is possible to reconstruct the temporal evolution of 10 whole-body muscles with a maximum normalized estimation error of 13%, using only 7 EMG sensors.
Introduction
Surface Electromyography (sEMG) serves as a critical tool for assessing physical fatigue and preventing Work-related MusculoSkeletal Disorders (WMSDs). In industrial environments, comprehensive monitoring of a worker's muscular state requires a high density of sensors. However, deploying a large array of wearable sensors is often impractical due to high costs, maintenance difficulties, and restrictions on the user's range of motion. This research addresses the challenge of maintaining high-fidelity biomechanical monitoring while significantly reducing the number of physical sensors required.
Methodology
The proposed framework reconstructs missing muscle data through a three-phase process:
- Encoding (fPCA): Functional Principal Component Analysis (fPCA) is used to handle the time-series nature of EMG signals. It represents complex, time-varying activations as a set of static weights based on common patterns (motor synergies) identified in training data.
- Estimation (MVE): Minimum Variance Estimation (MVE) utilizes a database of muscle co-activation patterns to estimate the weights of unmeasured muscles based on the data from the active sensors.
- Decoding: The estimated weights are transformed back into the time domain to reconstruct the full temporal evolution of muscle activity.
To optimize sensor placement, a Genetic Algorithm was employed to minimize the Schatten p-norm of the posterior covariance matrix, which represents the overall estimation uncertainty.
Experimental Setup
The study utilized a dataset from 10 right-handed subjects performing various industrial tasks, including lifting/lowering boxes, drilling, and painting. Recordings were taken from 10 muscles: Deltoids (anterior, middle, posterior), Biceps, Triceps, Trapezius, Erector Spinae, Gluteus Maximus, Rectus Femoris, Biceps Femoris, and Tibialis Anterior.
Results and Discussion
The results demonstrate that human motion redundancy allows for significant sensor reduction. The system successfully removed up to 3 sensors (a 30% reduction) without substantial loss in estimation quality.
Key findings include:
- Low Error Rates: The maximum normalized estimation error was 13% for the Gluteus Maximus, with most other muscles showing errors around 6%.
- Temporal Fidelity: Reconstructed signals closely matched ground truth recordings in both magnitude and timing, capturing essential muscle burst patterns.
- Cross-Subject Stability: The fPCA components were consistent across subjects, suggesting that models trained on general populations could be applied to new individuals with minimal calibration.
Conclusion and Future Work
While the current results are promising, the study's scope was limited to 10 subjects and 10 muscles. Future work will focus on scaling the system to a larger number of muscles and developing real-time estimation capabilities, as the current fPCA approach requires post-processing of completed movements. Additionally, integrating kinematic data from Inertial Measurement Units (IMUs) could further enhance estimation accuracy and provide a more comprehensive ergonomic assessment tool.
Summary updated