Fatigue Monitoring in Running Using Flexible Textile Wearable Sensors

Mohsen Gholami, Christopher Napier, Astrid García Patiño, Tyler J. Cuthbert, Carlo MenonSensors 2020 · OpenReview

TL;DR

This paper introduces a novel system using flexible textile strain sensors woven into running garments to continuously monitor lower extremity kinematics during running. By applying a stacked random forest machine learning model to this kinematic data, the researchers can accurately estimate a runner's fatigue level, measured by the Borg Rating of Perceived Exertion (RPE). This work paves the way for non-invasive, real-time fatigue monitoring to potentially reduce running-related injuries.

What Problem Does This Paper Address?

Running is a popular activity, but it carries a significant risk of injury, often exacerbated by fatigue. Fatigue in runners leads to changes in gait mechanics, increasing the mechanical forces on the body and raising the likelihood of overuse injuries. While methods like heart rate monitoring or blood lactate tests exist, they are either indirect or invasive. Subjective measures like the Borg RPE scale are useful but rely on self-reporting. The challenge is to find an objective, non-invasive, and continuous method to monitor fatigue during running, especially outside of a lab setting. Previous research has explored wearable sensors like IMUs for this purpose, but the use of flexible textile strain sensors, which can be seamlessly integrated into clothing, for fatigue estimation remained largely unexplored.

Methods

The core of this study lies in the combination of flexible textile strain sensors and a machine learning approach.

  • Flexible Textile Strain Sensors: These sensors are resistance-based and made from spandex yarn coated with a carbon black thermoplastic elastomer composite. They are designed to be woven into running garments (specifically, running tights in this study). The sensors measure changes in resistance as they stretch, which directly correlates to changes in joint angles and kinematics. The system utilized nine sensors placed on the hip, knee, and ankle.
  • Data Collection: Five healthy female recreational runners participated. They ran on a split-belt treadmill while wearing the sensor-instrumented tights. Their speed was gradually increased as their perceived exertion (RPE) rose. RPE was recorded every three minutes, and the treadmill speed was adjusted to reach a target RPE of 17 ("very hard").
  • Data Segmentation: Two methods were explored: stride-based segmentation and a moving window approach. The moving window method, with a 1-second window length, yielded the best results.
  • Feature Extraction: Statistical and temporal features were extracted from the sensor data, including mean, minimum, maximum, and range of motion, as well as their variations over time. Features were also extracted from the first derivative of the sensor signals.
  • Machine Learning Algorithm: A stacked random forest regression model was employed. This involved a two-step process:
  • First Model: Trained on extracted features to provide a rough estimation of the RPE level.
  • Second Model: Trained on the predicted values from the first model to refine the estimation and match it more closely to the actual RPE. This two-step approach helps interpolate RPE values between reported scores. The final fatigue level was determined by applying a voting window to the predicted values and rounding to the nearest integer RPE score.

Datasets

The study utilized data collected from five healthy female recreational runners during a prolonged running session on a treadmill. This dataset, comprising kinematic data from flexible textile strain sensors and corresponding Borg RPE scores, was collected specifically for this research. The authors note that the limited sample size of five participants was due to COVID-19 restrictions.

Key Insights and Ideas

  • Seamless Integration: Flexible textile strain sensors offer a promising avenue for unobtrusive, wearable fatigue monitoring, as they can be directly integrated into everyday athletic apparel.
  • Kinematics as a Fatigue Indicator: Subtle changes in lower extremity kinematics, particularly at the hip, are reliable indicators of increasing fatigue during running.
  • Participant-Specific Models: The effectiveness of different sensor locations varied significantly between participants, highlighting the individualized nature of fatigue and the need for personalized models.
  • Hip Dominance: Hip sensors proved to be the most crucial for estimating fatigue, suggesting that future iterations might achieve high accuracy with fewer sensors, potentially focusing solely on the hip.
  • Beyond Binary Detection: The study moves beyond simply detecting "fatigued" vs. "non-fatigued" to providing a graded, continuous estimation of fatigue level, which is more valuable for training and injury prevention.

Key Results

The stacked random forest model achieved impressive results in estimating the Borg RPE score:

  • Overall Performance: An average R-squared (R²) value of 0.96 and a Root Mean Squared Error (RMSE) of 0.06 were obtained across participants using the window-based segmentation method.
  • Participant-Specific Performance: Individual R² values ranged from 0.93 to 0.98, with RMSE values as low as 0.02 for some participants.
  • Sensor Importance: Excluding hip sensors led to a relatively small decrease in R² (0.01), while excluding knee or ankle sensors resulted in substantially worse performance. Ankle sensors alone yielded a very low R² of 0.10.

Limitations and Future Work

The authors acknowledge several limitations and propose future research directions:

  • Sample Size: The study was limited to five female participants due to the COVID-19 pandemic. A larger and more diverse sample is needed for more generalizable results.
  • Participant-Specific Models: While the current models are participant-specific, the authors aim to develop generalizable models that can predict fatigue across different individuals. This could involve exploring motor learning approaches.
  • Subjective Fatigue Measure: The study relied on the Borg RPE scale, a subjective measure. Future work could incorporate physiological measures (e.g., heart rate, blood lactate) alongside kinematic data for a more comprehensive fatigue assessment.
  • Sensor Placement Optimization: While hip sensor placement was optimized, further investigation into optimal sensor configurations for other joints and for generalizable models is warranted.
  • Broader Applications: Future research could explore monitoring bilateral lower extremity kinematics and heart rate using strain sensors to create a more robust and generalizable fatigue estimation model. The potential for real-time biofeedback applications to prevent running-related injuries is a key future direction.

Summary updated

Abstract

Fatigue is a multifunctional and complex phenomenon that affects how individuals perform an activity. Fatigue during running causes changes in normal gait parameters and increases the risk of injury. To address this problem, wearable sensors have been proposed as an unobtrusive and portable system to measure changes in human movement as a result of fatigue. Recently, a category of wearable devices that has gained attention is flexible textile strain sensors because of their ability to be woven into garments to measure kinematics. This study uses flexible textile strain sensors to continuously monitor the kinematics during running and uses a machine learning approach to estimate the level of fatigue during running. Five female participants used the sensor-instrumented garment while running to a state of fatigue. In addition to the kinematic data from the flexible textile strain sensors, the perceived level of exertion was monitored for each participant as an indication of their actual fatigue level. A stacked random forest machine learning model was used to estimate the perceived exertion levels from the kinematic data. The machine learning algorithm obtained a root mean squared value of 0.06 and a coefficient of determination of 0.96 in participant-specific scenarios. This study highlights the potential of flexible textile strain sensors to objectively estimate the level of fatigue during running by detecting slight perturbations in lower extremity kinematics. Future iterations of this technology may lead to real-time biofeedback applications that could reduce the risk of running-related overuse injuries.
BibTeX
@inproceedings{gholami2020fatigue,
  author = {Mohsen Gholami and Christopher Napier and Astrid García Patiño and Tyler J. Cuthbert and Carlo Menon},
  title = {{Fatigue Monitoring in Running Using Flexible Textile Wearable Sensors}},
  booktitle = {Sensors},
  year = {2020},
  url = {https://openreview.net/forum?id=gs-ee0e023},
}
Figure