Hasso-Plattner-Institut
Prof. Dr.-Ing. Bert Arnrich
  
 

Predicting Fatigue in Strength Training using Machine Learning on IMU and HRV Data

Master's Thesis

Arne Herdick, Supervisor: Justin Albert

Quantifying load during physical activity has been of high interest to the research community for a long time. Athletes should optimize the exercises to align the applied training load most closely with the training plan's value. For the general population, exercise load during, e.g., rehabilitation, is also essential. Quantifying load is usually done by considering two values. The first is the external workload, e.g., distance traveled, the travel speed, or the weight of strength training equipment. The second is the internal load, often measured as the Rating of Perceived Exertion (RPE). RPE specifies how exhausting an exercise was and is different between people. Measuring external load can be achieved using sensor devices such as Inertial Measurement Units (IMUs) or Linear Transducers to calculate, e.g., current extension during a barbell squat [1, 2]. Measuring subjective exertion with an RPE scale, such as the Borg scale, has the advantage of implicitly containing all relevant parameters, even in sessions or situations containing various exertion types. Usually, RPE values are retrieved between 15 and 30 minutes after the training session [3]. This strategy is employed to avoid influence on the entire session's result by the last few minutes. There is, however, interest in getting RPE ratings on the fly, which is why, similar to the measurement of objective values, research investigated the prediction of RPE values. 

This thesis investigates the prediction of subjective RPE value ratings using IMU and Heart-rate variability (HRV) measurements during resistance training. The motivation for predicting RPE values during training is that overexertion has adverse effects. If an RPE value can be predicted and signals exertion over the expected training threshold, users can be warned to change their behavior. Predicting RPE values from sensor data is informative during exercises and might also help athletes track their exertion more accurately. Given accurate predictions for the RPE, athletes should set the desired exertion according to their training plan and follow it accurately. Coaches might also use the produced data to interpret whether the training person might be cheating on their RPE scores to modify further training [4]. 

References

[1] Fernando Naclerio and Eneko Larumbe-Zabala. “Relative load prediction by velocity and the OMNI-RES 0–10 scale in parallel squat”. In: The Journal of Strength & Conditioning Research 31.6 (2017), pages 1585–1591.

[2] Brendan R Scott, Grant M Duthie, Heidi R Thornton, and Ben J Dascombe. “Training monitoring for resistance exercise: theory and applications”. In: Sports Medicine 46.5 (2016), pages 687–698.

[3] Justin A Kraft, James M Green, and Kyle R Thompson. “Session ratings of perceived exertion responses during resistance training bouts equated for total work but differing in work rate”. In: The Journal of Strength & Conditioning Research 28.2 (2014), pages 540–545.

[4] Joseph OC Coyne, G Gregory Haff, Aaron J Coutts, Robert U Newton, and Sophia Nimphius. “The current state of subjective training load monitoring—a practical perspective and call to action”. In: Sports Medicine-Open 4.1 (2018), page 58.