Press "Enter" to skip to content

Calibration of a Kinetic Cycling Pedal

Jonas Ebbecke 3

Last updated on 4. July 2021

Abstract of my Master’s thesis at the Institute for Biomechanics and Orthopaedics at the German Sport University Cologne.

Introduction: In order to understand the underlying mechanical principles of cycling, a kinetic cycling pedal has been developed in which two piezoelectric force sensors were embedded. The aim of this work was to calibrate this pedal by training a multilayer perceptron (MLP) in order to model the non-linear relationship between force input and sensor output and to calculate three force components as well as the point of force application (PFA). Subsequently, the outputs were validated.

Methods: For the calibration, an MLP with one hidden layer and 28 neurons was trained with a learning rate of 0.01. To assess die network’s mechanical validity, a set of unseen data was fed into the MLP and the root mean squared errors (RMSE) between inputs and outputs were calculated. Subsequently, the pedal was mounted on an SRM ergometer. Kinematic and kinetic data was recording as a subject pedalled at 100 W, 200 W and 300 W in order to assess the in-situ validity. The tangential forces of the pedal and the ergometer were calculated and compared. Further, the subject’s shoes were equipped with novel pedar pressure-measuring insoles. These were used for direct comparison between the vertical forces and the comparison of centre of pressure with the PFA.

Results: The results showed high accuracies in the mechanical validation for all force components (vertical: 1.44 ± 1.04 N, anterior-posterior: 2.72 ± 1.62 N, medio-lateral: 2.60 ± 1.04 N) and the PFA (1.85 ± 1.33 mm). In-situ, satisfactory accuracies of the PFA during the loading phase could be measured. Here, the trends of the vertical and tangential forces were similar and correlations high (100 W: r = 0.9975, p < 0.001; 200 W: r = 0.9698, p < 0.001; 300 W: r = 0.9728, p < 0.001), but the error values were unacceptably high.

Discussion and Conclusion: The results of the mechanical validation show that the approach of using MLPs for the calibration of the kinetic pedal is promising. Here, higher accuracies could be achieved than with most force-measuring pedals developed so far. However, in order to achieve equally high in-situ accuracies, adjustments have to be made in the training data acquisition. Nevertheless, the kinetic pedal shows great potential for understanding biomechanical determinants in cycling.

Main References

  1. Alexander, N., Strutzenberger, G., Jenny, H., Augustin, H., Schwameder, H., 2015. Static and dynamic evaluation of a pedal system for measuring three-dimensional forces in cycling. Proceedings of the Institution of Mechanical Engineers, Part P: Journal of Sports Engineering and Technology 229, 222–230. https://doi.org/10.1177/1754337115577029
  2. Bini, R.R., Hume, P.A., Croft, J., Kilding, A.E., 2013. Pedal force effectiveness in Cycling: a review of constraints and training effects. Journal of Science and Cycling 2, 11–24. Researchgate
  3. Broker, J.P., Gregor, R.J., 1990. A Dual Piezoelectric Element Force Pedal for Kinetic Analysis of Cycling. International Journal of Sport Biomechanics 6, 394–403. https://doi.org/10.1123/ijsb.6.4.394
  4. Gautschi, G. (Ed.), 2012. Piezoelectric sensorics. Force, strain, pressure, acceleration and acoustic emission sensors, Springer
  5. Goodfellow, I., Bengio, Y., Courville, A., 2016. Deep learning. MIT Press, Cambridge, Massachusetts, London, England. Deep Learning (deeplearningbook.org)
  6. Jabbar, H.K., Khan, R.Z., 2014 – 2014. Methods to Avoid Over-Fitting and Under-Fitting in Supervised Machine Learning (Comparative Study). In Computer Science, Communication and Instrumentation Devices. Research Publishing Services, Singapore, pp. 163–172. Research    Publishing Services, Singapore. Researchgate
  7. Lu, T.-W., Chang, C.-F., 2012. Biomechanics of human movement and its clinical applications. The Kaohsiung journal of medical sciences 28, S13-25. DOI: 10.1016/j.kjms.2011.08.004
  8. Schmidhuber, J., 2015. Deep learning in neural networks: an overview. Neural networks : the official journal of the International Neural Network Society 61, 85–117. DOI: 10.1016/j.neunet.2014.09.003
  1. Hartmut Ebbecke Hartmut Ebbecke

    Ich bin mal sehr gespannt auf die Ergebnisse des Projektes – werden bestimmt hier auch veröffentlicht…

Leave a Reply

Your email address will not be published. Required fields are marked *