Improving cycling force sensor accuracy using multilayer perceptrons

To study musculoskeletal loading in cycling, the accurate determination of pedal reaction forces is essential. Therefore a pedal including a three-axis piezoelectric sensor [9251A, Kistler, Winterthur, Switzerland] was developed. Pilot testing showed that with uniaxial force application, the imbedded sensor exhibits excellent linearities in all three spatial axes using a standard calibration matrix. However, with combined loading in multiple axes, as is common in cycling, linearity decreases and error values increase unacceptably. Here, there seem to be complex interactions between sensor and pedal body that cannot be solved by a linear calibration. Therefore, a neural network approach was chosen to increase accuracy. The aim of the present study was to validate the pedal sensor and its error correction.