[home] [Personal Program] [Help]

---

Session: Neuromuscular – upper extremities
Session starts: Friday 25 January, 10:30
Presentation starts: 11:00
Room: Lecture room 535

---

Mark van de Ruit (Delft University of Technology)
Winfred Mugge (Delft University of Technology)
Alfred Schouten (Delft University of Technology)

Abstract:
Humans optimally interact with their environment by adapting mechanical properties of their joints to successfully execute movements e.g. knee stiffness is modulated during the gait cycle. Time-varying system identification (TV-SysID) allows to quantify joint mechanics during dynamic motor tasks, useful when mimicking human joint behaviour in active prostheses. In recent years numerous TV-SysID methods have been developed, each with their own properties. In this study six methods are compared using simulations and experimental data. Available TV-SysID methods that were tested are: (1) Linear Parameter Varying (LPV) method; (2) Kernel Based Regression (KBR); (3) Ensemble Impulse Response Function (eIRF); (4) Basis Impulse Response Function (bIRF); (5) Short Data Segments (SDS); and (6) Ensemble Spectral Method (ESM). The limitations of each method to estimate joint stiffness were verified with simulations, where the underlying system is known. For the experiments, six healthy human participants were seated with their ankle strapped to a manipulator (Achilles, MOOG). The manipulator imposed ankle rotations (filtered Gaussian white noise: 0.1-40 Hz – RMS of ~1 deg) while participants performed a torque task by tracking a 0.5 Hz sinusoid (5 to 50% of dorsiflexion MVC). The simulations demonstrated that LPV and eIRF can identify instantaneous adaptations in joint stiffness, at the cost of requiring a lot of data. SDS, ESM, KBR and bIRF are able to identify rapid - but not instantaneous - changes in joint stiffness with less data, as they possess some smoothing. For the experiments, the joint stiffness estimated by SDS, ESM, bIRF and eIRF has a comparable sinusoidal pattern, as expected from the participants’ task. Whereas the TV-SysID methods provided similar estimates of joint stiffness, their robustness to noise and amount of data differs. Ideally, a TV-SysID method: (1) tracks rapid changes, at least as fast as humans can adapt; (2) can be applied on a single trial of data to investigate trial-by-trial variability; and (3) requires little to no prior information. To facilitate comparison of TV-SysID methods authors should publish their algorithm as well as their experimental data so new methods can be validated and compared to existing methods.