Page 29 - ASME DSCC 2015 Program

Technical Program

nonlinear Model Predictive Control of functional Electrical Stimulation robust Bayesian Sequential Input Shaping for optimal Li-Ion Battery
Invited session paper. DSCC2015-9762 Model parameter Identifiability
nicholas Kirsch, naji Alibeji, nitin Sharma, University of Pittsburgh, Contributed regular paper. DSCC2015-9942
Pittsburgh, PA, United States Michael Rothenberger, Hosam K. fathy, Pennsylvania State University,
University Park, PA, United States
One of the major limitations of functional electrical stimulation (FES) is the
rapid onset of muscle fatigue. Minimizing stimulation is the key to decreas- This paper examines the challenge of shaping a battery’s input trajectory to
ing the adverse effects of muscle fatigue caused by FES. Optimal control (i) maximize its Fisher parameter identifiability while (ii) achieving robust-
can be used to compute the minimum amount of stimulation necessary to ness to parameter uncertainties. The paper is motivated by earlier research
produce a desired motion. In this paper, a gradient projection-based model showing that the speed and accuracy with which battery parameters can be
predictive controller is used for an approximate optimal control of a knee ex- estimated both improve significantly when battery inputs are optimized for
tension neuroprosthesis. A control Lyapunov function is used as a terminal Fisher identifiability. Previous research performs this trajectory optimization
cost to ensure stability of the model predictive control. for a known nominal parameter set. This creates a tautology where accurate
parameter identification is a prerequisite for Fisher identifiability optimiza-
Analysis of a Tele-operated MRI-Compatible Vane Actuator for
tion. In contrast, this paper presents an iterative scheme that: (i) uses prior
neuromuscular facilitation in Hemiparetic limbs
parameter probability distributions to create a weighted Fisher metric; (ii) op-
Invited session paper. DSCC2015-9992
timizes the battery input trajectory for this metric using a genetic algorithm;
Melih Turkseven, Jun ueda, Euisun Kim, Georgia Institute of Technology, (iii) applies the resulting input trajectory to the battery; (iv) estimates battery
Atlanta, GA, United States, Ilya Kovalenko, Georgia Institute of Technology, parameters using a Bayesian particle filter; (v) re-computes the weighted
Martinez, GA, United States
Fisher information metric using the resulting posterior parameter distribution;
Analysis of a Tele-operated Mri-comaptible Vane Actuator for Neuromuscu- and (vi) repeats this process until convergence. This approach builds on
lar Facilitation in Hemiparetic Limbs well-established ideas from the estimation literature, and applies them to the
battery domain for the first time. Simulation studies highlight the ability of this
ConTRIBuTED SESSIon
1-4-2 WM3 estimation and Identification 2 iterative algorithm to converge quickly towards the correct battery parame-
ter values, despite large initial parameter uncertainties.
George Bellows E 1:30pm–3:30pm
Expectation Maximization Method to Identify an Electrically Stimulated
Session Chair: Jongeun Choi, Michigan State University Musculoskeletal Model
Session Co-Chair: nabil Chalhoub, Wayne State University Contributed regular paper. DSCC2015-9956
harish ravichandar, ashwin dani, jacquelyn khadijah-hajdu, University
Reconstruction of nonlinear Characteristics by Means of Advanced of Connecticut, Storrs, CT, United States, nicholas Kirsch, Qiang Zhong,
observer Design Approaches nitin Sharma, University of Pittsburgh, Pittsburgh, PA, United States
Contributed regular paper. DSCC2015-9897
A system identification algorithm for a musculoskeletal system using an
fateme Bakhshande, dirk Söffker, University of Duisburg-Essen, Duisburg, approximate expectation maximization (E-M) is presented. Effective control
Germany
design for neuroprosthesis applications necessitates a well defined muscle
The paper introduces into the concept of model-based force estimation of model. A dynamic model of the lower leg with a fixed ankle is considered.
flexible structures with impacts. The advantage of using extended PI-Ob- The unknown parameters of the model are estimated using an approxi-
server (MAPIO) in comparison with the well-known PI-Observer is investigat- mate E-M algorithm based on knee angle measurements collected from
ed with respect to the reconstruction of a complex nonlinear unknown input. an able-bodied subject during stimulated knee extension. The parameters
Therefore a clamped beam example is considered. The task of the observer estimated from the data are compared to reference values obtained by con-
scheme is to estimate the tip position and the effecting input as unknown ducting experiments that separate the parameters in the dynamics from one
input. This task is also realized in the presence of measurement noise. The another. The presented results demonstrate the capability of the proposed
reconstruction of complex nonlinear spring behavior is performed to prove algorithm to identify the parameters of the dynamic model from knee angle
the performance of MAPI-Observer in comparison to PI-Observer. measurements.

29