PDA

View Full Version : calibration using ANN



Christopher Johnston
09-14-1997, 07:13 PM
Dear Listers
We are a group working in horse biomechanics and among our "toys" we have a
force measuring horse shoe. The horse shoe is calibrated for every horse
over a force plate with a specific surface relevant for the research
question. We would like to then to use the force shoe in another enviroment
after calibration. We would like to use Artificial Neural Networks (ANN) to
produce our force data from the "other" enviroment. We have and would like
to use commercial software to train/test/predict. Some of the limitations
to date though is the difficulty in using the "network" programs with time
series. Our first attempts have therefore been to use each data point as a
node with 3 layers. The force shoe produces 6 channels of information (3
vertical and 3 horizontal). Thus we may have 180 input nodes and output say
30 for a signal stance. We use back propagation to test/predict. This seem
to be the most common method used in EMG/Force prediction studies we have
found inthe literature. The results are not totally satifactory though. We
would appreciate some help on the following
Which ANN commercial packages are available for this application?
Does anyone use a time serie treatment to train/test/predict?
Is there a better propagation method used?
Will using more information/other sources prevent memorisation as we wish a
generalisation?
I can summarize results if desired at a later date
Thank you for your help

Chris

Christopher Johnston DVM
Center for Biomechanical Studies (CBS)
Department of Anatomy and Histology
Faculty of Veterinary Medicine
Swedish University of Agricultural Sciences
750 05 Uppsala, Sweden
Tel +46-18672147
Fax +46-18672852
EMail Christopher.Johnston@ah.slu.se