Name: Muhamad Komarudin

Email: m.komarudin@stud.umist.ac.uk

CompanyName: UMIST

Country: United Kingdom

Abstract: Hardware for Finite Element Modelling

Muhamad Komarudin 1, Trevor York 1, W.R.B. Lionheart 2

1 Dept. of Electrical Engineering & Electronics,
UMIST, PO Box 88, Sackville St., Manchester M60 1QD, UK
Tel. (44) 0161 200 4729, Fax. (44) 0161 200 4789
m.komarudin@stud.umist.ac.uk, york@fs5.ee.umist.ac.uk

2 Department of Mathematics
UMIST, PO Box 88, Sackville St., Manchester, M60 1QD, UK
Tel. (44) 161 200 8978, Fax. (44) 161 200 3669
Bill.Lionheart@umist.ac.uk


Abstract:

This paper introduces a new approach for solving the forward problem in EIT
using Finite Element Modelling (FEM) implemented in custom silicon hardware. In
iterative image reconstruction algorithms, forward solutions, that are used to
determine expected measurements from suggested distributions of material, are
compared to actual measurements and the resulting errors are used to derive
improved estimates of the cross-sectional distribution. FEM is commonly used
to solve the forward problem in order to produce high quality images.
Non-linear iterative approaches, such as the regularized Newton's method, offer
significantly improved accuracy and resolution compared to linear
reconstruction algorithms, at the cost of processing time. Consequently it
is difficult to realise real-time solutions, to satisfy the demands of many
applications.

To improve processing time, one approach is to employ networks of digital
computers that together compute the solution in parallel. It is difficult to achieve
significant degress of parallelism using this approach. An alternative
approach is to exploit the fact that a non-obtuse finite element model is
equivalent to a resistor network. The proposed technique employs a network
of programmable components that emulate the elements of a conventional mesh.
Design considerations for VLSI implementation, targeted at applications of
industrial process tomography, are described. Software simulations explore
the requirements for accuracy in processing and it is suggested that, for
tomographic applications, 10-bits may provide adequate performance. Four
programmable circuit elements, resistor/conductor ladders, capacitors and
switched-capacitors are presented. Monte Carlo circuit simulation is used to
investigate the expected variation in the values of implemented resistors
due to process tolerance.

Close Window