Project Description

This research combining new mathematical signal processing techniques and existing techniques to enhance sensors and instruments that are, or will be, manufactured in NZ. Specifically, we are focussing on the new concept of Bayesian Model Prototyping (BMP) which is of direct relevance to existing sensors.  Secondly, we are employing more specialised techniques for time-invariant deconvolution and accelerated convergence for inverse problems, primarily concerned with measurements of moisture distribution in a range of composite materials.  These techniques too, are aimed at adding value to New Zealand's instrumentation and other sensor-based industries.  The work is divided into three main tasks:

1)  Bayesian model prototyping:  BMP is being configured to form a flexible tool ready to be incorporated within our processing platform to augment sensor outputs for sensor manufacturers.  Validation uses simulated and measured data.  The processing platform is also being configured with well-known linear methods, including factor analysis and principal component analysis.  The techniques will provide mixed model descriptors of sensor performance with mechanistic, deterministic and Gaussian terms, to provide product quality assurance.  Sensor clustering will also be enabled for enhanced sensing accuracy.

2)  Advanced signal processing:  Very high speed sampling requires detailed design, calibration, and reduced sensitivity to instrument limitations, but existing transforms which are used for calibrating and deconvolving time domain signals distort timing.  We are configuring a new, time-preserving transform, and validating using time domain reflectometry and measurements in known dielectrics.

3)  Accelerated convergence:  New methodologies are being investigated to accelerate and optimise the convergence of non-linear problems for faster modelling and tomographic inversion.  To meet this objective, we are exploring the relationship between moment method basis function order and cell size, considering various Jacobian surrogates for rapid execution, investigating accurate but rapidly calculated priors, eliminating time-consuming line searches in the conjugate gradient method, and optimising the number and position of measurements by generating a suitable measure of data ill-posedness.

Page last updated on 12 April 2005