Multiple regression based prediction correlations for enhanced sensor design of magnetic induction tomography systems

Magnetic induction tomography (MIT) is an imaging technology that measures changes in the electric properties of a sample located within the imaging region. Measurement of low conductivity contrasts such as biological tissue or ionized water flow in pipelines requires highly accurate systems due to the small amplitude of the measured signals. Optimisation of the sensors results in enhanced MIT performance. Geometric characteristics of MIT sensors impact the intensity of the electromagnetic field, and the inductive coupling between (a) the sensors and (b) the sensors and the medium. Three correlation models are derived to help developers predict the relative performance of MIT systems for a given set of coil characteristics. Bivariate and multiple regression analyses are performed on a dataset from finite element method simulations to validate the relationship between the sensor geometry and three performance parameters for a given set of uniform background distributions. Correlation models are provided for prediction of induced voltage level, eddy currents and system sensitivity relative to the geometric characteristics of the sensors. The performance of the computed models is validated using a dataset of 180 coil designs and four uniform electrical conductivity distributions. Predictions from the developed correlations are compared to reference data from simulations and experiments. Errors estimated for the predicted performance parameters together with the variance for each correlation are presented. The predicted data fitted the reference values within ±15%, showing reasonable accuracy of the models and a balanced variance-bias trade-off. It was found that the performance of MIT systems is largely affected by the coil dimensions and the number of turns, as well as by the coil shape and wire diameter to a lesser degree.