SENSOR OPTIMIZATION USING NEURAL-NETWORK SENSITIVITY MEASURES

A new method of optimizing a multi-sensor geometry using neural networ k function fitting acid sensitivity measures is described. The method is applied to a multi-angle optical scattering nephelometer for which theoretical scattering intensities are generated for distributions of spherical dielectric particles. Neural networks are trained to invert these angular intensities to determine accurately the size distributio n of normally distributed particles. The nephelometer model is optimiz ed to a minimum configuration using the sensitivity analysis. The meth od is further validated on experimental data by identifying essential channels in an on-line nephelometer used to determine concentration an d species of oil-in-water suspensions.