Artificial neural network based coincidence correction for optical aerosol spectrometers

Abstract

Many applications such as filter testing, healthcare, quality monitoring, and environmental measurements require precise aerosol quantification by optical aerosol spectrometers. This type of measurement equipment is capable of in-situ measurements and provides easy access to the size distribution of the particles. Due to the coincidence error, optical aerosol spectrometers are limited to applications with relatively low concentrations. At high concentrations, the counting efficiency is reduced, while the size distribution is shifted towards larger particles. In 1984 Raasch and Umhauer proposed an analytical correction method for the size distribution. Although the approach is easy to implement, it has some disadvantages. In this work, an alternative correction method for the size distribution is presented, which is based on neural networks. The performance of both correction methods is evaluated on the cumulative distribution of raw detector voltages. The relative error of the median, as well as an error integral over the whole distribution is used as a measure. The neural network-based method gives a correction result that shows approximately half the relative median error, and a third of the error integral compared to the method of Raasch and Umhauer, for high concentrations.