Multivariate limit of detection for non-linear sensor arrays

Abstract

With the increased development of low-cost and miniature devices, sensors are increasingly being deployed as arrays of redundant sensors. However, little work has been done characterizing properties of these arrays. Here, we develop and test a Bayesian algorithm for estimating the limit of detection of sensor arrays. The algorithm is applicable for single sensors as well as sensor arrays, and works by reducing a vector in the signal domain to a univariate response in the measurand domain. We show that the new algorithm can reproduce results from a benchmark algorithm for single sensors, and then demonstrate the benefit of adding additional sensors to an array. Then, we provide guidelines that achieve numerical stability while minimising computational cost. Finally, we provide a real-world example using an array of ion-selective electrodes measuring carbonate in seawater. This application demonstrates how incorporation of a set of individual low-quality sensors into an array leads to a substantially reduced LOD that clearly meets the demands of the application.