Software sensor design based on empirical data

Software sensor design consists of building an estimate of some quantity of interest. This estimate can be used either to replace a physical measurement, or to validate an existing one. This paper provides some general guidelines for the design of software sensors based on empirical data. When the model is a priori unknown, the problem can be stated in terms of non-parametric regression or black-box modelling. Complexity control is the main difficulty in this setting. A trade-off must be achieved between two antagonist goals: the model should not be too simple, and model identification should not be too variable. We propose to address this issue by a penalization algorithm, which also estimates the relevance of input features in the identification process. A data-driven software sensor should also provide accuracy and validity indexes of its prediction. We show how these indexes can be estimated for complex non-parametric methods, such as neural networks. An application in environmental monitoring, the design of an ammonia software sensor, illustrates each step of the approach.