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 measureme nt, or to validate an existing one. This paper provides some general guidel ines for the design of software sensors based on empirical data. When the m odel is a priori unknown, the problem can be stated in terms of non-paramet ric regression or black-box modelling. Complexity control is the main diffi culty 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 alg orithm, which also estimates the relevance of input features in the identif ication process. A data-driven software sensor should also provide accuracy and validity indexes of its prediction. We show how these indexes can be e stimated for complex non-parametric methods, such as neural networks. An ap plication in environmental monitoring, the design of an ammonia software se nsor, illustrates each step of the approach. (C) 1999 Elsevier Science B.V. All rights reserved.