Soft sensors for processing plants

Abstract

Soft-sensors assist in solving the problem created by the unavailability of a sensor by providing a software backup for it, thus allowing a reduction of losses in plant performance. Similarly, the use of a soft sensor to estimate a plant variable for which no sensor is installed can improve plant performance. The core of a soft-sensor is a partial plant model allowing the generation of a estimated measurement to replace missing measurements. Coupled with the model is a problem of signal estimation, interpolation and prediction. Modes considered here are black box models and gray models which include phenomenological knowledge. Also, comparisons are made concerning the requirements for soft-sensor models and for models used in model based control. An approach to an integrated view of the various soft-sensor modeling methods is attempted. Other aspects considered include use for interpolation and prediction of measurements having a sampling rate which is too low; performance of a control loop; performance of the soft-sensor indication in the period following the removal of the actual sensor, as plant characteristics change; complementary considerations for ensuring the availability of soft-sensors in industrial environments; and problems related to the use of soft-sensors in automatic control loops. A review made of a sample of the technical literature on applications as well as of research and development in this field, is commented in the text and summarized in a table.