Soft Sensing - Advanced Control and Supervision of Mineral Processing Plants - page 153

Tabl e 4. 5 Soft sensors in general

Reference Model class Main features Applications

NARX Identification of gas turbine using

NARX model.

NARX model of an industrial gas
turbine, developed using plant
data, shows better performance
than ARX models.

Basso et
al. , 2002
[4]

State model Infrequent sampling. Use of Fiber rate in sugar cane mill.

phenomenological knowledge.

Crisafulli,
1996 [33]

Kalman filter.

PCA Missing measurement recon-

struction (soft sensing) for
sensors within a set of sensors.
Types of faults. Fault detection.
Identification of failed sensor
through sensor validity index.
Effect of filtering. Separating
sensor fault detection from plant
changes.

Data from boiler process with a
nine sensor set.

Dunia,
1996 [34]

Various Topic containing treatment of

different topics concerning soft
sensors: justification, SS mod-
els, measurement selection, sen-
sor fault detection, etc.

Examples of soft sensors in in-
dustrial plants, except mineral

Fortuna
et al. ,
2007 [1]

processing plants.

Various Paper containing a review of

soft sensors including topics
in SS model design, different
model classes (especially LIP
or NARX), clustering, effect on
control loop due to replacement
of sensors by SS, the need for a
SS administration system.

Examples of soft sensors in min-
eral processing plants and other
plants. Control.

Gonzalez,
1999 [3]

Various Examples on benefits provided by

soft sensors in several industrial
plants. The need for data pre-
processing and adaptation.

Example of commercial packages
for SS development. Soft sen-
sor basics is addressed. Examples
showing SS displays in industrial
environment.

Control
Eng.
Europe,
2001 [2]

Transfer func- Effects of replacing actual sen-

sors by soft sensors in control
loops.

Tests using simulation.

tion

Gonzalez
et al. ,
1996 [35]

ARX Design problem when soft sensor

replaces sensor in control loop.

Study using model and simula-
tion.

Gonzlez,
1992 [36]

Notation: LIP = Linear-in-parameters; NARX = Nonlinear ARX; NN = Neural Network;
SVM = Support Vector Machine SS = Soft Sensor; PCA = Principal Component Analysis;
RMS = root mean square; LS =least square; RBF = radial basis function