01 March 2005

ISA Certified Automation Professional (CAP) program

This question is from the CAP study guide, Domain II.

Definition: Identify customer requirements and complete high-level analysis of the best way to meet those requirements.

CAP question

What is the MOST appropriate technique for modeling linear relationships for a large number of correlated inputs where the equations are unknown?

A. Artificial neural networks

B. Multivariable statistical process controls

C. Step response models

D. First principle models

CAP answer

The correct answer is B, multivariable statistical process controls.

A natural neural network

Artificial neural networks (ANN) excel at modeling nonlinear relationships for a relatively large unknown number of inputs. However, the inputs don't correlate, and the training data must cover the whole region. An ANN cannot extrapolate values outside the test region and doesn't handle large lags well.

Multivariable statistical process control excels at modeling unknown linear relationships for a large number of inputs that correlate.

Step response models excel at linear relationships for a small to moderate number of uncorrelated inputs where dynamics are important. Step response models work for linear dynamic on-line property estimates.

First principle models require known equations and parameters that use process principles and material and energy balances.

Reference: Blevins, et al. Advanced Control Unleashed: Plant Performance Management for Optimum Benefit, ISA Press, 2003. 

Nicholas Sheble (nsheble@isa.org) edits the Certification department. For information about the CAP program, go to www.isa.org/CAP .


A natural neural network