Refiner Creates Property Estimators with Neural Networks

by | Jun 6, 2007 | Control & Safety Systems, Downstream Hydrocarbons, Industry, Services, Consulting & Training

Jim Cahill

Chief Blogger, Editor

Last October, I featured one of Emerson’s advanced automation service consultants, Lou Heavner, and how he worked with Lukoil to create virtual sensors based on neural networks.

Their efforts were told in more detail in the March 2007 issue of InTech magazine. The article, entitled, Crude gets smart, described the Russian refiner’s challenge to keep their refined products within specification. They had been relying on lab samples that came back from the lab to the operators only once or twice a day.

To get feedback on product quality and composition more frequently, Lou and the team used neural network blocks in their DeltaV system’s controllers to create property estimators. As the article states:

The goal of a property estimator is to provide an accurate gauge of product quality, especially after lab results have become stale, which is most of the time. Property estimators are not intended to eliminate lab analyses, although the frequency of analyses may lessen once estimators are proven. Even though estimators may not be as accurate as lab analyses, they can be worthwhile calculated variables to help engineering and operations personnel monitor, troubleshoot, or understand and control the process.

The article describes the steps the team took to collect the data to train the neural network models. It offers guidance for those looking to implement property estimators. Some examples of their recommendations include:

  • The time stamp should reflect the time of data extraction from the process–not when it was scheduled for sampling, or when the lab technician performed the analysis, or when they reported the lab results.
  • Avoid filtering or manipulating the process data. Raw snapshot data usually makes for the best models.
  • If the process does not vary much, the model will not be reliable if the process wanders into a range with no collected data… the model will be changed to “Uncertain” and the operator can be alerted.

The team believes they may have one of the world’s largest installations in terms of neural network models. Currently operating models include ones measuring boiling points, flash points and viscosity on the pre-flash, atmospheric, and vacuum towers.

If operators at your plant are waiting on lab information to make quality adjustments to the process, you may have a business case for creating property estimators to augment the lab sampling process.

Submit a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Submit a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Subscribe for Updates

Follow Us

We invite you to follow us on Facebook, LinkedIn, Twitter and YouTube to keep up to date on all the latest news, events and innovations to help you take on and solve your toughest challenges.

Want to re-purpose, reuse or translate content?

Please do, Just link back to the post and send us a quick note so we can share your work. Thanks!

Our Global Community

Emerson Exchange 365

The opinions expressed here are the personal opinions of the authors. Content published here is not read or approved by Emerson before it is posted and does not necessarily represent the views and opinions of Emerson.