'. PHP_EOL; } elseif ( strpos( $page_path, "deutsch") !== false) { echo ''. PHP_EOL; } elseif ( strpos( $page_path, "francais") !== false) { echo ''. PHP_EOL; } elseif ( strpos( $page_path, "italiano") !== false) { echo ''. PHP_EOL; } ?>

Case Study in Carbon Black Process Optimization

by , | Sep 25, 2019 | Event, Production, Services, Consulting & Training

Jim Cahill

Jim Cahill

Chief Blogger, Social Marketing Leader

Emerson's Dr. Tiffany TangEmerson’s Dr. Tiffany Tang presented Novel Idea of Model Predictive Control on Fast Process! First Application at the 2019 Emerson Exchange conference. Here is Tiffany’s abstract:

Carbon black formation reaction is a fast process in the order of seconds. An application of model predictive control (MPC) on such a fast process was recommended. MPC was proposed to close the loop on product quality, Iodine number, and oil-to-air ratio to eliminate manual control and enhance product quality. A neural network model was developed to provide continuous prediction of Iodine number, which will be used as a control variable (CV) in the MPC design. This will be one of the first implementations of MPC in the carbon black industry. The Phase I of this project – Regulatory control audit and Inferential sensor development has been successfully implemented.

Tiffany is an Operational Certainty consultant in process optimization and advanced process control and has a PhD from Texas Tech. She opened describing this carbon black project challenges and the approach to address these challenges.

This carbon black producer manufacturers this product for the tire industry. Challenges in production were variations in quality and areas of manual operation that could be improved by automation. The objective for the project was to reduce off-spec products and improve operator efficiency. Having to rework off spec product doubled the energy consumption in this energy intensive process.

The carbon black process is a very fast process. The time to steady state is around 15-20 minutes. Model Predictive Control (MPC) is mostly applied is processes with large deadtime. MPC for fast processes requires special design and consideration. An online analyzer was needed to measure product quality. Delays in getting lab sample results from the prior manual sampling process would cause difficulties for an MPC solution. Finally, the MPC control strategy had to mesh with the regulatory control.

The project has two phases. In the first phase, modifications were made to the regulatory control strategy and the loops were retuned. A neural network-based inferential soft sensor was created to measure carbon black grade quality in real time.

Tiffany described each of these 3 changes in phase 1 of the project. For the MPC in phase 2, three manipulated variables (MVs) were identified—gas flow loop, oil flow loop, and process air flow loop. Bumpless transfer between the MPC and local control was important to avoid any upsets in the case of a mode change.

To address the loop tuning, the Entech Toolkit was used to identify the process dynamics. The MV setpoints are changed to understand the changes in the process variable over time. When the project team audited the regulatory control loops, transmitter and control valve issues were also identified. By addressing the valves, measurement devices and loop PID tuning parameters using the Lambda tuning method, the regulatory control was significantly improved. Before a control valve performance issue could be addresses, Tiffany bumped the PID reset parameter higher to compensate for the performance issue. It will be reset to the recommended level after the valve is fixed.

Tiffany next recapped the inferential sensor developed. The measurement was for Iodine Number (I2). By inferring this number through the neural network calculations the control strategy could adjust the oil feed flow rate based on this calculated value. The neural network was trained with the history of laboratory sample data and the key process parameters. A sensitivity analysis in DeltaV Neural is done to identify these key process parameters. Low impact parameters are eliminated during the model building process. The model was trained with a month and a half’s worth of data. For a more accurate model, they are currently pulling 6 months of data to improve the accuracy of the model.

The next phase will be to incorporate Model Predictive Control. Hopefully I’ll be able to provide an update in a future post.

Popular Posts


Related Posts

Follow Us

We invite you to follow us on Facebook, LinkedIn, Twitter and YouTube to stay up to date on the latest news, events and innovations that will help you face and solve your toughest challenges.

Do you want to reuse or translate content?

Just post a link to the entry and send us a quick note so we can share your work. Thank you very much.

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.