Dynamic world of modeling and control

Aug. 2, 2016
Greg McMillan and Stan Weiner discuss dynamic modeling with Julie Smith, global automation and process control technology leader at DuPont.

[sidebar id =1]Greg: Early in my career while I was the lead instrument and electrical engineer on what at the time would be the world’s largest acrylonitrile plant, I found the compressor control systems most challenging because of how fast things could go wrong and shutdown the plant. I had a degree in engineering physics with electives in chemical engineering but nothing prepared me for what I needed to know in terms of control system setup and dynamics. In my spare time I built a dynamic model of a compressor system. It changed my life. I developed a much better understanding of causes and effects and an appreciation of dynamic modeling as an essential learning experience. It caught the attention of the process control group in the Engineering Technology (ET) section. At the end of the project, I was offered a job in ET with what were the best minds in developing software for process design and writing dynamic models for process control. Forty years later, I'm still making the most out of the synergy between dynamic modeling and control. Most of my deep understanding of process dynamics stems from building and using dynamic models.

Stan: To help us all realize this opportunity, we're gaining the perspective this month of Julie Smith, global automation and process control technology leader at DuPont. Julie, what's the size and function of your group?

Julie: We have 25 specialists in first principle dynamic modeling and recently integrated 10 specialists in multivariate statistical process control (MSPC). There are other groups doing steady state modeling for process development and design. We use dynamic models to develop process understanding and process control improvement (PCI) to increase process performance and provide training to increase operator performance. We use software developed in-house for dynamic modeling. The objects offer the easy setup and dynamic behavior of many chemical and biochemical unit operations, including reactions and thermodynamics.

Greg: I've found it insightful to realize that for PCI you can use dynamic models to explore => discover => develop => prototype => test => deploy => train. The progression of these activities is synergistic and iterative for continuous improvement. Knowledge is gained and solutions are developed with no disruption to the process. You can explore effects of process relationships and process, equipment and piping design on the dynamic response and objectives. You can discover solutions that address inherent challenges and deliver improved performance. You can develop the details of the control system including feedforward and feedback control parameters. You can prototype the solutions and refine the details of the implementation, realizing success is often in the details. You can test the virtual and actual implementation. You can provide more reliable deployment by comparing actual to virtual performance. The same dynamic models can be used for operator training before and after startup, where you not only improve operator performance, but gain the input from operator expertise that leads to better models and PCI. Periodic operator training nurtures the process of continuous improvement by restarting the progression of activities. For more details on how online metrics can identify the benefits of modeling and control, see the April 2015 Control Talk column “Getting innovation back into process control.”

Stan: Julie, how do you use dynamic models?

[sidebar id =2]Julie: We use dynamic models to improve all aspects of plant performance. The models can be as big as you want depending on the capability of your computer’s central processing unit (CPU). The focus is more on studying operability and developing control systems for new growth businesses. Here we may emulate control systems and basic graphics to run totally independently. As we move into testing, deployment or training, we use an object linking and embedding (OLE) for process control (OPC) interface to the actual distributed control system (DCS) configuration and operator graphics. For some very fast dynamics, such as burner management systems, the OPC interface is not fast enough. We also make sure the integration step size is small enough to prevent any cycling from integration of the material and energy balances. For our more mature businesses, the focus is more on training and optimization. Here, it's important to run faster than real time, so you don’t have to, for example, wait three hours for a furnace to heat up. In all cases, we strive for best practices.

Greg: What's an example?

Julie: We use dynamic models to develop plantwide control strategies. For example, in bioprocesses, water balance is critical to understanding and improving them. There's a lot of water recycling, and we need to recover some valuable components that may be slightly soluble.

Greg: I was impressed how Luyben used dynamic models to reveal the snowballing effect of recovering and recycling excess reactants to a continuous reactor. He developed some fundamental, insightful requirements, such as the need for a flow controller somewhere in the recycle path of reactant. This could be achieved by using a flow control loop of reactor discharge flow to set production rate. The level controller then manipulates the reactant feed with a level setpoint to set residence time. For more on this solution and simple techniques to improve the control of continuous reactors, see the recording of my short course, “Part 3: Reactors, Fermenters, Crystallizers, Evaporators, Distillation Columns, Dryers, and Neutralizers,” for AIChE's St. Louis chapter in April 2016. 

Julie: We find dynamic models extremely useful for dealing with the confusion, complexity and interaction in plant response from extensive recycle streams. We study operability, including how effectively can you run the process with the given difficulties, and will the process tend to settle out. One of the first things we do before we start PCI is make sure the inventory control is in place and working well. In dynamic models the level and pressure controllers must be working well to proceed with adding and improving the other control loops.

Stan: How does what you find affect plant design?

Julie: We often find tank sizes are critical for smoothing out variability in composition. Dynamic simulations help prevent too small or even missing tanks that result from attempts to reduce the cost of equipment in a “value engineering” initiative.

Greg: For pH control, I've found this attenuation of pH oscillations by volumes and the mixing and injection delays from equipment, piping and valve design can make or break a system. Dynamic fidelity depends on getting these delays right that determine the loop deadtime and matching the slope of the laboratory titration curve that sets the process gain. For strong acids and bases, getting the model slope to match the lab slope depends on including the significant moderating effect of carbon dioxide absorption and conjugate salts. For more on the critical role of modeling for pH control, see the March 2016 Chemical Processing article, “Improve pH Control.” Backlash, stiction and poor positioner sensitivity can increase the dead time by orders of magnitude, and cause cycling juts from just putting the controller in automatic. While particularly troublesome for pH loops with high process gains, poor valve response is a major source of variability in all loops. To learn how a valve response model can identify the often unrecognized causes of oscillations, see the March 2016 article, “How to specify valves and positioners that do not compromise control” and the associated whitepaper, “Valve Response—Truth or Consequences.” Dynamic simulation is incredibly valuable for defining the details of the process, mechanical and valve design as well as the control system design.

Stan: What is the focus of your multivariate statistical process control (MSPC) specialists?

Julie: We see MSPC as being particularly useful for predicting batch endpoints with the end goal of supplementing or enhancing  performance of on-line, at-line or off line analyzers. We also see opportunities discovering correlations and assisting in the design of experiments (DOE). This can increase the focus on suspicious relationships, identify abnormal conditions and significantly guide troubleshooting. 

Greg: How can we help others to realize the value of modeling and control?

Julie: We need to advertise success stories, as well as make the setup and use of the models easy. The tools have come a long way, and you don't need a PhD to get started. We need to encourage people to take the plunge

Greg: I think there should be an ISA focus group and ISA WebEx presentations on modeling and control. I think online process performance metrics in the model and in the plant before and after the deployment can greatly help the plant recognize the benefits.

Top 10 reasons not to use dynamic models

  • 10. You're a steady state type of gal/guy
  • 9. Management thinks you already know everything
  • 8. Ignorance is bliss
  • 7. How else can you do a copy job?
  • 6. You like startups that reveal true personalities like reality shows
  • 5. You like the excitement in the control room of operability crises
  • 4. Virtual reality is for gamers and “Big Bang Theory” characters
  • 3. The Internet of Things will solve all our problems
  • 2. Your motto is "What me worry?"
  • 1. The startup is in Hawaii!

[sidebar id =3]

About the Author

Greg McMillan | Columnist

Greg K. McMillan captures the wisdom of talented leaders in process control and adds his perspective based on more than 50 years of experience, cartoons by Ted Williams and Top 10 lists.