How does control using first-principles model controllers differ from other controller types being used? Why are they better?
Most controllers, including model-based controllers, are linear and locally valid. They are tuned or calibrated for one operating region. But when tank levels or flow rates change, the process gain, time-constant and dead times all change. This requires retuning or recalibration, which can be expensive and time-consuming. First-principles, model-based controllers will work properly throughout the entire operating range. One benefit is that once tuned, they remain tuned until substantial physical changes occur in the process.Another benefit is that the first-principles modeling effort has the process engineer focus on a mechanistic understanding of the process. Knowledge discovered from the modeling effort also benefits process troubleshooting and online analysis, which management would prefer over the engineer focusing on understanding Laplace transform mathematics or methods to calibrate dynamic matrix-type controllers.
Another benefit is that first-principles, model-based control only has one tuning coefficient per controlled variable (CV). Proportional-integral (PI) control has two. First-principles, model-based controllers naturally feed forward the impact of measurable disturbances and naturally decouple interacting controllers. If a process control application is 2X2 with one feedforward variable, the two PI loops would have 4 tuning coefficients, the feedforward impact on each CV would have 8, and a simple static decoupler would have 2 more. That is a total of 14 tuning coefficients. A controller based on first-principles models would have 2. The simplicity of tuning is another advantage.
Finally, supervisory process optimization (RTO) usually uses a steady-state, first-principles model to define set points for the process regulatory controllers. But the low level controllers are derived from linear dynamic models (such as first-order plus dead time [FOPDT]). There are inconsistencies in gain and dynamics, which concern managers about the true optimization and handling of constraints. If the first-principles dynamic models are used for control and the steady-state version of the same model is used for RTO, then the two functions are consistent.
What might be prime opportunities for a controller based on first-principles models?
If a controller must be retuned frequently because the process gain and dynamics continually change, consider a first-principles, model-based controller.
Unit operation equipment often comes with on-board controllers. Here, the cost of designing a first-principles controller could be allocated to many units, and diagnostics from the first-principles models (efficiency, fouling, constraints, etc.) could be very useful to the manufacturer.
Could controllers based on first-principles models be undesired?
- The model would be unique for each application. The controller model and other choices on how the calculations or functions are performed must be clearly documented.
- The control approach is novel, and each operator or process engineer would have to understand how it works.
- If PID control works and maintenance is only a minor headache, use PID.
Does developing the first-principles models take substantial engineering effort?
It could take substantial effort to start from scratch. However, the models may already be available in other software engineers use for design, analysis and optimization. Note that other advanced regulatory or advanced process control (APC) models are not work-free. Consider the process step-testing time and cost to generate all the coefficient values in the FOPDT models for controller tuning, feedforward and decoupling action or the cost to calibrate finite impulse response (FIR) models for APC. Then, consider that these empirical models are only locally valid, which means gain scheduling rules or model calibration will be needed for several operating ranges. There may be a substantial up-front effort to generate first-principles models. However, they have only a few coefficients with values that need to be determined by process data, and they will represent the process over the entire operating range without extensive process testing to generate gain scheduling or multiple models.Is a model-based controller combined with other techniques, or does it replace them?
Model-based controllers could replace existing regulatory and advanced regulatory control structures because these controllers can send their outputs directly to the final control elements. However, it is recommended to keep primitive PID loops on flow rate, temperature and pressure, and to have the supervisory model-based controller send set point signals to the primitive PID controllers. The devices provided for PID control have many auxiliary features (filtering, communication protocol, validity checks, etc.) that one would want to preserve. The model-based supervisory controllers would replace the advanced regulatory control structure (ratio, cascade, feedforward, decouplers, etc.).Who will benefit most from this book?
The intended audience includes practicing engineers who want to solve nonlinearity issues, students in an advanced control course, and R&D folks in vendor organizations who wish to consider providing products that can perform model-based control.Tell us a little about your background and experience with model development.
I started my career in the chemical industry. After 13 years, I switched to academia, where I received numerous recognitions for exploring using first-principles models in control. Grounded in what is practicable, I have investigated and validated methods on several pilot-scale to full-scale processes. The material in the book represents my application views of best-in-class, practicable techniques. My chemical engineering degrees are from the University of Maryland and North Carolina State University. I have served in various ISA leadership positions and as the President of the American Automatic Control Council. Now that I am “retired,” I focus on providing training materials for practicing engineers.How did you get interested in using first-principles models in control?
In the early 1980s, my manufacturing site began upgrading all our analog process control devices with a digital distributed control system (DCS). At that time we were using the company’s new business/payroll computer (when they gave us time on it) to execute FORTRAN (yes, all letters capitalized back then) for modeling, design and statistical algorithms. Also, I bought our first home computer (a Radio Shack 64 K machine using the home television for display, using BASIC for programming, and booting from a 7-inch floppy disk) and created Atari-like games for my children to play. I also programmed the computer for work-related data processing algorithms I had written for my hand-held programmable calculator. It just clicked. We could code first-principles models in the DCS for online, real-time process analysis and perhaps control. My management did not understand the concept. It is fun to claim that I explained it saying, “We could code the engineer’s mind inside the computer.” I was told, “Russ, we don’t invent technology; we buy it when needed to make products.” So, I moved to an academic career where I could explore the possibilities.What key points do you want to highlight for potential readers?
A key theme of the book is that the mathematics for these first-principles, model-based controllers do not require differential equation solutions or transformed variables (Laplace) or empirical models (FOPDT, ABCD, ARMA, Neural Networks, FIR, time series, etc.). The controller model could be the same model that is used in supervisory steady-state optimization, real-time process analysis, training, troubleshooting and process design. The vision is to have one model to supplant all the others. The advantages are that the process engineer does not have to learn all those diverse mathematical methods, and using first-principles models reinforces mechanistic cause-and-effect process understanding. Further, it uses the familiar mathematical methods engineers used in their college courses to reveal engineering principles. Paraphrasing J.R.R. Tolkien, “The one model to rule them all.”Where can readers get supporting information?
Visit my website, www.r3eda.com, to download Excel/VBA simulations on several processes (the code is open, and how-to guides are provided). There are also some application papers.About R. Russell Rhinehart
Dr. R. Russell Rhinehart, professor emeritus in the School of Chemical Engineering at Oklahoma State University, has experience in both industry (13 years) and academia (31 years) and was head of the school for 13 years. Russ is a past president of the American Automatic Control Council, was editor-in-chief of ISA Transactions from 1998 to 2012, and Director of the ISA Automatic Control Systems Division (now Control and Robotics). He is a Fellow of both ISA and AIChE and a Process Automation Hall of Fame inductee. He received the 2009 ISA Distinguished Service Award and the 2013 Fray International Sustainability Award.Inspired by his industrial experience, his mission has been to bridge the gap between industry and academia. Russ was the codirector of two industrial consortia (one at Texas Tech and a second at Oklahoma State) and built pilot-scale laboratories for dual use in undergraduate education and graduate research. He left industry in 1982 with a vision to use engineers’ process models in control and pursued many aspects of doing so in his academic research career. This book is his collection of practicable methods. His goal is for it to be a useful guide to others seeking to use nonlinear models in control.
His 1968 BS in chemical engineering and subsequent MS in nuclear engineering are both from the University of Maryland. His 1985 PhD in chemical engineering is from North Carolina State University.
He maintains a website (www.r3eda.com) to provide open access to software (including simulators to support this text) and technique monographs.
In “retirement,” he offers consulting services related to engineering analysis, and serves on several ISA, American Automatic Control Council (AACC), and International Federation of Automatic Control (IFAC) committees.