Principles and Practice in Process Control

Principles and Practice in Process Control – updated 2020-09-26

I frequently taught the process control course to chemical engineering seniors and graduate students, and versions of the course to those in industrial practice.  There are some good textbooks on the topics, but I found that the use of differential equations and Laplace and z-transforms, which satisfy a professor’s sense of what is important, misdirected student understanding from the principles to the glorious and perfect (but only when idealized) abstraction of math.   So, I supplemented the textbooks with my own lectures to clarify the principles and reveal the practice.  I have now used several of those lecture supplements as articles in the Develop Your Potential series in CONTROL magazine, and am grateful that I can place versions here.  Hopefully, this series of articles will reveal some of the mysteries to the visitor. 

Use the links below this list to access articles on:

  1. Process Control is Inventory Control – This is a perspective to understand what to control and how to pair manipulated and controlled variables.
  2. Understanding the P, I, and D, of a PID controller – This is an intuitive development of the PID algorithm to understand what each term does.  It has been one of the most popular articles for Control Global.
  3. Tuning PID controllers – This is a guide to using heuristics to direct tuning procedures, which are faster and surer than methods grounded in fancy math.
  4. Filtering – This reveals conventional methods to remove random noise and spurious events.
  5. Laplace Transforms – This is the conventional language for communicating control techniques in nearly all vendor manuals and bulletins.  You do need to understand what it means, but don’t need to use calculus to do so.  Researchers, of course, find the mathematical convenience of Laplace transforms to be a convenience in analyzing and proving their important concerns; but, the concerns of those in the practice are different.  Engineering-Mathematicians enjoy analytically inverting the transforms, but I find that to be a tedious distraction from practice relevance, and only applicable to a few trivial influence patterns.  This article explains Laplace transforms, shows how to interpret them, and shows how to reconstruct the implementation code from them for any input pattern.
  6. FOPDT Modeling – Many control techniques are grounded in a First-Order-Plus-Deadtime (FOPDT) model of the process response to the controller or disturbance.  Accordingly, obtaining a FOPDT model is one step in implementing a technique.  The conventional “reaction curve” approach to a single step-and-hold in the MV may have been best practice in the 1940’s pre-computer era of Zeigler and Nichols, but I think that nonlinear regression to a skyline input is best practice today.
  7. Calculations with Measurement Signals – This article is about converting scaled transmission signals back to the measurement and coping with issues related to noise, digital discretization, and calibration options.
  8. Orifice Calibration – This article suggests that in installations that are non-compliant to ISO standards, a power-law model of the orifice dP to Flow Rate relation is better than the ideal square root model.
  9. Dimensionless Group Models – This empirical modeling approach is often overlooked when a process engineer has been trained to use regression software and power series models. In my experience dimensionless group models (using Reynolds, Nusselt, or Prandtl numbers, and friction factor as variables) are usually better in several ways.
  10. APC Maintenance – Model-Predictive Control has large economic justification, but the dynamic models loose fidelity as the process ages or is changed, or as operating conditions change. Depending on the situation, the models may need to be re-calibrated on a 6- to 24-month schedule. This needs to be included in the capital planning and in the plant schedule and budget.
  11. Artificial Intelligence – It is not magic nor is it a sentient machine. Using linguistic rules or networks can solve problems that traditional regression modeling cannot. But, understand how to properly use the AI approach. Each modeling approach has its unique attributes and procedures.
  12. Valve Characteristics – Why do we need equal-percent valves?
  13. Output Linearization (output characterization) – how to transform the controller output so that the loop response is linear.