Seasoned Greg: There is so much I have learned that was never taught in university and even industry courses. Much of what I learned, often the hard way, is still not widely understood. I have tried to pass on this knowledge in my articles and books, but often the critical concepts do not stand out. Plus, the practitioner is overloaded and the literature intimidating—especially since project schedules and budgets make free time a premium. Here I consider what if I had the chance to tell a newbie Greg today who wanted take an optimum path to maximizing his technical career.
Newbie Greg: How can I accelerate my learning of what is really needed?
Seasoned Greg: Seek out and talk to experienced practitioners, keeping an open mind and realizing there are assumptions and exceptions because of the diversity of applications.
Newbie Greg: Where do I find them?
Seasoned Greg: Besides automation engineers, talk to process, mechanical, instrument and maintenance engineers and technicians. Find technical experts at suppliers and talk directly with them. I think you may be surprised at how willing they are to talk to you. Everyone generally likes to share technical expertise. Just start with a reasonably good question and follow up with questions showing you are truly exploring their answers as time goes on. If you are actually working on automation systems in the process industry, you can join the ISA Mentor Program to gain the advantage of getting answers to your questions and those by other members posted on my ISA Interchange site for the Mentor Program blogs at https://blog.isa.org/author/greg-mcmillan
Newbie Greg: How can I increase recognition?
Seasoned Greg: Document and present what you have learned and accomplished internally and generically externally. Articles and presentations by practitioners are greatly needed. Engineers in marketing and sales touting their products and professors touting their research tend to dominate publications and conferences. Writing and presenting increases your communication skills and helps to spread unbiased knowledge of what is truly important. Learn to explain your accomplishments in terms that managers without automation experience can understand. Develop metrics to show the benefits of what you have accomplished.
Newbie Greg: What do I do about mistakes?
Seasoned Greg: Be up front about them, telling all affected and finding out what led to the mistakes, how to avoid them, and possible ramifications to other applications. You tend to learn the most from your mistakes. I learned to use simulation where you can make mistakes without consequences and gain a deeper understanding through experimentation that is not possible in the plant making mistakes in the actual plant far less likely as noted in the Control article “Simulation enhances career performance”.
Newbie Greg: How do I avoid distractions and realize what is truly important?
Seasoned Greg: This is quite a challenge because of the lack of practitioner-based practical articles. Further, the increasing desire for the next big thing and almost magical solutions have led to the information technology (IT) perception that more data is the solution, as demonstrated in the Industrial Internet of Things (IIoT) and Industry 4.0 movements. What is critical is getting the best measurements and final control elements (e.g., valves and variable frequency drives) and using feedback control supplemented by feedforward, ratio, and override control for PID control and the possible advancement to model predictive control (MPC) for multivariable control and optimization.
If measurement accuracy and, even more importantly, repeatability are poor, or there is no feedback controller to make corrections for unknowns, or the final control element is not precise, you are out of luck. More and more data points are worthless unless they can be used to improve (not functionally replace) the automation system. I have not seen IT articles understand the transfer of variability from the controlled to manipulated variable, limit cycles, window of allowable PID gains, and the critical impact of the 5Rs (Rangeability, Resolution, Repeatability, Response Time and Reliability) as discussed in the Control Talk blog “What is Truly Important for Measurements and Valves.” Back in the 1980s and 1990s, colleagues of mine wasted years of engineering and millions of dollars on software devoted to expert systems, fuzzy logic, and neural networks. Fortunately, I only had a passing interest and spent a few months on using fuzzy logic to control a multistage waste treatment pH system. It worked, but operators and automation engineers didn’t understand what was going on. I helped get it replaced with MPC that could be better monitored and improved over time. There are many reasons why PID and MPC have been the prevalent, most widely used control technologies for the past 50 years as noted in the Control Talk blog “Keys to Successful Process Control Technologies.”
Newbie Greg: What are some good sources of knowledge?
Seasoned Greg: The ISA Guide to the Automation Body of Knowledge Third Edition provides a basic understanding of almost every aspect of an automation system design. The ISA 101 Tips for a Successful Automation Career, inspired by the Mentor Program, provides concise career and technical guidance. The McGraw-Hill Process/Industrial Instruments and Controls Handbook Sixth Edition provides essential knowledge and best practices for achieving the best automation system performance.
Newbie Greg: What are some misleading concepts?
Seasoned Greg: The biggest one is the prevalent academic idea that PID works in engineering units whereas nearly all PIDs in the process industry work with signals in percent of measurement scale and final control element range. PID tuning is dramatically affected unless the engineering units’ scales are all 0 to 100. Aggravating this tuning problem is that academic PIDs tend to use the Parallel Form with an integral gain and derivative gain whereas industry PIDs in DCSs nearly all use the Series or Standard Form with an integral time (reset time or its inverse) and a derivative time (rate time).
Another misleading concept with widespread implications is the idea—seen particularly in academic publications originating in electrical engineering and systems engineering departments—that disturbances are on the process output. Process disturbances, most notably load disturbances (e.g., changes in stream flow or composition), are process inputs. There are disturbances on the process output but these are mostly due to poor mixing or other forms of noise. This misconception has led to tests and tuning for setpoint response by changing PID setpoints instead of load response by momentarily changing the PID outputs and then using a setpoint lead-lag or a PID structure to get the desired setpoint response.
The literature generally associates oscillations with too high of a PID gain. The more prevalent cause of severe oscillations in important loops, such as composition and temperature control on volumes, is too low of a PID gain. These loops have near or true integrating or even runaway response requiring aggressive proportional action to make up for the lack of process self-regulation. Too low of a PID gain causes very large and very slow oscillations that are not effectively filtered by downstream volumes. For runaway reactions, a point of no return can be created by excessive acceleration resulting in a safety-instrumented-system trip and the blowing of relief valves. Additionally, the amplitude of limit cycles caused by valve backlash and poor positioner sensitivity is larger for a smaller PID gain.
Variable frequency drives (VFDs) are often cited as being faster, more linear and precise than control valves. Not realized is that a system pressure drop that is appreciably less than 25% of the system static pressure can cause a quick-opening, installed-flow characteristic. Standard inverter I/O cards offer only a 0.35% resolution, and the deadband and speed-rate-limiting settings used are often excessive.
Dead-time dominance (total loop dead time greater than open loop time constant) is often stated as a reason to go to MPC. A better solution is feedforward control and, if necessary, dead-time compensation—preferably by simply inserting a dead-time block in the external-reset feedback path with block dead time updated to match total loop dead time. Not understood is that both MPC and dead-time compensators are more sensitive to an overestimate than an underestimate of the total loop dead time for dead-time-dominant loops. Besides an accurate estimate of dead time, the reset time of the PID must be greatly decreased to see the benefit of dead-time compensation. The benefit from dead-time compensation is actually greater for lag-dominant processes but since these can have a high PID gain, the need for dead-time compensation is generally not pursued.
Advocates of control algorithms often tune the PID to prove their point not as a deliberate deception but largely due to a failure to recognize that loop performance depends on PID tuning and there is a tradeoff between performance and robustness to changes and nonlinearities. Tests of an algorithm supposedly better than PID often show how much smoother the response would be than a PID with aggressive tuning (e.g., Ziegler-Nichols). Not shown is that halving or quartering the PID gain could give similar results. Also, tests are often for disturbances on the process output for self-regulating processes where the total dead time is greater than a quarter of the open loop time constant (e.g., not lag-dominant). Proponents of wireless measurements may show the consequence of a slower wireless update rate as inconsequential by much less aggressive tuning that corresponds to a greater dead time from the slower update rate.
Control valve rangeability is not often cited but when it is stated, the rangeability is often the ratio of the maximum controllable flow to minimum controllable flow. These flows often do not take into account the installed flow characteristic and resolution limit due to stiction that is particularly large near the closed position. Linear valves are sometimes cited as offering the best rangeability based on the inherent flow characteristic, not realizing a valve pressure drop appreciably less than 25% of the system pressure drop causes linear flow characteristic to distort to quick opening, making the minimum controllable flow for a signal corresponding to resolution near the closed position much larger. Also, valves are often sized larger than needed, making the valve pressure drop less and the distortion greater. Additionally, the limit cycles from resolution are larger, since the resolution is a percentage of valve capacity. Larger is not better. We would be much better off if valve specifications had an entry for allowable backlash and stiction at normal, maximum and most importantly minimum operating position. Right now, the emphasis on minimizing leakage increases the stiction and resolution near the closed position and the emphasis on capacity and cost increases the backlash and resolution in terms of the change in flow and causes greater distortion of the inherent flow characteristic by decreasing valve pressure drop.
Control valve 86% response time is often not stated for small and large changes in signal. The 86% response time due to poor positioner or actuator sensitivity can increase by an order of magnitude for small signal changes (e.g., < 0.25%). The 86% response time can also increase by an order of magnitude for large signal changes (e.g., > 25%) due to slow slewing rate caused by a large actuator.
There is a rule dating back 50 years that positioners should not be used on fast loops and that a volume booster be used instead of a positioner. Positioners must be used on piston actuators. Boosters used instead of positioners on diaphragm actuators can cause a butterfly valve to slam shut. Plus, the resolution and backlash is worse from the omission of the positioner. For valves that need to be faster, a volume booster is put on the positioner output with a booster bypass valve slightly open.
10. PID algorithms work in percent signals instead of engineering units
9. Most loops don’t use the PID Parallel Form
8. Most loops are more affected by automation system dynamics and PID tuning than process dynamics
7. Most process disturbances are on the process input
6. You can’t do better than a PID for unmeasured disturbances on process input
5. You can prove anything you want by how you tune a PID
4. More problems are caused by too low a PID gain than too high a PID gain
3. You don’t need to tune for a setpoint change
2. If your automation system is doing poorly in the 5Rs, you are out of luck
1. Without dead time, we would be out of a job