The following discussion is part of an occasional series showcasing the ISA Mentor Program, authored by Greg McMillan, industry consultant, author of numerous process control books, 2010 ISA Life Achievement Award recipient, and retired Senior Fellow from Solutia, Inc. (now Eastman Chemical). Greg will be posting questions and responses from the ISA Mentor Program, with contributions from program participants.
Greg McMillan’s Question
What is your personal wish list of prospective achievements in process automation? What are the future capabilities, practices, and management needed to make this a reality and to make your job and the plant more productive?
Greg McMillan’s Answer
Personal: I hope that people would take the time to read my books. I am willing to a send free copy of a specific book to any address in North America.
Better Practices and Management: See the Control Talk columns, “The concealed PID revealed parts 1-5” and the ISA TR5.9 PID Algorithms and Performance technical report for the knowledge needed for many of these practices:
- External-reset feedback (ERF) should be more widely understood and used in proportional–integral–derivative (PID) to inherently prevent oscillations from slow valves, slow secondary loops and valve resolution, split range points, valve input or measurement output signal interruption, and to provide better ratio control coordination and intelligent optimization by directional move suppression.
- The two degrees of freedom PID structure (2DOF) should be more widely understood and used to provide the desired setpoint response after PID is first tuned to provide the load response needed, which is done by momentarily putting controller in manual and making a small output change.
- Specification sheets for control valves should include the max resolution, max lost motion, max 86% response time (T86) and minimum and maximum test step sizes per guidance and examples in the latest draft of ISA TR75.25.02 Annex A.
- Users and suppliers should seek to minimize the total loop dead time from transportation delays, sensor lags, automation component update rates, and filters.
- Users, suppliers, and academia should realize that process disturbances enter as process inputs and not process outputs, and consequently realize the value of large process time constants to slow down disturbances and the corresponding benefit of high PID gains.
- There should be a greater use of procedural automation to automate startups and proactively deal with transition and abnormal operation.
- There should be more access to software and education for dynamic simulation and digital twin to explore, develop, prototype, and continuously improve innovations in process control.
- Practitioners should hire, train, and mentor to reduce stress and workload.
- Users and suppliers should give practitioners at least 10 hours each month to seek new knowledge, explore and implement innovations, and provide a path of advancement based on technical achievements and innovation to offer higher grade levels (e.g., senior and distinguished technologist or fellow) with corresponding pay increases to encourage practitioners to pursue a technical career rather than a managerial career.
- Practitioners should be encouraged to publish and present generic knowledge gained.
My number one wish list item is the preservation of knowledge, manufacturing, and automation engineering jobs in the U.S.
- One of the things that I am seeing is a gap in technical understanding between executives and engineering on what artificial intelligence (AI) and machine learning (ML) can actually do and when they should be applied.
- There seems to be a continued trend in offshoring engineering to use resources that do not have the technical depth or industry experience required to do the work. This isn’t just happening at integration or service vendors. Even large leading petrochemical companies are moving their specialty engineering centers to China. Long-term, this is creating many vulnerabilities for our future. I wish Dick Morley was still around for us to get his take on what is really happening.
Hector Torres’ Answer
External Reset Feedback and 2DOF
I find interesting it previous wish list items include the wider use and understanding of external reset feedback as a tool to improve control. I am in the same boat. I have been recently studying this feature’s performance on a split range temperature control in a batch application. Overshoot reduced drastically while reaching setpoints at an adequate pace by enabling the external reset feedback in a PID controller and inserting a deadtime block between the output splitter BKCAL_OUT and the BKCAL_IN at the PID’s input. When setting the deadtime block at three to four times the open loop deadtime, the overshoot and rise time behave the best.
I am sure that there are different process needs and applications that can benefit from the external reset feedback feature. I would like to keep applying it and continue learning. I would like to understand how the internal control algorithm processes the delayed BKCAL_OUT, how it uses it to calculate the integral error, and determine the controller’s output.
Using a 2DOF PID structure would have been an option to explore as well. I understand that 2DOF also offers a fast approach to setpoint at minimum overshoot. This would be another wish, getting to understand and apply this concept and add it to my toolbox.
Smart PID and Decoupling Techniques
I have read literature around smart PID controllers proposedly able to effectively deal with fluctuating process variable update rates from offline analyzers or wireless transmitters. A couple of months ago, I implemented a control application where I had to program a good amount of code, first to ensure only genuine and timely values were processed, and then to calculate the controller’s output. The algorithm included calculations to level out the integral action based on the time between process variable updates. The algorithm had to calculate the output for two highly coupled controllers and thus, the algorithm included the ideal conventional decoupling technique.
I would like to have the opportunity to learn in detail how the smart PID feature works and how to commission it. I wonder if enabling feedforward on both would be as effective as decoupling these. If I could implement it in future applications, it would save a good number of hours in code and configuration.
Another potential use of this smart PID could be the control of quality parameters. In many manufacturing sites, lab results are normally used by the operators to manually adjust and correct process conditions to maintain evaluated parameters into specifications. Despite this, lab data might have a defined frequency of sampling; the time in between evaluations would still vary. Having a PID that automatically adjusts process conditions based off these unregular lab data would reduce off-grade, relieve the operators from updating statistical process control (SPC) charts and reacting to tendencies, and provide the customer with more consistent products.
In terms of decoupling techniques, I would like to better understand how to calculate the different gains and how to estimate the feedforward gains. I feel that there is a lack of understanding in how to accomplish this. The literature I have read elaborates on general concepts but does not provide one with practical and detailed exercises to completely grasp the concept.
Noisy Signal Filtering
I would like to learn the different techniques of filtering for noisy signals and find one that comes at low impact on the control performance. If the control system comes with well documented and built-in filter options, it would allow the practitioner to select in an easier manner the one that is suitable for the application right off the box.
Some systems would have options like exponential filtering or even dynamic filtering. The foremost generates a weighted average of the current raw and past filtered measurement values. The dynamic filtering also generates a weighted average of the current and past measurement values. However, the filter factor is not static, and instead it adapts to the level of process variation in each measurement.
Some systems use what it is called “adaptive tuning.” This is basically the lambda tuning concept. However, instead of having a fixed lambda factor, the algorithm adjusts it based on the error at the moment. When the error is within a given dead band, the lambda factor is increased to its maximum value, making the controller action almost dormant. Conversely, when the error is high, the lambda factor is taken to its minimum. Thus, the controller becomes pugnacious.
I would like to try this concept and see how it works for different applications. In a way, this concept might be similar to what other systems offer where the reset action is suspended while the error stays within certain limits. However, the adaptive tuning concept also affects the proportional portion of the control algorithm. I wonder if the gain scheduler option offered in some control systems would give a similar response. This would be another feature I would like to gain experience on.
Peter Morgan’s Answer
I have had the good fortune to be engaged in industries that valued, promoted, and rewarded expertise in the field of process control and automation. For those entering the field now, staying the course in engineering through enquiry, continuous learning, and practice is discouraged by corporate polices which do not offer advancement on the technical ladder within the organization. This, in the long run, disadvantages both the organization and the employee. To develop and retain in-house expertise, corporations need to create opportunities for recognition and career advancement as automation experts.
The ISA 5.9 committee was established in 2019 to develop a technical report which, for the first time in automation history, would provide comprehensive descriptions of PID algorithm forms, their implementation, and application to process control. The fact that there's been no standard convention for naming each of the algorithm forms has led to confusion and sometimes heated debate, when the adopted form name suggests the property of the algorithm. Adding to the confusion is that the same algorithm can have different names, suggesting a property that, depending on interpretation, can be contradictory. The terms “interacting” and “non-interacting” are prime offenders in this category. The ISA 5.9 technical report, now undergoing its final review, establishes a standard for PID algorithm form names which I hope will be adopted by the industry.
Contrary to the expression, “You can’t teach an old dog new tricks,” I have recently learned the value of implementing the PID algorithm by using filtered positive feedback to provide integral action. This method, used in pneumatic controllers (out of necessity) and easily implemented in today’s digital systems, was recognized by Greg Shinskey as providing an opportunity to improve closed loop performance and stability when external reset feedback is used. Greg McMillan has already mentioned some of the benefits of this implementation. The method also simplifies the implementation of override logic when two or more controllers act through a signal selector. Disappointingly, few automation suppliers provide this implementation of the PID algorithm as an option, a situation that I would like to see changed.