Invisibility of process control

Greg: Specialists doing process control improvements are doing a disappearing act. While the biggest loss of expertise is in application of advanced process control (APC), the efforts to improve control strategies, tuning, measurements and valves also are suffering. All of this comes at a time when tools and technologies offer the greatest flexibility and capability. The symptoms are the disappearance of process control improvements, copy jobs in migration projects, and the expectation that solutions simply lie in extensive use of the Industrial Internet of Things (IIoT). A tempting thought for management trying to reduce staff and budgets is that IIoT will tell you everything you need to know and do about the process. The real case is that there should be a synergy between intelligence offered by automation professionals and the availability, visualization, integration and coordination of data offered by IIoT that leads to more extensive use and appreciation of people doing process control.

Stan: Sigifredo Nino, a protégé of Greg Shinskey featured in two previous Control Talk columns, Distilled Analysis of Interaction, and “How to Get the Most Out of Your PID by Better Tuning," is similarly concerned about the lack of appreciation of process control skills and accomplishments. Our follow-up column, “Tuning Finale,” with Mark Coughran, sought to help eliminate some of the confusion about tuning that's a contributing factor.

Sigifredo: The problem is perhaps best summarized by Karl Åström in his statement, “Control has become a hidden technology.” What he wanted to highlight is the fact that if all the controls work fine, nobody notices their existence, and that people have always been accustomed to defining control as devices and equipment rather than ideas. More attention should be paid to popularizing what control is. I personally experienced this when I saw that the benefits of $2.5 million per year for an improvement in a distillation tower top product recovery control did not get the attention deserved. An award for greatest achievement that year was given for replacing process equipment (e.g. heat exchanger), even though the return on investment (ROI) was much less than for the improvement in the distillation tower controls. Plant engineers, operators and managers can physically see and understand new process equipment. The same can't be said for process control. Process engineers and operators are more focused on steady state and whether the process is running. Not appreciated is the need to move process inputs and outputs for a more productive plant.

Greg: Process support and operations personnel like the idea of setting constant flows, and are basically uncomfortable with a process control system that they never did really learn conceptually or practically being given the freedom to manipulate process or utility flows. This is seen in flows being set manually or to a preconfigured value during transitions and for batch operations. Not realized is that a process controller is needed to immediately deal with the inevitable unknowns and disturbances, and that flows exactly matching a process flow diagram (PFD) developed during process design is a rare coincidence, and may not be optimum for current objectives and operating conditions.

Stan: The remedy to this problem should start in university courses.

Sigifredo: My experience from being on the firing line is that engineers at the plants don't have a working understanding of what process control is, nor what can be accomplished by using it. The emphasis at the universities was always on developing new control algorithms, even though it's been proven that PID can achieve optimal unmeasured disturbance rejection and setpoint response if one learns how to tune it, and effectively use the many features, as documented by Shinskey in his many articles and books. It may not even be possible to sell to universities the idea of practical courses on what the engineer needs to know to at least understand and see the value of process control as practiced in industry. Students are basically taught how to learn with a PFD mindset. Academics generally like to talk to academics and publish in academic journals. Books by Shinskey are not used or appreciated. There are notable exceptions to the lack of university focus on how to get the most out of a PID, as seen at the University of the Lund in Sweden, and the book “Advanced PID Control” by Karl Åström and Tore Hagglund.

Greg: I appreciated the use of ordinary differential equations (ODE) in the book, “Process Dynamics and Control” by Seborg, Edgar, Mellichamp and Doyle, as the starting point for understanding process dynamics. I particularly found valuable the papers by Bill Luyben for understanding how thermal lags and deadtime can close the window of allowable PID gains in an open-loop, unstable, highly exothermic reactor, and how a recycle stream can cause a snowballing effect in reactors in general. Then, of course, we have Cecil Smith, who went from university professor to industry consultant, focusing on the time domain and PID control, and publishing practical books and articles.

When I took over teaching a course on modeling and control at Washington University in St. Louis, it was totally based on using Laplace Transforms and Z Transforms. I decided to cover the basics of these transforms in a one-hour lecture, then move onto to the use of PID, focusing on the time domain, and ending the course with an introduction to model predictive control (MPC). A virtual plant was used in a weekly laboratory for applying PID and MPC strategies. The students really got into the functionality and value of process control, which are important in whatever job they might have because the control system is the window into the process and the means of making changes to increase process efficiency and capacity. To my dismay, the other professors did not see the necessary value, and decided the course should be an elective.

I think there is some importance to understanding and using the frequency domain. I've used Bode plots to understand resonance where the oscillations of one loop are amplified by another loop being in automatic, and the effect of filters on oscillations as discussed in Control Talk's December 2014 blog entries, “Controller Attenuation and Resonance Tips ” and “Measurement Attenuation and Deception Tips .” I also used Bode plots for self-regulating processes and Nyquist plots for integrating and runaway processes for a deeper understanding and development of an estimation of the effect of deadtime and secondary lags on the ultimate period and ultimate gain. This last aspect is not something I'd expect 99% of process control engineers to be interested in or need to develop skills in. I think the more useful approach is the time domain, since the time response is what we see in trend plots, and we can get the values of process time constants from the ODE for material and energy balances, as seen in Controlglobal.com's online paper, “First Principle Relationships for Process Dynamics .”

The problem with the Laplace transform analysis is the simplification needed and the consequential omission of complex valve dynamics, as discussed in Control Talk's March 2017 entry, “How to get the most out of valve positioners” and Control's May 2016 article, “How to specify valves and positioners that do not compromise control .” The dynamics of sensors are also complex depending on the direction of the change, fluid velocity, installation detail, location and coatings, whose consequences are noted in the Control Talk's February 2017 column, “26 things to not do in temperature, flow, valve, differential pressure and pH applications .”

Stan: How is IIoT seen as the solution?

Sigifredo: Unfortunately, IIoT is seen by some as the total solution, even to the point of reducing the need for process control engineers. There are positive aspects of IIoT that could increase the remote availability of data and remote guidance by specialists that could actually increase the visibility and appreciation of process control. Unfortunately, what's often emphasized is that simply dumping data to a cloud can result in software telling you everything you need to know, not realizing it can possibly overload you with what you don’t want to know. This overload is particularly the case if there isn't significant user intelligence by engineers, who understand the process and control system.

Greg: Given sufficient data security, assurance, range, validity and immunity from interference, as noted in the article, “Master the IIoT,” the addition of user intelligence can enable coordination of maintenance with operations to be part of the solution instead of part of the problem, which is further discussed with Brian Hrankowsky in “Drowning in Data, Starving for Information 4."

[sidebar id =1]

Data analytics is the most powerful tool commonly cited to determine if a process is behaving differently and provide key inferential measurements such as product composition. Not realized is that data analytics identifies correlations, and not causes and effects. The training sets need to be intelligently selected to include operating ranges and desired control system modes, and eliminate outliers. There really needs to be a design of experiments (DOE) to better identify the full range of possibilities. The results need to be analyzed based on an understanding of the process, equipment and control modes often best achieved by a high fidelity dynamic simulation in a virtual plant. Application of data analytics to continuous processes requires dynamic compensation to synchronize process inputs with process outputs. The task is comparable to identifying the deadtime and time constants in the process and automation system for feedforward dynamic compensation. The use of tuning software to do this identification for a single feedforward input is not well understood and is time-consuming. Imagine trying to do this for all the possible inputs for continuous data analytics, particularly considering some inputs may not truly be cause-and-effect. This is difficult for a single unit operation and nearly insurmountable for the total process. For batch processes, dynamic compensation isn't needed for detecting batch abnormalities and predicting batch end points. Consequently, I see significant value in data analytics for batch chemical reactors, and an even greater value for bioreactors for biologics due to the complexity of the mammalian cell response of batch data analytics that use Hadoop data processing clusters in a private cloud. However, for fed-batch operation, sometimes called semi-continuous operation, some dynamic compensation may be needed for analysis and prediction of process variables as a batch progresses.

The computation of online metrics for process efficiency expressed as a ratio of a product flow to a raw material or to a utility flow for continuous processes needs dynamic compensation to prevent short-term irregular responses. The computation of ratios using running totals of a product and raw material or utility over the last shift, day or week can reduce the dynamic compensation needed, similar to the totalization effect in a batch operation. The largely unrecognized challenges of online metrics have resulted in rather limited use, and more reliance on audits for documentation of process control benefits.

A virtual plant in a private cloud offers the needed experimentation that couldn't be done in an actual plant. This allows the discovery, development, prototyping, demonstration, testing and training of process control improvements by all types of technologies, including online metrics, showing the benefits in terms of process efficiency and capacity.

Stan: What about using the cloud for doing closed-loop control?

Sigifredo: The speed, reliability and predictability of the data rate for mesh networks aren't adequate for many control loops. Even more troublesome is the concept that a program in the cloud can readily replace all the intelligence built into automation systems for continuous and batch control of industrial processes.

Greg: I'd hope that any attempt to extensively implement process control in the cloud would include the best of what industrial system suppliers have learned, and what's been developed by way of ISA standards and technical reports. Personally, I'd hope the PID controller has the positive feedback implementation of integral action enabling external-reset feedback. The enhanced PID gained from external-reset feedback is so important for systems where the process variable originates from a wireless device or an at-line or offline analyzer, where the time interval between updates is greater than the process response time. For more on this enhanced PID, see Control Talk's July 16, 2015, entry, “Batch and Continuous Control with At-Line and Offline Analyzer Tips .” For much more about deadtime-dominant processes, see Control Talk's Dec. 1, 2016, entry, “Deadtime Dominance—Sources, Consequences and Solutions .”

Stan: There are some IIoT applications for making remote device data available for use by a DCS or PLC for control in long pipelines and distant tank farms. Technically, if nodes are used rather than a mesh network, keeping the update time less than 20% of the total loop deadtime should prevent additional delays from appreciably degrading control.

Greg: There are also potential advantages to be gained by using portable wireless devices for detecting process conditions, such as fouling and demonstrating predicted performance (e.g., batch reactor cycle time and endpoint by the simple use of measurements of coolant into and out of a reactor coil or jacket).

Some questions that remain to be addressed in the extensive use of IIoT are: Is the process industry ready for IIoT? Are we willing to trust wireless? Are we ready to use cloud-based Layer 3+ equipment (operator stations, historian, etc.)? Are we ready to embrace IT security and create a security model that works for the process industry? Are we willing to collaborate with other plants to share sensor data? Can IIoT standardization drive down greenfield construction cost and/or brownfield operational costs?

There's some negativity about the term “advanced” as discussed with ISA Mentor Program participants in the Control Talk's July 2015 column, “Why all the hatin’ on advanced control ?” The attitudes of people toward APC who are not well versed in APC, as noted by Brain Hrankowsky, are "It’s too complicated," "No one will understand," "No one else will be able to support it," "It will cost a lot and be hard to justify" and "We've gone this long without needing advanced controls, so we don't need it." My subsequent thought was, "Why not call it 'smart process control?' and ask if users want 'dumb process control.' ” There remains the principal concern that APC may be sought as the total solution by adding more sophisticated software, overlooking the need to improve the foundation (e.g., control valves, measurements and PIDs) on which APC depends. In a way, looking to more sophisticated higher-level software is also indicative of the dubious idea that IIoT is the total solution, decreasing the need for automation and process control professionals.

What's needed is for practitioners to publish and to offer webinars to increase the visibility of process control. Since the bottom line is what management best understands, we must promote the extensive use of key performance indicators (KPI) centered on process efficiency and capacity that can show the “before” and “after” cases. These metrics and the control strategies to maximize them can be developed in a virtual plant as discussed in “Getting innovation back into process control.” IIoT may not have the dynamic compensation needed for short-term KPI indication, but it can possibly be used to provide long-term KPI trend and trajectory knowledge.

Top 10 songs to brighten your day by promoting positivity and creativity