Control Strategies

As we get further away from our college years, sometimes misconceptions can solidify. For instance, most engineers that had a basic control theory class may recall that increasing the gain in a feedback loop at some point will introduce instability/oscillations. The misconception is that this is not universally true for all loops, such as level loops.

ModelingAndControl.com's Greg McMillan has a great article co-developed with researchers at India's MIT Anna University on ControlGlobal.com, Adaptive Level Control: Exploring the Complexities of Tuning Level Controllers and How an Adaptive Controller Can Be Used in Level Applications. Greg, Sridhar Dasani and Dr. Prakash Jagadeesan clear up the gain misconception:

...the opposite correction is more likely needed for integrating processes. Most level loops are tuned with a gain below a lower gain limit. We are familiar with the upper gain limit that causes relatively fast oscillations growing in amplitude. We are not so cognizant of the oscillations with a slow period and slow decay caused by too low of a controller gain. The period and decay gets slower as the controller gain is decreased. In other words, if the user sees these oscillations and thinks they are due to too high a controller gain, he or she may decrease the controller gain, making the oscillations worse (more persistent).

The authors describe some challenging level control applications such as continuous reactors, crystallizers, and material balances in unit operations that require extremely tight level control. For process vessels such as horizontal tanks, drums, and spheres, the level change for a given flow rate is not linear because of the geometry of the vessel. Changes in the fluid density and use of non-linear valves can also increase the challenge to perform tight level control.

The authors note that adaptive level controls, built on adaptive control software such as DeltaV InSight, can

...not only account for the effect of vessel geometry, but also deal with the changes in process gain from changes in fluid density and nonlinear valves. Even if these nonlinearities are not significant, the adaptive level control with proper tuning rules removes the confusion of the allowable gain window, and prevents the situation of level loops being tuned with not enough gain and too much reset action.

The article highlights process dynamics related to conical tanks. These tanks have extreme changes in cross sectional area as the level changes. The MIT Anna University research lab used the embedded DeltaV InSight software to automatically identify the process dynamics around changes to the level setpoint within the conical tank. The authors describe how this is done:

The adaptive controller employs an optimal search method with re-centering that finds the process dead time, process time constant, and process gain that best fits the observed response. The trigger for process identification can be a setpoint change or periodic perturbation automatically introduced into the controller output or any manual change in the controller output made by the operator.

The article is complete with equations for integrating process gains, conical tank dynamics, and controller tuning rules. These may help awaken those brain cells if you're like me and have let these cells remain dormant over the past several years.

The authors summarize their findings:

Adaptive level controllers can eliminate tuning problems from the extreme changes in level control dynamics associated with different equipment designs and operating conditions. The integrated tuning rules prevent the user from getting into the confusing situations of upper and lower gain limits and the associated fast and slow oscillations. The smoother and more consistent response allows the user to optimize the speed of the level loop from fast manipulation of column reflux and reactor or crystallizer feed to slow manipulation of surge tank discharge flow control.

MP3 | iTunes

Here's another great presentation from the recent Emerson Exchange from refining & chemicals industry solutions director, Pete Sharpe. You may recall Pete from earlier process optimization-related posts.

His presentation, SmartProcess Distillation Application Improves Recovery and Saves Energy--A Case Study, describes a project to improve operations on a high-purity distillation process at a Goodyear facility. These distillation columns, 11 in total, were part of a purification unit with multiple trains. The ultra-high purity product specifications required very tight quality controls.

From a control strategy standpoint, the process had multiple large, 200+ tray columns with extremely long time constants. Also, different feedstock suppliers provided feed with different qualities. The operators had to have large safety margins to compensate for disturbances caused by the feed variability. These conservative margins reduced the overall recovery rate and increased the energy required per unit output. Overall, the purification unit was a large energy consumer within the plant.

Pete and the APC consultants worked with the engineers to scope a project that included a functional specification that included the design for all 11 columns in the purification unit. The initial implementation phase covered only the first column in the series to:

• Have the Goodyear team gain experience with the technology
• Develop acceptance by the plant operators
• Demonstrate the value of the distillation optimization to the management staff.

The plant engineering staff with support from the Emerson APC consultants would implement subsequent columns.

During the functional design specification phase that included an on-site audit of the installed instrumentation and control strategies, the team identified regulatory control issues including sticking control valves. As I've mentioned in an earlier post, process variability can often be traced back to valve performance problems and these should be addressed first.

After these issues had been addressed, the team installed a SmartProcess Distillation Optimizer that embeds DeltaV model predictive control and neural network function blocks on the first column. The control strategies were based on a "what comes in must go out" approach that included material (overhead to feed ratio) and energy (reflux to feed ratio) balances. The objective of the optimization was to minimize overhead product loss while controlling bottoms impurities to target.

The manipulated variables (MV) for the MPC controller included the distillate rate and reboiler steam. The control variables (CV) included the lower tray pressure-compensated temperature (PCT) and the overhead PCT. The disturbance variables (DV) included the feed flows and reflux temperature. The column constraint variables included measurements from on-line analyzers for overhead product loss and light impurities in the bottoms as well as internal flux rate, column delta pressure and the reflux/distillate ratio.

Pete and the team had to work through reboiler steam temperature instability caused by a process design issue. Also, the overhead pressure controller range was insufficient to handle the difference in temperature between day and night. Two valves controlled the pressure in a condenser system, one large and one small. The pressure swings caused changes in the temperature and quality of the column output. The team developed a control strategy to adjust the large valve to keep the small valve within the correct control range.

A two-week period on-site was required to address these issues, commission the optimizer, and train the operating and engineering staff. The results from this project were impressive. The optimizer immediately started reducing distillate rate and overhead product losses. The average overhead product loss was reduced 22% while the impurities in the bottoms were maintained within specification. Steam usage dropped 7%.

For this one column, this optimization project delivered an estimated annual value of $700,000 (USD) through increased recovery rates and decreased energy usage.

MP3 | iTunes

Update: I've uploaded a copy of Pete's presentation to Slideshare and embedded it within this post. It's also available for download from Slideshare.

Last year I summarized an Emerson Exchange presentation by James Beall, Advanced PID Functions for Improved Control Performance. At this year's Emerson Exchange, he did a reprise version, Interesting and Useful Features of the DeltaV PID Controller. I'll highlight some additional nuggets of wisdom that he imparted.

When it comes to limits being placed on setpoints and for a loop that is in automatic mode, it's usually best to let the operators have total control on the setpoint and not apply setpoint limits to the PID control when it is in the cascade mode. He did add that when it comes to cascaded loops like a level control master PID loop and a flow slave loop, it's a good idea to set limits on the output of the master loop rather than the sepoint of the slave loop, and that there is a lot of flexibility in the DeltaV PID control algorithm to do this.

With cascade control, James noted that mode tracking and bumpless transfer is provided and that limited conditions in the slave loop are taken care of through the BKCAL feature. DeltaV Books Online describes the input and output:

BKCAL_IN is the analog input value and status from a downstream block's BKCAL_OUT output that is used by a block for bumpless transfer. This connection is necessary if the PID is a master to another controller in a cascade. Without the connection, the slave controller will not make the transition to CAS and the master PID will never be active.

BKCAL_OUT is the value and status sent to an upstream block to prevent reset windup and provide bumpless transfer to closed loop control.

In conditions where the slave loop is limited, you can enable PID external reset. You would most often use it in the primary loop of a cascade and have it compensate for unexpected slow secondary-loop response. This is done by selecting the FRSIPIDOPTS "Dynamic Reset Limiting" on the master loop and the CONTROL_OPTS "Use PV for BKCAL_OUT" on the slave loop.

I summarized James' discussion of gain scheduling, which provides up to three regions of different PID tuning with a smooth transition between regions. This year, he described a parameter FRSIPID_OPTS that modifies the proportional gain as a function of the error (process variable, PV minus setpoint, SP). You can use this non-linear gain function to make the tuning more aggressive as the separation between PV and SP increases. It also can be used to create the "error squared" PID function.

James cautioned that using this non-linear gain function on an integrating process, like levels, can cause oscillations at the reduced gain. For these applications, the reset time should be based on the product of gain and the minimum gain modifier (NL_MINMOD), which will result in a larger reset time to prevent oscillations. He suggests using the gain scheduler to provide non-linear tuning on integrating processes.

There's more wisdom shared on valve output characterization, anti-reset windup limits, adaptive control, and loop simulations, which I'll leave with you should you choose look at the embedded presentation.

MP3 | iTunes

For those attending the Emerson Exchange with distillation processes, you don't want to miss Emerson's Lou Heavner's presentation, Advanced Distillation - 102. Lou is an advanced control consultant, whom you may recall from earlier posts. He'll be presenting Wed, 9/30 at 4:15pm in the Sun 4 room and again Thurs, 10/1 at 9:00am in the Captiva 1 room.

Lou's preference is to have each session be an open, interactive discussion on distillation control challenges and solution approaches. He does have a presentation and will talk if that is what the session attendees would prefer, but he's hopeful an interactive discussion will blossom. I've gleaned some highlights from the presentation.

This is a picture of a traditional binary distillation column. The feed may be vapor, liquid, or a mixture, but is most commonly a liquid. Its composition is usually variable. Liquid in the bottom of the column is boiled up through the column and vapor leaving the top is condensed and returned as reflux. The product flows must equal the feed flow or the process won't operate very long. So, the liquid level in the bottom of the column and the liquid level in the reflux accumulator are controlled by manipulating product flows or sometimes by manipulating heat to the reboiler or reflux to the column.

Usually, composition is inferred key from temperatures in the column and controlled by manipulating heat to the reboiler and reflux flow to the column. Pressure can be controlled by venting non-condensibles (if they are present) or by controlling the amount of condensing in the column overhead.

Feed is usually not available for control, but may be in some cases. When it is available, it can be a good choice for optimization--maximizing throughput. In some cases, online analyzers are available and if they are, they may be used for control or simply monitored.

Lou stresses the fact that you can't control something that you can't measure. Online analyzers or product purity measurements are one of the key requirements for good distillation control. If there is not an appropriate online analyzer, then some kind of inferential measurement will be required. Other measurements such as flooding (a column operating constraint) and reflux ratio can be used to track performance.

Interaction is one of the defining challenges of distillation control. Interestingly, there are many ways to pair controlled and manipulated variables. Some will work well in one column and poorly in another. The whole study of relative-gain array (RGA) analysis has been developed to understand the best way to pair control and manipulated variables. This is largely dependent on factors like product purity specs, feed composition, number of theoretical stages in the column and typical reflux ratio.

Another source of interaction outside of many columns is thermal integration. It is common for the hot product to be cooled against the feed to provide some preheat and efficiency to the column. Variability in the bottom product temperature or flow rate will be recycled back into the column through the feed. Sometimes a heat pump arrangement is used to boil the bottoms against the overhead vapors, which have been compressed. This configuration is rarely seen and only practical when the overhead and bottom products have similar boiling points and non-condensables are not present.

Another factor that makes distillation control difficult is the actual process dynamics such as long time delays associated with the time it takes for liquid reflux to cascade all of the way down to the bottom of the tower. This is more problematic when one considers that vapors will rise much more quickly up the tower. The controls need to respond to both of these kinds of dynamic responses.

When looking at the distillation process, it's a classic multi-variable process with controlled and manipulated variables. You can include the material balance loops (i.e. the level control loops) in the model predictive control (MPC) strategy. The levels would be the control (or constraint) variables and the product flows would be the manipulated variables. There are extra, manipulated variables, so an opportunity exists to include optimization. There may also be additional constraints (e.g. flooding or valve positions) and measured disturbances (e.g. reflux temperature and feed temperature).

Lou goes on to describe interaction and the relative gain array analysis process to identify the best pairings to minimize the effect of loop interaction and simplifying the matrices. He also covers pressure compensated temperature pros and cons, multi-component distillation, level control, azeotropes, and batch distillation.

If you've been battling distillation issues in meeting quality specs, energy usage, yield, and/or capacity, bring these to one of the two sessions and see what thoughts Lou and fellow attendees have to offer. It will hopefully be worth your while!

Update: Here's Lou in action:

A great question came in the Process Automation Usability Project forum asking why there is not more control in the field being done. If you're unfamiliar with the term, it's where a Foundation fieldbus (FF) transmitter, digital valve controller, or other instrument runs the control logic for the loop, instead of this being done in the automation system controller.

When I saw the question, I bounced it over to Emerson fieldbus consultant Dan Daugherty, whom you may recall from earlier fieldbus-related posts. Dan made a couple of great points, the first on segment design:

If you do control-in-controller, and you don't care a whole lot about process segregation (only do it on a coarse, controller level), then you can skip all the concerns regarding segment design except for voltage drop and number of devices. Many engineering companies didn't want to learn segment design, so they stuck with control-in-controller.

His second point dealt with the complexity of the control strategy used:

...some control strategies are complex enough they can't be done on a single segment (prior to 2007) because not all the control functions they wanted to use were available in devices. That is no longer true with Emerson because we have ability to pick up functions in the H1 card, effectively filling out any gaps that used to exist in the control-in-field palette of functions.

His third point is that there are economic and performance advantages to using control in the field. In his response, he references a presentation that he and some colleagues are giving at the upcoming Sep 28-Oct 2 Emerson Exchange conference, Effects of Macrocycle Time and Sampling Rates on Control Loop Performance.

The team set up controlled tests in Emerson's Marshalltown Flow Lab to compare performance between Foundation fieldbus and conventional 4-20mA analog loops for control response period, load frequency response, and setpoint step response.

Their tests showed that with FF control in the field, the control response period equaled the macrocycle and they could get 0.18 seconds, which is adequate for almost all loops. The exceptional loops that require faster response times, such as surge control and compressor lube oil, typically have dedicated controllers. Dan does however reference a 2008 Emerson Exchange presentation where field-based control was successfully performed in a compressor anti-surge application.

Another important finding is that there is even more reason to use control in the field as the number of loops on a fieldbus segment increases. The control response period is much greater when control is performed in the automation system controller than when control is executing in an FF device when they looked at 8 loops on a fieldbus segment.

If you'll be joining us in Orlando at the Emerson Exchange, you may want to check out Dan and team's presentation. One session will be Sep 29 at 11am and the second on Oct 1 at 10am. Visit the Personal Scheduler to plan the all sessions you want to see.

MP3 | iTunes

If you're a process automation professional and you haven't been following Greg McMillan on the ModelingAndControl.com blog and his "What Have I Learned?" series of posts, you're not in the flow of great knowledge sharing. Here's the current list of posts:

• What Have I Learned? - Sharing Knowledge
• What Have I Learned? - Einstein and the Ultimate Limits for Loop Performance
• What Have I Learned? To Lag or Not to Lag?
• What Have I Learned? - Unexpected Wireless Benefits
• What Have I Learned? - Cascade Control -Part 1, Part 2, Part 3
• What Have I Learned? - Manipulation of Multiple Flows Part 1 - Split Ranged Control, Part 2 - Valve Position Control, Part 3 - MPC
• What Have I Learned? - Ratio Control Part 1, Part 2
• What Have I Learned? - Cost and Source of Oscillations Part 1, Part 2, Part 3, Part 4
• What Have I Learned? - Wiring and VFD Problems
• What Have I Learned? - Writing

If you have wrestled with any of these control strategies, I hope you'll find some of the experience Greg shares helps shortcut a path to a solution.

The first post in the series, Sharing Knowledge, ends with this exhortation for you to join the ranks of those who'll share their expertise:

What distinguish humans from other animals are the gifts through art and science to discover, create, and disseminate knowledge and beauty expanding our understanding and perception of the universe. Art and science can both get at the essence and create new entities that take on an essence of their own. Both improve the quality and level of life. For me, good technical writing is both art and science. Try doing a weekly blog on what you have learned. I bet if you stick with it you will find it rewarding and create something that takes on an existence of its own.

I liked Greg's shift from the technical aspects of process control to today's post on writing. Greg gives his reasons for sharing these posts:

The main point of this blog like all of my writing is to share what I have learned. My goal for next year is to help prevent significant expertise and knowledge in process automation from being lost forever. I would guess 100 or more automation professionals are retiring each year who have published at best an infinitesimally small portion of their expertise for posterity. Also, new engineers are facing special challenges. My sense is the new kid in the control room doesn't have the mentors or the internal technical training programs I took for granted. They may be thrown into the midst of a difficult problem with no guidance.

He then shares ten points on writing. I especially like number 9, Don't get hung up on perfect grammar or a perfect piece. It reminds me of the unsolicited advice I offered a fellow Twitterer expressing his frustration with writers block. I pointed him to the Cult of the Done Manifesto.

I'd like to say great minds think alike, but I'll not go anywhere near there when I'm writing about Greg!

MP3 | iTunes

I caught up with Emerson's Lou Heavner the other day and we traded a few teenager "war stories." Lou is an advanced automation consultant and I've shared some of his expertise in posts over the years. He mentioned he had done a very basic presentation to show the interaction of operations, control, advanced control, and control strategies using the DeltaV system as his example.

I asked if I could post his presentation in my SlideShare account and discuss it here in this post. He was kind enough to agree.

Lou starts by describing key areas of the operator graphics describing the navigation, toolbars, alarm banner, and buttons to the model predictive controller (MPC) display that he typically will add to the advanced control project. He shows the loop faceplate which comes up when the operator clicks on an alarm. He notes that the operator:

...can change mode, SP, etc. He has one click access to loop tuning, alarm acknowledgement, trending, and with the appropriate privilege, he can access the engineering environment.

He shows the operator faceplate for a PredictPro MPC controller where the operator can view the optimizer, change modes and setpoints, and view the trend prediction horizon. For those who may be unfamiliar with an MPC controller, Lou shows the optimizer, which shows the variables being maximized or minimized and their associated economic value.

Lou next switches to the engineering environment where the modules and their associated parameters are located. Advanced control functions like loop tuning, neural networks, and MPC are available along with the regulatory control options. Lou shows the engineering side of creating MPC controllers from initiating automation step tests to creating and downloading the MPC controllers into the Delta controllers.

He shows an example of a composite block to calculate heater efficiency using the heat loss method. The calculation nests these composite blocks and Lou shows how to drill down and back out. He closes the brief presentation showing an on-line view which aids in troubleshooting the control strategy. This on-line view is available when the strategy is in simulate mode or actually running on-line in the controllers.

I hope Lou's simple descriptions and screen captures helps show the interaction of the advanced controls from their design through to their operation. We'd also like your thoughts on if you'd find this valuable to be seen in another form like a screencast.

I recently exchanged some emails with Emerson's Sergei Kuznetsov, part of TAG projects organization, and based in Minneapolis, Minnesota. Sergei is principal control systems engineer, certified professional engineer, and has an MSEE degree.

He shared with me an article, Staying In Control that he had written for Engineered Wood Journal magazine. The article describes ways to improve flake blending and mat forming in older oriented strand board (OSB) mills. For those unfamiliar with OSB, Wikipedia defines it:

Oriented strand board, or OSB, or waferboard, or Sterling board (UK) or SmartPly (UK & Ireland) is an engineered wood product formed by layering strands (flakes) of wood in specific orientations.

The issue with many older OSB mills built in the 1970s and 1980s is that they have large transport delays in the conveyors, which connect process equipment spread across the mill. Sergei notes that the problem most adversely impacts the blender inflow control and mat forming bin level control. These areas have large impact on the quality and consistency of the final product.

Such a problem of course is not limited to OSB production lines. Any process that involves a particulate material via conveyers can potentially have its deadtime affecting efficient control of related process variables.

From a control strategy perspective, Sergei described the challenge and solution:

A conventional PID (proportional-integral-derivative) feedback controller will not work well in applications with long process deadtimes. Good control can be accomplished, even in older mills, by employing the Smith Predictor control algorithm to address processes with significant transport delays or deadtimes.

In some extreme cases, this deadtime can be five minutes from the dry wood bin to the blender and then to the forming bin. If this deadtime is ignored in the tuning of the forming bin level controller and wood flow controller, process changes will prompt overcorrections and likely oscillatory conditions, unless the controllers are substantially detuned. Detuning causes sluggish response to changes and impacts the quality and consistency of the strand board.

Sergei detailed how the Smith Predictor algorithm addresses this deadtime:

The Smith Predictor uses a process model to calculate predicted process change in response to a control action as if there is no deadtime. This change is added to the PID process variable so the controller is made to "believe" that the corrective action actually took effect immediately, and thus will not take additional action. With such a modification, the PID controller can be aggressively tuned so it can provide good control of its process variable.

For the blending wood flow control, the flow can deviate due to the woodpile shape or differences in the bins that feed the conveyor. With a Smith Predictor accounting for transport deadtime, the loops can be aggressively tuned to handle the natural deviations in flow and bin switching. By closely controlling the wood flow, the proper ratios of wood to wax/resin can be maintained in the blender.

For the forming bin, controlling this level in older mills is notoriously difficult and typically requires a high level of operator intervention. Deadtime from long conveyors and blender retention time is a large part of this control challenge. A high forming bin level can cause unplanned shutdowns and bin level deviations can impact quality and consistency. A PID-based level controller with a Smith Predictor can account for this deadtime so that the level loop can be tuned aggressively to handle changes in the process and hold the level steady.

Sergei shared how these two loops are cascaded where the level controller is the master loop and the flow controller is the slave loop. He wrote:

When a forming bin level gets too high, the master sends a lower flow setpoint to the flow controller. If the level gets too high, flow setpoint is reduced. Both slave and master have their respective process deadtimes compensated by the Smith Predictor algorithm, so the cascaded pair works almost as if there is no deadtime at all.

The Smith Predictor does very well in processes with a fixed deadtime. When the deadtime varies, advanced process control (APC) strategies like Model Predictive Control can help provide reliable control.

MP3 | iTunes

Wow, what a fast and furious week... I'll close the week by highlighting a recent article by ModelingAndControl.com's Greg McMillan. He coauthors the article, Virtual Plant Provides Real Insights, which appears in the January 2009 edition of Chemical Processing magazine.

Greg's a published author on the topic of pH control and widely known his expertise. Working with Monsanto engineers, they sought a better way to control the pH of a wastewater pit. Maintaining pH between a permissible range of 6 and 9 was a labor-intensive activity. It also required veteran operators. Inexperienced operators, typically working night shifts, would call the plant engineers with a nearly full pit with an out-of-range pH and an imminent pumpout about to happen. Plant engineers typically don't enjoy being awakened to hear this.

The initial solution was to replace the pit with two 40,000-gallon tanks. The issue with this solution was high capital costs and limited plant real estate for the tanks. Plant engineers worked with Greg to see if a better solution could be modeled and developed. The goals were to minimize capital costs, provide reliable operations, and be easy enough to operate, even for less experienced operators.

Using much of the wisdom he freely shares at ModelingAndControl.com, Greg developed a virtual plant running a DeltaV system with embedded simulation on his notebook computer. Virtual plant means it runs both the simulation of the plant and the control system logic. Greg describes the setup:

The virtual plant included a dynamic model of the process with material and charge balances as well as mixing and injection delays, and a dynamic model of the control valves with deadband and resolution limitations. The models were configured and embedded in a distributed control system (DCS) along with the control strategies. The integrated nature of the virtual plant eliminated the need for separate programs, interfaces and emulations. We could develop and test the actual control modules and displays used in the plant.

Working with the lab data history, the team developed titration curve tabular data. They next matched the titration curve of the process model with the laboratory titration curve. They ran the demineralization unit batch sequence for different equipment, injection and automation system designs. The model showed where the biggest causes of upsets to pH level occurred.

They could also do what-if analysis to see if fast inline pH control could catch the disturbances and smooth them out. The result of the modeling and control analysis was that the pH could be controlled with 10,000-gallon tanks instead of the original 40,000-gallon tanks for project capital savings of more than 50%.

The article gives the design details of what process designs, process instrumentation, and control strategies were required to achieve the initial objectives sought.

If you've got a retrofit project ahead, you might consider a modeling and control analysis to see if large capital savings are possible. In today's global economic environment, this could make you a hero.

I also wanted to pass along that Greg was conducting a pH survey for a revision to his pH book:

Help Greg McMillan fine-tune his focus on pH issues by answering a few questions online. Taking part will give you a chance to win a copy of "Advanced pH Measurement and Control" as well as other prizes. When taking the survey, if you don't know the answer to a particular question, just select the 0-1% choice.

MP3 | iTunes

One only has to come across reports like the Institute for Supply Management's Manufacturing ISM Report On Business to see the slowing global economic conditions. The report notes:

Manufacturing activity continued to decline at a rapid rate during the month of December. The decline covers the full breadth of manufacturing industries, as none of the industries in the sector report growth at this time.

So if you're a process or plant or automation engineer, what do you do?

I exchanged emails with Emerson's James Beall to ask him what changes he's hearing about from process manufacturers since economic conditions began slowing in the fall. James and the variability management consultants help process manufacturers find ways to optimize their process.

James wrote:

Certainly, the focus has shifted from increased production to decreasing the costs of goods sold. Energy savings have even more emphasis than before. Distillation process are heavy energy users and are often using 5-25% more energy than is required--unless they are using a properly designed Model Predictive Multivariable Controller (MPC) or advanced regulatory control strategy.

Besides lower energy usage, improved control performance reduces variability allowing the process to operate closer to its constraints, which can improve yield.

In an earlier process variability post, James cited a study from the team's work that showed major causes of variability include control valve performance (30%), improper tuning (30%), and improper process and/or control scheme design (20%). In the ideal world, you could optimize your plant and reap these benefits throughout the plant's lifecycle. For the 20% process design issues you can. Unfortunately for the rest, the law of entropy being what it is--production processes tend to disorder over time. Valves stick as they wear, sensors plug, vibration on rotating equipment increases, etc.

Where digital instrumentation exists, the devices can report these issues to the operations and maintenance staff. A 4-20mA analog input signal provides a process variable, but not if the measurement signal is good. Likewise, a 4-20mA analog output signal to a control valve without feedback does not let the control algorithms know if the valve has moved to its intended location. These issues have to be uncovered through offline analytical techniques.

These all combine to change process dead times (the time delay from an output change to a change in the process variable) and the control dynamics of the process. The goal is to try to make the process dynamics as linear as possible and minimize dead time.

James recommends that you measure these changes in dynamics, annually at a minimum, and more frequently if the ROI justifies it. He and the team use Emerson's Entech Toolkit to identify common dynamics such as first order, second order overdamped and integrator+lag. This helps identify the process dynamics so that the control loops can be properly tuned.

With the process dynamics clearly understood, and final elements and measurement devices repaired or replaced, James and the team help plant engineers select the proper control algorithms for the process dynamics and tune the loops for best response without oscillation.

With the process properly lined out, it can now operate closer to operating limits due to reduced variability. Also, waste is reduced and less energy is typically consumed. These all directly impact the bottom line--a very good thing for these economic times.

Well-tuned regulatory control opens up the opportunity to also apply advanced control algorithms like MPC at a process unit level to further improve control performance and reduce operating costs.

MP3 | iTunes (I'm trying a trick from Gary Mintchell to stand to see if that adds more energy to my voice in these podcasts.)

Update: I just saw ARC Advisory Group's Larry O'Brien reference the ISM report. I was hoping my post was first because I found it by Googling around. Alas, Larry's post was the day before. Read it for more on the ISM findings and trust that I'd link to his post, if he was my original source!

I saw Emerson's James Beall the other week and asked him for a copy of his Emerson Exchange presentation, Interesting and Useful Features of the DeltaV PID Controller.

Every year, James presents to standing-room-only crowds and his presentation (given twice) this year was no exception. PID or proportional-integral-derivative control is definitely not a new concept. I did some Googling around and found references to it dating back to 1922 when N. Minorsky published an article on its use for automatic ship steering control.

While PID control has been around for a long time, technologists keep adding innovations, like degrees of freedom to the proportional action and the derivative action.

James began by describing three common PID forms: parallel, standard (a.k.a. ISA form), and series (a.k.a. classical form.) The standard form is the default choice in the DeltaV PID function block and the series form is an option. James counseled that the choice is based on your prior experience and personal preference. The series and standard forms are identical if the derivative action is not used. Also, your choice of forms can impact the conversion of tuning constants from a previous control system.

The PID function block also has a STRUCTURE parameter that provides two degrees of freedom for the proportional and derivative actions. On a change of setpoint (SP), you can scale these actions (BETA = proportional action scaling, GAMMA = derivative action scaling) between 0 and 100%.

The PID function block has an integral dead band (IDEADBAND) for when the error (SP minus PV) gets within this dead band. At this point, the integral action stops. James described a level controller application that feeds a downstream unit in order to reduce the movement of the controller output when the error enters the dead band.

James discussed three setpoint filters based on rate of change. One filter provides a time constant in seconds of the first order SP filter (SP_FTIME). Another provides a ramp rate at which downward setpoint changes (SP_RATE_DN) or upward setpoint changes (SP_RATE_UP) are acted on when the loop is in automatic mode.

Limits can also be placed on highest and lowest setpoints allowed, whether or not these limits are obeyed when the loop is in cascade or remote cascade mode, or whether output limits of the master loop in a cascade pair are used to limit the setpoint to the slave loop in cascade and remote cascade mode.

On the subject of cascade-control loops, James shared how mode tracking, bumpless transfers, and other loop interactions are automatically handled by the PID block's BKCAL interblock communications.

Gain scheduling is another PID control innovation for loops with nonlinearities where different regions of the PID controller can have different PID tuning parameters. The DeltaV PID function block can have up to three regions with different tuning parameters, based on a selected state variable (output, process variable, error, production rate, etc.) The algorithm provides a smooth transition between the regions.

James also provides guidance on valve output characterization and anti-reset windup limits in the presentation. Although these advanced PID functions can appear quite technical, they can significantly improve the performance of PID control and provide ways to handle difficult process dynamics. The bottom line to getting this control right is better control performance and a more efficient process.

You can read about the full capabilities of the PID function block in the 9.3 version of DeltaV Books On-line.

MP3 | iTunes

A great presentation at the recent Emerson Exchange was one that discussed the results of applying model predictive control (MPC) in a challenging gas plant process. Model predictive control has been available for decades and used in very large applications such as refinery process units, but its use in smaller applications found in the oil and gas sector is relatively new.

This plant was limited by its field compression capability, and throughput could be increased if they could reduce the differential pressure (DP)--mainly by lowering the inlet pressure to the gas plant. Swings in gas flows were introducing disturbances to downstream equipment such as carbon dioxide removal trains, causing them to trip. This caused further disturbances and in turn caused the operators to run these CO₂ removal trains very conservatively to minimize the risk of cascading train trips. The result was that overall throughput was reduced--directly impacting the bottom line financial performance of the plant.

Sarah Perkins and Andrew Taylor of ProSys Engineering, based in Australia, were called in to work with the gas plant's engineering staff to develop control strategies to maximize throughput while minimizing upset conditions and their cascading effect on other process units.

They saw three areas where advanced process control, specifically MPC, could be applied to meet the overall objective of maximizing plant throughput capacity. These included minimizing the differential pressure across the CO₂ removal trains, minimizing the liquid recovery plant (LRP) inlet pressure (to reduce overall plant inlet pressure), and using the plant's incoming pipeline as a surge vessel to eliminate spikes in inlet pressure which might trip the reciprocating compressors.

Let's dig in a little deeper in one of these areas--CO₂ removal trains. This gas plant had a number of these trains in parallel. The objective of the model predictive control strategy was to maximize the flow through the train either to a specified high limit or to valve saturation--whichever constraint was active first. Satisfying this objective effectively minimizes the differential pressure across each CO₂ removal train.

For each train, they designed and implemented a dedicated DeltaV PredictPro MPC control block running in their DeltaV controllers. To minimize the differential pressure, the goal was to maximize the butterfly valve opening coupled with the need to quickly cut back the flow in case another CO₂ removal train trips. With this MPC-based flow controller, the butterfly valves were linearized where the output was expressed as a % of flow capacity, instead of a % of valve position. Each butterfly valve had different flow characteristics, so each MPC flow controller was individually characterized.

The feed gas flow was the manipulated variable; the constraints were setpoint (SP) minus process variable (PV) error, and an operator-entered maximum flow rate. These constraints help to detect trip conditions and honor process limits like flow rates at which foaming begins to occur. In abnormal situations, the maximum flow rates are set to current flow rates to allow the operators a chance to make decisions about redistributing the flow rates.

The team made custom graphics for the operators to see the CO₂ removal trains on a single view, to quickly recognize patterns of abnormal situations and to take manual corrective action.

The payback on increased throughput was less than two months and even more throughput will occur when all of the reciprocating compressors are reconfigured for the reduced operating discharge pressure.

With MPC available at the DCS controller level, it can be applied to many smaller and mid-size applications in oil & gas and other industries. Engineers like Andrew and Sarah are helping process manufacturers solve challenging problems like these that also deliver fast financial results.

MP3 | iTunes

I won't spoil the press release in the works about the donation of a DeltaV system to a major university for use by a consortium of universities, but I will share that Emerson's Terry Blevins was at the university last week. He was there to provide an introductory process control and DeltaV system overview.

Since a consortium of universities is involved, a neat things done for this installation was to setup VPN access and Windows Remote Desktop access into the system to configure, test, and run the DeltaV control logic and calculations created using MatLab. In preparation for teaching the course, Terry used VPN to connect to the university's intranet. He then opened a remote desktop connection into the DeltaV system to prepare the models and simulations he was going to use to teach the course. The other universities' graduate students will use this same method as they collaboratively advance their research.

I thought I share some of these basics in case you are a college student or new to our world of process automation. Fair warning--for those experienced automation professionals, quickly hit the "back" button to avoid going any further into this post.

Terry begins his introductory presentation with organization and layout of a process manufacturing plant with the caveat that there is no "typical". Plants are divided into process areas and these areas are defined based on the equipment or process grouping. Examples are tank farms, boiler houses etc.

Terry gave a field device and wiring overview, showing examples of two-wire, four-wire, HART and Foundation fieldbus devices and how they connect into an automation system's I/O.

Next, he covered documentation of the plant control and instrumentation. Typical documentation includes a plot plan, which is a physical layout of the plant. Process flow diagrams show the major pieces of equipment in a process area and their design operating conditions. A P&ID (Process & Instrumentation Diagram) shows the piping and instrumentation installed. Loop sheets show the details of instrumentation and field wiring. Terry referenced the ISA S5.1 tag number convention standard that helps identify I/O as pressure, flow, temperature etc. and its readout and output function.

Terry showed the change in technology in distributed control systems over time from a hardware and field wiring perspective from individual wiring per device, to bus-based I/O. To familiarize the students with the hardware they might see in plants, he showed pictures of controller and I/O cabinets, marshalling panels, junction boxes, and panels with connections to other intelligent devices.

He then got specific with the hardware components and software applications in the DeltaV system and showed how the students could set up virtual plants with simulations of a running process against their control strategies.

After I passed a draft Terry's way for review, he pointed me to one of his earlier ModelingAndControl.com posts, Control Basics and Terminology that covered these basics plus more including:

• Introduction Historic Perspective
• Field Devices and Wiring
• Documentation of Plant Control and Instrumentation
• Characterizing the Process, Terminology
• Control System Objectives
• Manual and Feedback Control
• Feedforward Control
• Cascade Control
• Override Control
• Split Range Valve Position and Ratio Control

He even posted a test for those of you really ambitious new learners out there. If you're new to the world of process control, take a look at these links when you have some bandwidth and see if you find them valuable.

Update: A colleague from our DeltaV Twitter community points out that my hyperlink to Characterizing the Process, Terminology was not linked correctly. I've now updated it. Thanks for keeping me on my toes!

Update 2: Another reader found my incorrect use of "are" instead of "is" in the second paragraph. Specifically:

Consortium is a collective noun and therefore singular, not plural. The same applies to nouns such as group, herd and flock. The predicate (are/is) relates back to the singular subject (consortium), not to the plural object (universities).

As regular readers can attest, I need all the help I can get when it comes to grammar!

ModelingAndControl.com's Greg McMillan and Solutia's Mark Sowell will be presenting at the upcoming ISA 54th International Instrumentation Symposium. Their paper, Advances in pH Modeling and Control, describes the use of embedded simulation, coined "Virtual Plant" and model predictive control to improve the control of pH levels in a plant waste water treatment application.

The authors begin by describing the challenge of pH control:

The pH electrode offers by far the greatest sensitivity and rangeability of any industrial process measurement in terms of the measurement of concentration (hydrogen ions). To realize the full potential of this opportunity requires extraordinary performance of mixing equipment, control valves, reagent delivery systems, flow meters, control system design, and controller tuning.

The virtual plant is described:

A virtual plant can be used to sort out fact from fiction important for insuring performance and reducing capital and operating costs. The virtual plant consists of a download of the actual control system configurations and displays, embedded advanced control tools, and a dynamic process model running on personal computer...

The articles details the control strategy used:

We developed and prototyped model predictive controllers (MPC) to replace the fuzzy logic control system. MPC-1 adjusted the 1st stage pH set point to keep the second stage reagent valve at a minimum position for good response and reliability. MPC-2 trimmed the 2nd stage set point to keep the pH in the tank at an optimum pH.

The authors describe the interaction of the virtual plant with the real plant. They write:

In order to study and improve performance of the control system and the fidelity of the process model for actual process conditions, we put the virtual plant in a read-only mode online running real time. A simple interface module was configured that used object link[ing and] embedding for process control (OPC) to read indicated waste flows, controller set points, and controller modes from the actual plant.

If you are battling pH control in a waste water treatment application, you'll want to give this paper a read. You might also want to get your hands on one of Greg's books, Advanced pH Measurement and Control, if pH control is currently vexing you.

Update: Greg wrote me that the presentation went well and the room could have been bigger to hold all the folks interested in hearing about this topic. He has done a slight revision on page 1 to better tie in the results to the general situation with pH systems. This version is now posted on the original hyperlink above.

Interphex2008, the Pharmaceutical and Biotech manufacturing conference is going on this week in Philadelphia. Before Emerson's Terry Blevins and Mike Boudreau left, they passed along the presentation they are giving on Thursday, March 27. It's entitled, Application of PAT in Product Development. They are joined by University of Texas at Austin PhD graduate student, Yang Zhang and Broadley-James' Trish Benton. Here's an excerpt from the abstract:

The Process Analytical Technology, PAT, initiative encourages innovation in pharmaceutical development, manufacturing, and quality assurance to enhance understanding and control of the manufacturing process. The challenge for many manufactures is to identify how best to address the opportunities that PAT offers. Broadley James, Emerson Process Management, and the University of Texas are working together to examine and quantify the potential to reduce cycle time and out-of-spec product through the use of high fidelity, dynamic simulation and multivariate analytics. The objective of this work is to show that the impact of PAT can be maximized through the integration of these tools during product development (PD).

In the presentation, Terry begins by discussing the U.S. Food and Drug Administrations' PAT initiative, which has a framework that identifies some of the tools they discuss in the presentation. These include:

• Multivariate data acquisition and analysis tools
• Process and endpoint monitoring and control tools
• Continuous improvement and knowledge management tools

Terry describes on-line process analytics including fault detection and quality parameter prediction. Tools for detection of abnormal operations vary for measured and unmeasured disturbances. For measured disturbances, principal component analysis (PCA) captures contributions that can be associated with process measurements. Deviations may be quantified using Hotelling's T-square statistic.

The residual space that is not captured by the principal component score space reflects changes in unmeasured disturbances that can impact operations. These deviations can be measured with the Q statistic, squared prediction error (SPE).

For the quality parameter estimation, detection of deviations is addressed using projection to latent structures (PLS).

Armed with these statistical tools, Mike shows how the basis for bioreactor process modeling. In the book Mike coauthored with Greg McMillan, New Directions in Bioprocess Modeling and Control, they present a first principal bacterial model that was developed for fungal, bacterial, and mammalian cell processes. The intent of the process model is to more quickly evaluate input step techniques and control strategies in the PD stage.

The BioProcess International magazine article, PAT Tools for Accelerated Process Development and Improvement describes this collaborative effort between Emerson and Broadley-James technologists and University of Texas researchers along with how these tools can accelerate life science manufacturers' PD phase.

Over the past several weeks, ModelingAndControl.com's Greg McMillan has shared a three-part series on common control myths with his readers. In the initial post in the series, Common Control Myths - Part 1, Greg offers five myths, which he then shoots down, one by one.

Three of his five involve disturbances in the process. On unmeasured disturbances, Greg offers the following myth and rebuttal:

Unmeasured disturbances are a side issue - if there were no unmeasured disturbances, control would be a non issue because you could home in on the controller output that corresponds to the desired set point for a process variable. You would just need to run some data fitting algorithm one time and the loop would be set for the life of the process. In reality, there are always unmeasured disturbances.

Often the best-laid designs on newly commissioned loops require adjustments over time as the control engineer learns about the unmeasured disturbances impacting the loop.

In the second post in the series, Greg gets deep into the equations to show the scan time effects on peak and integrated errors. Since I've forgotten more than I've remembered, I'll trust him that the math checks out... J

In the final post of the trilogy, Greg debunks control myths six through ten. These include control valve performance, pH sensors, and thermocouples versus RTDs. For example, on pH sensors, he writes:

The most accurate type of pH sensors are used most often - the most popular sensors are the ones that require the least amount of maintenance, such as references with solid electrolytes, even though these may require more time to equilibrate and have a more variable junction potential. The flowing liquid junction reference for the right materials of construction and electrolyte is generally the most accurate but the least used type of pH electrode in industry because of the need to pressurize and refill the reservoir.

If you're early in your career as an automation engineer, you definitely want to subscribe to the ModelingAndControl.com RSS feed as one of your shortcuts to rapid learning.

Update: I was a bit too hasty calling it a trilogy! Greg has unleashed Part 4 of his common control myths.

I received an email from Anand Iyer. He's a certified project management professional (PMP) and a project manager in Emerson's engineering center in Pune, India. His project experience covers the gamut from pharmaceuticals, bulk drugs and intermediates to oil, gas and petrochemicals.

He's sent me a paper he's written entitled, Collaborative Measurement Control System Engineering. It describes how measurements close to one another in the process can collaborate with one another to verify their operation. He describes an example around a distillation column:

Now let us take two temperatures (bottom temperatures) in a distillation column and a level measurement. When the level is normal, the two temperatures are same or have a fixed relationship between them. TI1 is placed at a lower level in the column (near bottom) and TC2 is at a higher level (and used for Temp. control). Now TC2 is generally used for control. We can safely say that if Level is normal, and TC2 is under maintenance, TI1 can be used for control (with a minor adjustment to Setpoint if required). Thus Level and Thermocouple TI1 put together can "collaborate" the measurement of Temperature-measurement TC2.

Anand contrasts the traditional approach to a failure with how collaborative measurement strategies can be used in control strategies to avoid outages or process disturbances. In the traditional approach:

...the first thing done if an element were to fail was to swap the elements (either during the shutdown caused by the failure) or by a planned outage or having the loop in manual and doing the swap. At times, we have also used our ingenuity and just swapped the wires at the analog inputs and tuned control setpoints to have the plant up and running in a very short time. And hopefully, in all that chaos, someone had the presence of mind to record the swap on the wiring diagrams.

Using a collaborative measurement strategy:

...says that if level is not low and TC2 is not available then TI1 can be a valid measurement. We alarm the operator that TC2 is not available, fine tune the setpoint if required... All this occurs automatically and there is no outage or disturbance that could result in quality issues.

He extends the thought to Foundation fieldbus devices where the final control elements themselves can perform the logical evaluations and select the available primary or collaborated measurement, increasing the overall robustness of the control strategy. Anand also extends his thinking to wireless devices and how they could be used in a collaborative measurement environment--not as a primary measurement, but as a collaborative measurement to check on other devices nearby.

I hope you'll give Anand's paper a read and add your thoughts.

The Automation.com list server has an interesting thread, Three Element Drum Level Control Problem. The question asked was:

We have a waste heat recovery boiler that is supplied by exhaust of a 20MW Gas Turbine. We've seen that at lower turbine loads (75% and below) the three element drum level controller cannot maintain the drum level at desired setpoint. As soon as the load on the Gas Turbine is increased to more than 75% of rated load, the stability keeps getting better. At rated load (20MW) the drum level is very stable and close to the setpoint.

There have been several responses discussing the tuning at various loads. I asked around to see what advice we might have to offer. Emerson's Jack Tippett, a variability management consultant noted that it is critical to know your process dynamics. His point:

If you don't know the process dynamics, control tuning is an art not a science and good control performance is an accident not a certainty.

Once you know your process dynamics, it is important to design your strategy to assist in achieving the process objectives in light of those dynamics. Jack noted a similar situation from his past where he tuned the levels in a 450-megawatt heat recovery steam generator (HRSG) system.

There were six boilers including two lines with high, medium and low-pressure drums. This power producer was unable to achieve a station ramp rate of 25 MW per minute necessary for automatic generation control (AGC) due to serious swings in the drum levels.

After measuring and determining the process dynamics, the process was re-tuned and they were able to achieve the ramp rate and achieve good level control at less than 70% load.

Jack also noted that they chose a single-element control strategy for the following reasons:

1. Feedwater flow control requires a working flow meter: the sense lines for the flow transmitter were outside and were subject to freezing. The Fisher valve had a DVC positioner and AMS software to monitor incipient valve non-linearities (which are the main reason for the second element.)
2. The open loop dynamics (changing the feedwater valve position manually and watching the response to level) on all six boilers showed very small dead times (1 to 6 seconds). This meant that the proportional-integral (PI) level tuning could be very aggressive. As a result, there was no value in the third element (steam flow feed forward)--the level control could be fast enough to respond the changes in level due to steam demand changes. The real need for the feed forward from steam is when the level dynamics are very slow (30 - 90 seconds dead time) so that the feedwater flow can anticipate the long-term level changes (due to steam demand) in spite of the shrink/swell effect.

By having good measurement in the flow, valve position, and valve characteristics and good understanding of the process dynamics across its operating range, Jack and the plant engineers were able to successfully implement a simple single-element control strategy.

I came across an email that the ISA Honors & Awards Committee has selected the paper, Improving PID Control with Unreliable Communications, for its excellence in documentation award. Emerson's Deji Chen, Mark Nixon, Terry Blevins, Willy Wojsznis and the University of Texas, Department of Computer Sciences' Jianping Song and Aloysius K. Mok wrote the paper.

The paper examines PID control in a wireless network where intermittent loss of communications is likely to happen. It identifies the poor dynamic response of standard PID algorithms in this loss of communications scenario. The team proposed an enhanced PID algorithm to improve dynamic response under these conditions.

Terry Blevins summarized the paper well in an earlier post on the Modeling and Control blog. The post, PID Modifications for Unreliable Communications describes the situation:

One of the technical challenges is that the 2.4 GHz spectrum defined by IEEE 802.15.4 is also used by Wi-Fi and Bluetooth devices. Also, some electrical devices found in industry generate noise in this frequency band. Thus, at times it is expected that a transmission will be corrupted. To help minimize the impact of these other devices on communications, the Time Synchronized Mesh Protocol (TSMP) selected for wireless HART uses frequency hopping. Even so, at times it is expected that multiple transmissions of a measurement used in control or multiple communications of control actions to an actuator may be lost.

Terry describes how the loss of communications can cause the PID loop to continue executing and wind up due to the reset action. This reset action can be disruptive to the control of the loop. And, if the derivative (the D in PID) action is used, the loss and resumption of the control measurement signal can cause a spike, again bumping the control of the loop.

The Emerson and UT technologists worked through a solution to minimize the impact of this loss of communications. Terry sums up the change:

However, by modifying the reset and derivative calculation to account for the time since the last measurement update, then it is possible to minimize the impact of loosing multiple measurement transmissions.

If you want to look at the math behind this innovation, check out the overview presentation, PID for Unreliable Communications, given at ISA 2006.

Congratulations to the team for their contribution to furthering the advancement of wireless technologies in process automation!

In our continuing series of screencasts, I'm trying to give examples of how advanced diagnostics in Foundation fieldbus devices can be used in control strategies to avoid abnormal situations and potential losses in production.

Emerson's Rune Reppenhagen shows in this quick 2 minute, 47 second screencast, how an advanced model predictive control strategy in a DeltaV controller automatically recognizes a failure diagnostic in a temperature transmitter and switches the mode of control over to a manual state.

At the same time, this diagnostic alerts the operator of the situation, and the AMS Device Manager software shows the condition of the transmitter so it can be quickly repaired.

By using the advanced diagnostics from these intelligent field devices in the control and advanced control strategies, conditions which impact the availability and quality of the process can be avoided.

For those of you challenged with the vagaries of pH control, I wanted to make sure you had seen the news of an upcoming pH Control web seminar, arranged by ISA, featuring ModelingAndControl.com's Greg McMillan. The web seminar covers the root causes of poor performance in pH control systems.

In a recent post, Greg describes how he plans to share his experiences:

I spent a lot of time on pH startups. I found most of the key design concepts needed for success where not discussed anywhere, For example, the normal dip tube design for reagent injection is disastrous and the mixing and valve resolution requirements are exceptional. I discovered how I could reduce the number of stages of neutralization, offer inexpensive alternatives to the classical neutralization vessel, and decide when signal characterization could help or hurt your control objectives.

Unlike his recently released free eBook, this May 16 web seminar (2:00pm - 3:30pm Eastern U.S. Time) does have a cost. It's $195 (USD) for ISA members and $225 (USD) for non-members. If you're not the lone person in your organization who struggles with pH control, Greg suggests:

The seminar is much more cost effective if the registrant connects in a conference room with a computer projector.

If you can't make this event, Greg has also published a book on this topic, Advanced pH Measurement and Control, 3rd Edition.

Recently I discovered in my PAT RSS persistent search feed an article in Pharmaceutical Processing magazine entitled, PAT Solutions-Eight advanced process control technologies worth considering. This article was written by Rick Rys, President, at R2 Controls, and Janice Abel, Director, Global Pharmaceutical and Biotech Industries, at Invensys.

Since ModelingAndControl.com's Greg McMillan recently co-authored a book New Directions in Bioprocess Modeling and Control-Maximizing Process Analytical Technology Benefits and recently had an article published entitled Maximizing PAT Benefits from Bioprocess Modeling and Control in the November 2006 issue of Pharmaceutical Technology IT Innovations, I had to ask for his thoughts.

Greg sent me a great email which I'll pass along with my edits to insert hyperlinks:

The DeltaV systems offers the advanced control technologies mentioned in the PAT article, such as synthetic analyzers, feedforward and predictive control, dead time compensation, and model predictive control in its standard integrated graphical configuration studio that uses Fieldbus function blocks. The synthetic analyzers not only include online regression models such as Neural Networks but also embedded first principal models. Furthermore, innovative analytics, control systems, and models can be prototyped faster than real time in a virtual plant on a desktop or laptop PC anywhere. The virtual plant uses an exact duplicate rather than an emulation or simulation of the control system in the control room. Advanced technologies in the virtual plant can be developed and tested from the high speed play back of historical data from existing systems used to automate bench top fermentors. This includes a new adaptive control technology that identifies process dynamics and indicates the relative improvement possible from better control. The high speed virtual experimentation capability of the virtual plant is a key feature and may be the only way to provide enough historical data particularly on "what if' scenarios since a fermentor batch for most new bioprocesses takes 14-17 days.

The same virtual plant can be used for education of operations and technical support by the dynamic restore and high speed playback of instructive periods of operation.

The technologies can be connected to the bench top system for evaluation, verification, and adaptation of models early on in the commercialization process.

The uses and advantages of the synergistic environment of the virtual plant are explored in the book New Directions in Bioprocess Modeling and Control, the article "Maximizing PAT Benefits from Bioprocess Modeling and Control" in the November 2006 issue of Pharmaceutical Technology IT Innovations, and in the lectures on the Modeling and Control.com blog. The important practical implications of the extremely slow one direction integrating response of biomass and product concentration on modeling and control are also discussed in the book, article, and website.

The integration and knowledge management of a diversity of technologies in DeltaV addresses the essence of the PAT initiative as expressed in the following statements by the FDA:

Process Analytical Technology:

• It is important to note that the term analytical in PAT is viewed broadly to include chemical, physical, microbiological, mathematical, and risk analysis conducted in an integrated manner.

Process Analytical Technology Tools:

• Multivariate data acquisition and analysis tools
• Process and endpoint monitoring and control tools
• Continuous improvement and knowledge management tools
• An appropriate combination of some, or all, of these tools may be applicable to a single-unit operation, or to an entire manufacturing process and its quality assurance.

Batch process manufacturers have long understood that applications which require both sequential and continuous control have been a challenge. A typical example is a centrifuge application commonly found in Biotech and Pharmaceutical manufacturing processes. A centrifuge separates solid and liquid material by spinning a sieve-like device at high rate of speed, and recovering the liquid, solid or both materials.

I caught up with Brian Crandall, a Life Sciences industry project engineer. He said that proper control was critical since centrifuges are quite expensive, and sensitive to a variety of failure conditions. These conditions need to be addressed within seconds to prevent equipment damage and possible injury to operations staff.

Brian summed up the control challenge as the centrifuge having various operational states. Moving between these states is best done using sequential operations. However, monitoring of the failure conditions, which change in severity and action depending upon the operational state, must be done in parallel with the sequencing.

If a failure condition occurs, the current sequence has to be stopped, and the specific failure sequence started within a minimum timeframe. The S88 Batch Model defines a sequential state driven approach, but it does not fit the requirements of this application. The big issue is the failure monitor does not offer continuous monitoring required for quick reaction to the failure condition.

Using a DeltaV system for this particular Biotech application, Brian designed, tested and implemented a modified S88 state model that had the ability to stop a sequence without waiting for a transition thus meeting the high-speed timing requirements of the equipment. Multiple sequences would be required for the main sequence, shutdown, and E-Stop to allow stopping one sequence and starting another at the same time. Also the code design needed to modular to fit the rest of the S88 modular design philosophy. Also, this design placed control at correct level in S88 batch model, at the equipment module level.

Some failure conditions the design addressed included: high vibration, VSD fault, low seal water pressure, and low instrument air pressure. Depending on the state of operation, the failure conditions required different actions per a failure condition matrix.

Other industries have applications requiring this mix of continuous and sequential control. Some examples include a refiner in the pulp and paper industry, extruders in the specialty chemicals industry, and other state-driven processes or equipment.