Stan: November's column with George Buckbee, "How to Get the Most Out of Your Software for Loop Tuning" (November 2013), gave a good introduction to loop performance that set us up for the next step: How do we get from loop control performance to process economic performance? Today more than ever, we just can't make things look better. We have to show the impact on the bottom line. Unless automation engineers show the financial gain from process control improvements, this activity will never achieve its potential or get recognition for the value it delivers.
Further, quantifying the effect of better control on plant profitability lets us focus attention on the most important improvements–those that return the biggest benefits.The recognition gained is critical to support for further individual efforts and revitalizing our profession.
Greg: Lewis Gordon, a principal control systems engineer retired from Invensys after 38 years of experience, has submitted some comments on improving loop performance to increase plant profitability. Lew specialized in control applications design, tuning and loop performance monitoring software (ExperTune's PlantTriage). Here are Lew's comments on the column with Buckbee and more details from a subsequent conversation.
Lew: George is absolutely right in saying that improving loop performance is about more than tuning. He mentioned some specific examples, including loops in manual, root causes, interaction, stiction, backlash and faulty instrumentation and/or actuators.
In the case of controllers in manual, this is sometimes not a significant issue. A controller may be in manual simply because it has become obsolete (out of service) or because the current process operating mode does not require it. Certainly a controller that is always in manual is wasting the cost of its installation, as the column makes clear. But installation is a one-time, sunk cost—water under the bridge. As George says, the important questions are why is it in manual now, and what will be the benefit of getting it into auto.
The first issue is where transferring a controller into auto triggers an expanding oscillation, forcing a return to manual, even if only under certain operating conditions. This usually happens because the tuning is so tight that the loop is unstable with the controller in auto. But it is also possible for each of two reciprocally interacting controllers to be stable so long as only one is in auto, but for both to cycle when they are in auto at the same time. Both of these situations do happen, but not often. More significant for the long term are those loops that do not control well enough during process upsets. Often, on difficult-to-control loops, the controllers are in auto, but tuned sluggishly to keep the loop stable under most operating conditions and hold steady state in the absence of upsets. Then manual/operator control is used to handle occasional upsets. So two revealing metrics are the frequency of auto/manual transfers and the frequency of output changes in manual, which most loop performance software also tracks.
Stan: I like the idea of metrics on how often a controller is switched to manual and how often the output is changed. If the output of a controller in manual is not being adjusted by the operator, closed-loop control is probably not that important.
Lew: In either case, retuning such loops can improve control performance, especially during an upset, and reduce operator load. Nevertheless, this improvement is unlikely to have significant economic benefit without changes to the average operating point.
The other examples mentioned certainly exist, and when they are found and corrected, the benefits will be very welcome. Backlash and stiction are almost pervasive. They cause oscillation too, but these situations can be distinguished from gain-driven oscillations because the oscillations they cause are self-limiting. If the deadband and stick-slip are less than 1%, the limit cycle may be more of a nuisance than a significant economic problem (unless they affect one of the variables mentioned in the next paragraph). They do create a maintenance issue because of the associated wear and tear on the valve; however, such cases can go on for months and even years without being addressed because their economic and operational impacts are not large enough to demand a solution. Similarly, instances of faulty instrumentation that have significant effects such as George describes provide surprising and dramatic improvements. However, they are usually singular situations—not really part of a long-term continuous improvement concept, except as gateway events.
A process of continuous improvement is the real money-maker, enabled by tuning and control loop performance monitoring (CLPM) software. The article pointed toward the most beneficial reason for using CLPM software to improve control performance in auto: to minimize variation (reducing standard deviation) in the process variables (PV) that affect plant economic performance. The key point is that more stability often allows the operating point for economically significant variables to be moved to values that provide increased production rates, higher yields, lower energy costs per unit of production, longer equipment life, more uptime, and fewer emissions violations and fines. These improvements may be less dramatic than killing a troublesome oscillation, but they are where the big money is because their return steadily accumulates over time. An improvement in any of these areas of just $150 an hour will return over $1.25 million a year. Every plant is a "target-rich environment" for such opportunities, to borrow George's phrase. A plant of any size will make millions of dollars' worth of product a year and consume millions of dollars' worth of energy. Even small percentage improvements in either of these factors can generate a huge return on investment (ROI) for the effort required at no capital cost.
Stan: The type of applications may make some problems just an aggravation. We need to know how a deficiency affects plant profitability. Greg has had some bad experiences when plants tried to save money by not using positioners or by using piping valves as control valves.
Greg: There are many cases where the process gain is low enough and primary process time constant large enough so that limit cycles are filtered by back-mixed volumes to the point of inconsequence in terms of economic impact. I have also seen a plant that could not start up because positioners were purposely left off fast loops, requiring PID outputs of 25% or more just to open a valve. I have also seen many problems created when on-off valves (with stiction and backlash of 8%) were used for control, lowering the yield caused by a limit cycle in reactor feed and pressure control. Smart positioners don't show this problem because the feedback is on actuator shaft position rather than on the internal closure element (ball, disk or plug). The November 2012 Control article, "Is your Control Valve an Imposter?" relates the incredible deception. A throttling valve with a digital positioner and diaphragm actuator from a manufacturer whose heritage is control valves rather than piping valves can achieve a precision of better than 0.5%. Imposters often have deadband and resolution (stick-slip) that is an order of magnitude worse.
For split-range applications, stiction is greatest near the shared closed position, and any resulting limit cycle across the split range point wastes reagent or utilities even if the controller is adaptively tuned to deal with different dynamics of the streams and valves.
Likewise, pH applications with strong acids and strong bases sometimes show pH limit cycles around setpoint even with fairly good throttling valves, requiring another process volume for smoothing and adjustment. When designing a new system, it may be possible to save considerable equipment cost by maximizing reagent valve precision and rangeability, perhaps allowing a smaller volume or even eliminating a stage of neutralization. My formative years spent solving pH problems sensitized me to valve sensitivity. Nearly all of the oscillations observed were due to valve or mixing problems.
Less obvious is the additional loop dead time associated with the time it takes for the PID output to get through the stiction and backlash so that the valve moves. This additional dead time can be approximated as the resolution limit or half of the deadband divided by the rate of change of the controller output. For unmeasured load upsets, the increase in dead time can be significant for a slow process or excessively slow tuning. For a highly exothermic reactor, a runaway condition can develop from a delayed coolant response.
Scenarios where an operator intervenes despite a well-designed and installed valve measurement and adaptively tuned controller occur when the controller is asked to deal with abnormal operation during start-up, transitions and changes in phases of batch operations. Michel Ruel gives some examples automation to eliminated manual actions with big economic impact in the two-part Control Talk column series from Nov. and Dec, 2009, "Show Me the Money" and Part 2.
I have also seen PID outputs on unit operations that were saturated because the piping system was not adequate to run the plant at more than twice the original nameplate capacity. The consequences were insignificant compared to the other problems associated with equipment in parallel unit operations periodically shutting down for defrosting.
Stan: How do you find out if a loop that is not performing affects plant economic performance?
Lew: Identify the process variables that relate to energy, raw material use and product value. Often these process variables are associated with product quality or composition. Maximizing an impurity is a prime opportunity. Two examples are maximizing the moisture in the product from a dryer and air in ice cream packed in a container, without violating quality specs. This optimization reduces the raw material and energy used per unit of product sold. A chemical engineering example is maximizing the impurity in a higher-value product stream. In distillation, this is often the overhead light fraction, and the impurity is the heavy fraction. Reducing over-purification can save a lot of energy and increase product yield per unit of feed.
Greg: In my Control Blog, "Extending Process Control Benefits Analysis," one of the overlooked opportunities is the margin that the operator introduces by choosing an operating point much further from the optimum than what is permitted by reduced process variability. This seems largely due to a lack of process knowledge and the potential opportunities. Mike Brown addressed this at the end of the three-part Control Talk column series from July, Aug. and Sept 2010, "Process Performance Improvement" (Part 2 and Part 3).
Stan: Why are operators reluctant to push closer to a constraint?
Lew: Because process gains are often higher and quality violations more likely closer to a constraint. Also, the possibility of a safety system or relief device activation increases as you get closer to a constraint. This can get operators into trouble. However, better control performance allows operators to control at more profitable points with more stability, more safety and more confidence.
Greg: Online metrics that show the penalty of margins and highlight improvements are the best way of motivating operators and the process and automation engineers supporting operations. For a discussion of some of the practical considerations in implementing online metrics, see part six of my presentations for the mentor program at MYNAH Technologies, "How Do You Improve Plant and Model Performance?"
Can software automatically point the engineer to the root cause of a problem whose solution can provide significant potential benefits?
Lew: A single variable is often affected by a number of other variables. Loop performance monitoring software uses power spectrum analysis to point to other variables with common frequencies that may be major contributors to the variability in a key variable. Interaction maps use color-coded XY grids to indicate strength of interaction. Still, you need to know how the interactions flow. The software does not determine if the dog is wagging the tail, or vice-versa. Someone with process understanding, a person who can determine what is upstream and downstream and what happens first, needs to look at how mass and energy flow through the system.
The amplitude and period of an oscillation are also important clues. If the amplitude is stable, the cause is most likely deadband or stick-slip. If the period is quite large relative to the dead time, the cause is probably a load disturbance. If the amplitude decays or grows, the cause is probably related to controller tuning. Cycling from aggressive tuning is the least likely cause, since most loops are tuned sluggishly. The most common causes of oscillations are dead band and stiction. Next are interactions. In the case of interaction, you can sometimes minimize the interaction by tuning the two loops to cycle at more separated periods similar to what is done for cascade control.
Greg: Another situation from my experience is that loops often have too low a PID gain (a too high proportional band). Near-integrating processes (e.g., continuous processes with large primary time constants), true integrating processes (e.g., level, batch temperature and composition, and gas pressure), and runaway processes (e.g., highly exothermic reactor temperature) often have a PID gain that is one to two orders of magnitude lower than the maximum allowed for a critically damped response.
Oscillation amplitude gets larger and the decay slower as the negative feedback internal to the process is decreased. Thus, oscillations get worse as you move from near- integrating processes with weak internal negative feedback to true integrating processes with zero internal negative feedback, and on to runaway processes with internal positive feedback. For controllability, the PID controller must provide the negative feedback missing in the process through the proportional mode. The result is effectively a low PID gain limit (high proportional band limit), and for these processes, the normal practice of reducing controller gain (increasing proportional band) to reduce oscillation amplitude can actually make the oscillation worse.
What practitioners don't realize is that the reset time must be increased by the same orders of magnitude to prevent slow rolling oscillations with a period 10 times or more the ultimate period. For these loops, the solution is often to increase proportional action and greatly decrease integral action by increasing the reset time by two orders of magnitude. For more details on this severe problem in the process industries with polymerizations, large liquid volumes, and batch operations, see Part 2 in the Control Talk blog series 3/12/2013, "Processes with No Steady State in the PID Time Frame."
Most of the benefits we sought in my years at Monsanto and Solutia were associated with increasing production rates and on-stream times. We developed an opportunity sizing and assessment process that addresses the key questions of how to quantify the potential and realized benefits. The people issues are discussed in the July 2012 Control Talk column, "The Human Factor." Many of the solutions that resulted in benefits are discussed in a four-part Control Talk blog series 3/22/2102 to 4/12/2012, "How Can You Quickly Increase Production Rate and Efficiency?" (Part 2, Part 3 and Part 4) Feedforward control was a frequent solution. As plants minimize inventory and change production rates more radically, a plant-wide feedforward control strategy becomes advantageous, as described in the March/April 2011 InTech article, "Feedforward Control Enables Flexible, Sustainable Manufacturing." The November 2011 Control article, "Don't Overlook PID in APC," details how a simple configuration change to add a valve position controller can improve plant capacity and efficiency. Finally, you may also want to check out what Jacques Smuts has found in the Aug 2012 Control Talk column, "Control Loop Improvement."
Stan: It is important that plants understand that software does not replace a process control engineer, but actually increases the need for one to take advantage of the opportunities offered. Next month we continue our conversation with Lew to see what can be done in terms of reversing the decline in process control expertise triggered by retirement, "lean and mean" operations, and the narrow focus on project schedule and cost. Part of the answer is in the ability to quantify and prove the source of the benefits.
Greg: As retirees we look back fondly at the various goals programs. Here are the "Top Ten Uses of Old Performance Reviews."
"Top Ten Uses of Old Performance Reviews"
(10) Fire starters. For fun times next to the fireplace or barbecue reading performance enhancement process (PEP) stories. Heck, you can throw in the PEP stuff too.
(9) Insulation for your attic. The more paper work you generated, the more energy you save. Go for an R50 factor.
(8) Thank-you cards for your boss. Staple the X ratings to your raise notices.
(7) Landscaping for gerbils. Now, they know how to have fun with paper.
(6) A gourmet delight for vegetarians. Great fiber and an appropriate end.
(5) Puppy training papers. Give a performance review on the puppy's ability to hit your performance review.
(4) Engineering reunion decorations. Highlight the really imaginative phrases.
(3) Targets for skeet. Wad them up with glue and water and launch them.
(2) Science fiction story. Call it "Engineers from Other Worlds."
(1) Party hats for the next stockholder's meeting. Sit in the front row and toss them on stage. Get your CEO to wear one.