A plant’s process control system affects safety and environmental results, energy and operating costs, quality and production rate. Most sites have spent millions of dollars on control systems — an individual loop may cost upwards of $10,000 to implement when you consider expenses for engineering, sensor, wiring, controller, valve, configuration and programming. Yet, is this sizable investment paying off as well as it should? Do you really know if your control system is delivering results? Is it reducing variability and operating costs? Is it improving quality, safety and production? If your plant is like most, you don’t have a good way to measure control system performance. So, here, we’ll discuss how to measure, track and boost performance of your process control system. We’ll introduce a scoring system that will help you evaluate potential for economic improvements. When you measure and manage control system performance, you get better results from the process as a whole. Studies indicate that improved process control typically leads to reductions of between 2% and 6% in total operating cost .
Table 1 shows some control-system performance measures as well as results for typical and best-in-class plants. If you don’t know how well you’re doing, you’re hardly alone. Most plants aren’t monitoring even the most important of these metrics. This situation isn’t surprising. Work on new control systems focuses heavily on infrastructure, connectivity and security, with little emphasis on applications and performance. Thus, many plants find themselves with shiny new control systems that run the same old control applications and strategies. Also, most Distributed Control Systems (DCS) don’t have tools to calculate and report basic information on performance.
Service Factor Report
Figure 1. Abundance of loops with 0% service factor signals a serious problem.So where can you get performance information? Most plants have a process data historian actively gathering thousands of pieces of data. Analyzing this information provides great insight into control system performance. Specialized “control loop monitoring” systems can extract the most meaningful data from the process historian (or directly from the DCS) and present these data in a useful prioritized fashion. Let’s now look at a variety of parameters that can give you an indication of performance. Simple MeasuresThree specific factors provide a good starting point for assessing control system performance: Percentage of loops in manual. When control loops are in manual, they aren’t performing their basic function — they’re not controlling. Yet it’s typical to find 30% of control loops running in manual. The loss of control poses safety and environmental risks and, unfortunately, has taken a deadly toll. Furthermore, a loop running in manual usually indicates some other, underlying problem. If a sensor fails, a valve sticks, or controller tuning is unstable, the operator usually will place the loop in manual. A control-loop-monitoring system can provide a complete list of loops in manual, pinpointing many opportunities to improve operation. Valves at limit. When valves are fully open or fully closed, they’re not controlling. The process can wander without any ability to control it. When a valve is 100% open most of the time, it’s often a sign of an under-sized valve. Develop a list of these under-sized valves; you’ll have a great starting point for a debottlenecking study. Many plants have been able to increase production rate simply by focusing on key bottlenecks caused by under-sized equipment.
Oscillation ReportFigure 2. Identical oscillation periods indicate a single common cause. A control-loop-monitoring system can greatly simplify this task by using real-time data to calculate the percentage of time each valve is fully open. Operator actions. Another way to determine the control system’s effectiveness is to evaluate how many operator actions are required to keep the process running. Count each of the following as an operator action: • acknowledging an alarm or message;• altering a set point;• changing controller mode; and• varying control output position.This provides a simple, yet effective way to find where big problems lie. In most plants, less than 10% of control loops account for 80% of operator actions. In some cases, these “bad actors” become a regular distraction, keeping operators away from more important tasks. At SABIC Innovative Plastics, Selkirk, N.Y., operator Gene Lezatte says that one distillation column routinely caused major operational headaches: “We were chasing it around. Sometimes we’d have two operators to manage just that one column.” After some intervention by the controls engineer, Lezatte says, “That column is rock-steady.” Statistical MeasuresTraditional statistical metrics such as variability can help provide overall perspective on the process. After all, one of the primary functions of a control system is “to reduce variability.” Some simple, classic measures include: • variability or variance;• standard deviation; and • average, minimum and maximum.
Drill DownFigure 3. Process variable exhibits a large repeatable swing. Trends in any of these values can indicate problems in a process. Establish a baseline for variability of key factors and pay close attention when variability dramatically increases or decreases. When normalized by instrument spans or byproduct specification limits, the absolute value of these metrics also may hold some meaning. Academic MeasuresProfessors at engineering schools have emphasized certain control performance measures over the years. These metrics offer some value as indicators, particularly when doing an “apples to apples” comparison. However, be extremely careful about using them in a real-world process. Integral of absolute error (IAE). Almost every process control textbook talks about IAE as a measure of control effectiveness. IAE is the “sum of the absolute error,” i.e., the difference between the process variable (PV) and the set point (SP) over time. On the surface, IAE appears a very good measure of performance. Indeed, for a controller humming along in automatic, this metric can provide some insight. But in a real-world process, many factors can directly affect it: • During set-point changes, the PV may be far away from SP, leading to a false high value of IAE.• More or fewer load upsets will change the value of IAE.• When the loop is in manual, IAE has no real meaning as a measure of control performance.For these reasons, only use IAE as a measure for “apples to apples” studies. Harris Index. This measures performance as the ratio of the performance of the current controller to that of a Minimum Variance Controller (MVC). MVC is a theoretical controller that provides “best possible feedback control performance” — it tries to bring the PV back to SP in the shortest possible time by making relatively dramatic moves of the manipulated variable. Unfortunately, real-world control valves can’t move 100% in one second! So we must settle for somewhat-slower-performing feedback loops. The Harris Index offers some value a performance indicator. However, it has some restrictions: • Calculations require an accurate estimate of process deadtime. It’s not always possible to have this in advance for thousands of control loops in a plant.• Alone, it doesn’t account for the desired slowing of control loops to coordinate response with cascade, ratio or other loops.In practice, the Harris Index works fairly well on fast control loops when you have a good estimate of deadtime. Recommended Key Performance IndicatorsThree particular performance indicators have delivered proven results over time:1. service factor;2. oscillation significance; and 3. valve travel.Service factor. This is the percentage of time the control loop is fully in service — that is, operating in normal automatic mode with the control valve able to move and the PV within regular operating limits. In a well-performing loop, the service factor would be 100%. When it drops below 100%, start looking at each component of service factor individually — to see if you have a valve sizing problem, an instrument span problem or a tuning issue. Figure 1 shows a sample service factor report with many loops showing 0% of time in normal mode.Oscillation significance. Plant efficiency, quality and overall performance can suffer because of significant oscillations, i.e., ones that measurably affect variability and have a distinct period of oscillation. So, it pays to determine the most important oscillations. When you measure significant oscillations, you’ll find that many loops oscillate in tandem with others. So, when you can identify the root-cause oscillation, you can stabilize many loops with a single corrective action.
Process Interaction MapFigure 4. Strong colors indicate strong correlations, while left and right denote lead and lag times. The likely root cause shows strong colors on the right side of this map. Determining root cause of oscillations involves detecting the period and strength of each oscillation. When oscillations propagate through a plant, they always stay at the same period. A report of oscillations sorted by period will help find root cause of each oscillation. In Figure 2, for instance, several loops are oscillating with a period of exactly 34.13 min. —a telltale sign there’s a single common cause to the oscillation of all these loops. Drilling down on a specific loop can help confirm the diagnostics. Notice the large repeatable swing of the process variable in Figure 3.Valve Travel. The control valve is the control system’s workhorse. Valve travel is a good overall indicator of maintenance demand and loop performance. It’s calculated by summing the amount of valve travel over a day. Each time the control valve moves up or down, we can totalize the amount of movement and learn something about the controller and the valve. Excessive movement of a control valve creates two problems:• inordinate wear on the valve; and• process upsets induced by the extra valve movement.When a performance supervision system is first installed, you can find some major problems very quickly by looking first at the extent of valve travel. Other Performance Metrics. Some more-sophisticated measurements can help pinpoint specific process issues. A few examples include: • Oscillation detection and oscillation periods allow you to find the cause of routine process upsets.• Comparisons with process specification limits enable operators to push the process closer to optimal operation.• Interaction maps, such as the one shown in Figure 4, establish cause-and-effect relationships within the plant.Sometimes looking at metrics together can tell you something more about the process or equipment. For example, when the maximum value of the PV over some time period equals the minimum value, then the instrument has “flatlined” or failed. This is a common mode of failure for many types of instruments including thermocouples and level transmitters. Many plants continue running for days or even months without recognizing that some instruments have failed in this way. Yet a simple diagnostic comparison like this will lead your instrument technicians directly to failed instruments.Improving Control System PerformanceYou can’t control what you don’t measure. Start by measuring some simple statistics for your control system. With “Time in Normal,” “Time at Limits,” and a few other metrics, you’re on the way to discovering the key limitations of your process. Measure the right things. You easily can get side-tracked by calculating hundreds of meaningless metrics. Make sure that you’re measuring those things with the greatest impact on the bottom line. For example, if your business wants increased production rates, then you should monitor valves at limit. But if the focus is on energy reduction, then a better place to start might be oscillations in temperature loops. Always ensure that you can establish a link between the business goal and the metrics you’re using. Prioritize. Use both economic and technical data to focus efforts. Does it matter if the waste processing surge tank has wild variations? Probably not. Concentrate on processes and controllers that impact key parts of the business. Make sure you have a way to identify which control loops affect the bottom line and which ones don’t. Take focused action. Software alone doesn’t make the plant run better. You must take action. The metrics will show you where to take action but it’s up to you to ensure action is taken. Use a spreadsheet to track all actions needed, including valve repairs, instrument calibrations, loop tuning and control strategy changes. Assign responsibilities and deadlines for each item — then follow up to ensure they get done. Track and report results. Putting results in business terms is a weak spot for most engineers. However, always make an effort to track and report your results in such terms. For example, if you’ve reduced variability by 50%, has this resulted in fewer rejects or less recycled product? It’s usually much easier to assign monetary value to a reduction in rejects than to generic “variability reduction.” Once you have some bottom-line benefits, broadcast your success. Make sure that operations managers know. They may have other processes that need the same type of improvement. George Buckbee, P.E., is director of product development at ExperTune, Inc., Hartland, Wis. E-mail him at[email protected].References1. Ronka, M. and G. Buckbee, “Control Performance Supervision Enhances Revamp,” Chemical Processing, April 2008, www.ChemicalProcessing.com/articles/2008/052.html.2. Boyes, W. and G. Buckbee, “Why do you Need Performance Supervision,” Control, April 2006, www.ControlGlobal.com/articles/2006/070.html.3. Brisk, M. L., “Process Control: Potential Benefits and Wasted Opportunities,” 5th Asian Control Conference, Melbourne, Australia (July 2004).