Advanced process control metrics: Closing the loop on APC performance
For more than 30 yr, advanced process control (APC) has established itself as an important and relatively routine and valued part of the industrial process control and operation landscape.
For more than 30 yr, advanced process control (APC) has established itself as an important and relatively routine and valued part of the industrial process control and operation landscape. Most often, APC takes the form of model-based multivariable predictive control (MPC) technology. Though largely eclipsed by MPC, advanced regulatory control (ARC) technology has retained an essential role where smaller, higher-frequency, more specialized or higher-availability (base-layer) process automation solutions are desired.
APC comprises MPC and ARC.
Despite the widespread adoption of APC and its respected role in modern process automation, it continues to be viewed as a demanding and challenging technology to own and use. It continues to have high cost and maintenance, high-skill support requirements, and often contributes to, rather than reduces, operational complexity. Lacking proficiency in any of these areas is known to run the risk of APC applications rapidly degrading into low utilization and poor performance. A critical component of technology management, which remains limited or missing in the case of APC, is appropriate metrics—known as key performance indicators (KPIs). Metrics can “close the loop” on successful use of the technology, providing feedback to monitor and correct associated decision-making and work practices, leading to continuous improvement and sustained top-quartile performance. The following work presents readily available solutions for APC metrics to close this gap. Note: The term manipulated variable (MV) is used since it has become widely understood in industry. MV can refer to an MPC or ARC handle or to a single-loop controller output, which can be thought of as an MV in a 1 × 1 matrix.
Characteristics of a good APC metric.
Every technology has its own measurement and management challenges. Given the history and track record of APC, important characteristics of effective metrics include:
- Quantifiable (e.g., an actual number that can be historized, trended and reported)
- Transparent (e.g., a straightforward equation with general industry acceptance)
- Responsive and intuitive (e.g., can be seen to change reasonably with relevant process changes)
- Granular (e.g., applied at the variable level, but not at the overall controller or application level)
- Key (e.g., should address a minimum number of the most essential properties to facilitate high-level, face-value assessment)
- Automated (e.g., preferably a real-time online calculation, as semi-automated or periodic reporting brings many drawbacks and should be avoided).
Utilization metric.
This article proposes two metrics—utilization and benefits—that are necessary and sufficient for effective top-level monitoring of APC applications. The utilization metric is based on MV movement. This does not pertain to how much, how fast or how far an MV is moved by the APC application, but whether the MV is being moved at an appropriate time frame. This method is simple, intuitive and reliable. Just as a vehicle is being utilized and doing work only when it is actually moving, an MV is only being well-utilized if it is actually being moved—at least occasionally. The same can be said of single-loop controllers, which have outputs that do not move when in manual mode. Mathematically, in control systems, process data historians or spreadsheets, utilization can be expressed as in Eq. 1.
UX(t) = If (X.Value(t) <> X.Value(t-PUX),1,0) (1)
where:
UX(t) = Utilization of variable X at time t (1 = True, 0 = False) can be averaged over time to yield a 0%–100% value
X.Value(t) = Value of MV X at time t
X.Value(t-PUX) = Value of MV X one time previously
PUX = Utilization period for MV X; defaults = 1 hr, 1 shift or 1 d.
FIG. 1. APC earns benefits by holding the process closer to constraints than manual control, even as the process operating point and constraint limits move dynamically.
For MVs that normally move continuously, a default period of 1 hr is appropriate. For MVs that move only periodically based on process conditions, a period of 1 shift, or 1 d, may be appropriate. Using standardized time periods for various classes of MVs, where the choice of period for each MV is intuitive for people knowledgeable in the process, can add to ease of use, consistency of results and comparability of the metrics across different applications, company sites and industry.
Earnings metric.
The economic benefits of APC are also directly related to utilization and to the range of MV movement. Using APC, MVs can hold the process closer to constraints—the automated nature of the application means the MVs can back the process away from encroaching constraints in a responsive and reliable manner—should a disturbance or change in process conditions make this necessary. Likewise, the MVs can be automatically moved to capture additional earnings as process constraints move farther away. However, in the absence of an automated APC application, operators tend to adjust the MVs to keep the process relatively distant from constraints to provide greater response time and a wider margin of error for manual intervention. This greater distance from constraints typically translates into a cost penalty in terms of an incremental loss of capacity, yield, and energy efficiency. This is the most traditional and fundamental (and still one of the best and most valid) ways of summarizing the essential role of APC in real-time process control and optimization (FIG. 1).
Accordingly, APC benefits are reflected in the average value of an MV, relative to its economically minimum value, over a time frame that is based on how often operators typically adjust the MV manually in the absence of APC. The idea is that, lacking APC, the MV would typically be kept (at best) at the safe (minimum economic) value for the duration of the time (Eq. 2).
BX(t) = (Average(X.Value(t,t-PBX)) – Minimum (X.Value(t,t- PBX))) × CX × PBX × UX(t) (2)
where:
BX(t) = Benefits of MV X during time t, $ (dollars), can be totaled over time to yield total benefits, such as $/yr
PBX = Benefit period for MV X, default periods = 1 hr, 1 shift or 1 wk
CX = Cost, price or dollar value of MV X, $/unit.
This is an inherently conservative and realistic method of estimation because it considers the periodic manual MV adjustments that occur when APC is off or in the absence of APC. This method can be used to estimate the benefits of a proposed APC application, based on the data history of the proposed MV(s).
FIG. 2A. MV movement provides a simple and reliable basis to monitor APC utilization and benefits at the individual variable level. MV1 exhibits 100% utilization throughout the year.
As with the utilization metric, using standardized time periods can add to ease of use and to more consistent and comparable results. The utilization and benefit metrics will normally track together, since both derive from MV movement. The utilization metric is more useful in day-to-day support to flag problem variables. It is much more telling than the traditional service factor, which is often misleadingly true, even for MVs that are not actually being moved or utilized.
The benefit metric is useful as a standardized method of earnings estimation, which remains an important unmet industry need, to justify application cost of ownership and ongoing reinvestment decisions. The benefit metric also aids in understanding the relative value and performance of different variables, technologies and applications, and to prioritize the ongoing allocation of process control resources.
Case study.
Two MVs from an MPC application for 1 yr are shown in FIGS. 2a and 2b. MV1 had 100% utilization throughout the year. This can be immediately gleaned from its continuous movement, as well as from the continuity of its benefits. This MV was being utilized essentially every hour of the year, carrying out work and earning benefits.
FIG. 2B. MV movement provides a simple and reliable basis to monitor APC utilization and benefits at the individual variable level. MV2 was under-utilized for much of the year.
In FIG. 2b, MV2 appears underutilized for a large portion of the year. Most of its earnings occur from June–October. However, for the first half of the year and for the last 2 mos of the year, this MV was moved only very occasionally, perhaps only manually or by adjusting limits. The utilization metric can help identify problem variables and initiate follow-up, though it does not necessarily provide diagnostic answers.
This method can be readily implemented within control systems, data historians or spreadsheets, including retroactively requiring only that the MV be previously historized. These metrics are equally straightforward to deploy for multivariable controller MVs, ARC handles and single-loop controller outputs, allowing all three to be monitored and compared on a common basis.
In literature, APC application benefits were estimated based on MV movement, but utilization was estimated based on the percentage of controlled variables (CVs) at limits.1 This approach has its roots in multivariable theory and was adopted by several international refiners. However, experience has brought the insight that utilization and benefits are closely related. Utilization, as well as benefits, can also be more simply and reliably estimated directly from MV movement.
Performance metrics.
One reason for the slow emergence of effective high-level APC metrics has been industry’s continuing priority to better understand model-based control performance, which has proven to be more complex than expected. Control performance refers to measuring APC performance based on the behavior of the CVs, rather than the MVs. Several reasons explain the unexpected difficulty:
- Process models (e.g., actual process responses, normally identified through a plant step test) were expected to remain reliable for a lifecycle period of 2 yr–5 yr. However, experience found that many or most process responses change frequently, and often dynamically as a result of the very disturbances under control. When actual process responses change relative to the embedded controller models, this poses a fundamental conundrum for model-based control performance.
- Traditional error minimization performance criteria—known as aggressive profit maximization—has proven to be inappropriate for most multivariable control applications. In industrial process operation, transient error minimization is usually neither a large earner nor a desirable behavior. It is usually necessary to move the process more carefully, observe established process “speed limits,” and minimize overshoot and oscillation, which have criteria that conflict with aggressive error minimization criteria.
- Many factors affect a process’ CV, so that isolating and identifying the contribution of APC based on CV behavior is highly problematic (FIG. 3). Since so many parallel efforts are normally ongoing in operating facilities to effect process optimization, most CV-based metrics will yield sensible results even if the APC application is off.
FIG. 3. Many factors affect a process’ CVs, making it problematic to identify the specific contribution of APC. Conversely, MV movement is a direct measure of the APC contribution to process optimization, making it a much more suitable choice to form the basis of high-level APC metrics.
Meanwhile, MV movement is a clean measure of the contribution of the APC application. These considerations illustrate the wisdom and necessity of starting from MV movement—not CV behavior—as the basis for top-level APC metrics.
The path forward.
Standardized and effective top-level APC metrics are missing from the overall process control, operation and APC landscape. This continues to contribute to industry delay in finding answers to important questions, such as: What is the best approach to matrix design? Which variables are earning the most benefits? When is ARC appropriate? What are basic loops earning relative to advanced loops? To what extent does model-based control outperform traditional feedback control methods, which are much easier and less expensive to employ?
More effective APC metrics will help answer these questions. More rapid continuous improvement of APC practices and more efficient use of limited APC resources help close the loop on the successful use of the technology. After more than 30 yr of vigorous deployment of MPC applications, industry needs more transparent metrics to further advance understanding and progress in this critical sector of modern process automation. HP
LITERATURE CITED
- Kern, A., “Online monitoring of multivariable control utilization and benefits,” Hydrocarbon Processing, October 2005.
The Author
Kern, A. - APC Performance LLC, Red Lodge, Montana
Allan Kern is the owner of APC Performance LLC. He has more than 30 yr of advanced process control (APC) experience and has authored numerous papers on effective, low-maintenance APC solutions. He is the inventor of patented Rate-Predictive Control (RPC), industry’s only inherently adaptive control algorithm; and of XMC, industry’s only model-less method of multivariable control. Mr. Kern holds professional engineering licenses in control systems and chemical engineering, is a senior member of the International Society of Automation (ISA) and is a 1981 graduate of the University of Wyoming.
Related Articles
From the Archive