November 2021

Special Focus: Process Controls, Instrumentation and Automation

The path forward for process automation: Multivariable control as core competency

In process automation, multivariable control has always been considered an area of specialization and a luxury for those companies with sufficient scale and resources to justify its high costs of ownership.

Kern, A., APC Performance LLC

In process automation, multivariable control has always been considered an area of specialization and a luxury for those companies with sufficient scale and resources to justify its high costs of ownership. However, one of the key insights to emerge from the past few decades is that multivariable control is not a specialization. It is, instead, a fundamental aspect of virtually every process operation—ask any operator or process engineer.

The question is not whether or not a process has multivariable control, but whether it is carried out manually (by the operating team) or automatically (by a method of automated multivariable control). Moreover, multivariable control can be more easily understood and readily mastered by observing traditional manual multivariable control principles and practices throughout industry, rather than by struggling with the conventional, highly specialized multivariable control paradigm.

In the coming two decades, the dominant process automation activity will most likely be multivariable control. Industry has already undergone several decades of multivariable control activity, but the conventional paradigm (model-based multivariable control and real-time optimization, or MPC) has never evolved into the core competency needed to meet industry’s widespread scalable needs. Much work remains to be done, and more agile tools are needed to do it.1 With traditional manual multivariable control as a guide, automated multivariable control core competency will finally emerge.

What is multivariable control?

When console operators make setpoint and output adjustments to single-loop controllers in the course of process operation, that is manual multivariable control. Operators typically make multiple controller adjustments in concert, with multiple constraint limits and optimization goals in mind. This is the most practical and intuitive way to understand the role of multivariable control in process operation—the coordinated adjustment of groups of related controllers for constraint management and process optimization purposes—whether carried out manually by operators or with the aid of automated multivariable control technology.

Operators make controller adjustments based on many considerations, including their training and experience, feedback from process alarms, input from process engineers and supervision, and operating orders that typically flow down the chain of command from daily site-wide production planning and optimization (PP&O) meetings. The entire operating team works together in this way to meet production and optimization goals, while safeguarding reliable process operation. This inherent activity can be found in place on virtually every process operation in industry because the multivariable nature of most processes demands it, and because it is actually the most natural, effective, state-of-the-art way to manage and operate complex industrial processes.

In this light, manual multivariable control has always been a core competency of the process industries, and each site’s level of success is dependent upon its ability to master it. To take process operation, automation and optimization to the next level, automated multivariable control must also become an industry core competency.

Model-based control and real-time optimization

Model-based control and real-time optimization are double-edged swords. While powerful in concept, they have proven to be too complex and fragile—primarily due to the unreliability of the models upon which they depend—to evolve into low-cost, low-maintenance, high-agility, long-lifecycle technologies (i.e., into core competencies). Several articles report that more than half of MPC applications perform at pre-installation levels or have been removed within 18 mos–24 mos.2 This combination of high cost, high maintenance and short lifecycle is increasingly considered unsustainable.

Model-based control and real-time optimization are often considered synonymous with multivariable control, but they are actually just part of the way conventional MPC solves the multivariable control problem. Other ways exist to do multivariable control that do not entail these elaborate and fragile methods. For example, FIG. 1 outlines a method of multivariable control that basically mimics (automates) traditional manual multivariable control methods that do not depend on detailed models or embedded optimizers.

FIG. 1. The basic method operating teams have always used for manual multivariable control, which notably does not require process models or real-time optimizers. This method can be automated to provide a less fragile and more readily mastered multivariable control tool.

In retrospect, industry should not be surprised by the pitfalls of large-scale model-based control, because the same lesson was learned in single-loop control long ago. Feedforward is the single-loop equivalent of model-based control and is equally powerful—in concept. However, industry uses feedforward very sparingly—less than 5% of loops—due to the added cost, risk and maintenance associated with any model. MPC was expected to overcome this limitation by virtue of more careful plant step tests and better model identification tools, but instead industry discovered that models change frequently for a wide variety of reasons, so that model identification is a shifting target. Rather than solving process control tuning and performance issues once and for all, MPC took application support to vast new levels. Fortunately, experience and insights now show that multivariable control, just like single-loop control, can be readily accomplished using feedback control algorithms with or without the selective use of key feedforward models.

“Real-time” optimization, defined as optimization deployed at the control layer in conjunction with multivariable control, also may be on a declining trajectory. Aside from its part in MPC complexity and fragility, optimization at the control layer cannot begin to compare with modern optimization practices at the business PP&O layer in terms of sophistication of tools, site-wide breadth of scope, appropriate optimization time scales, etc. At the same time, modern connectivity makes it easy to share information between layers, so that multivariable control can directly utilize PP&O results, rather than be burdened with its own inferior optimizer. As a further emerging concern, the high support and maintenance of model-based control and real-time optimization put them at odds with modern control network reliability and cybersecurity principles.

Benefits and metrics

Industry adopted the intrinsic value of closed-loop control over open-loop control decades ago, so that “loops in manual” is one of the most common process control metrics among top-tier operating sites today. Industry also adopted the idea that console operators should not be distracted by excessive numbers of low-value alarms, so that “bad actor alarms” is another nearly universal alarm management metric.

This says a lot about the number of loop interventions (i.e., the number of controller setpoint, output and mode changes) that occur daily at an operations console. High numbers indicate excessive multivariable loops in manual and/or bad actor loops. Such a “loop intervention” metric is transparent, responsive and embodies a wealth of APC measurement and management information [as a good key performance indicator (KPI) should]. For example, it measures the activity load on console operators; indicates the health of existing multivariable control applications; identifies missing/needed applications; and identifies top bad actor loops. Industry now has an “alarms per hour” guideline for effective operation; should industry also have a “loop interventions per hour” guideline? Loop interventions, not service factor, should be industry’s go-to metric for APC, along with the MV utilization metric.3

The benefits of closed-loop (automated) multivariable control are fundamentally the same as for closed single-loop control: more consistency and timeliness, fewer alarms and constraint violations, and greater optimization. As automated multivariable control evolves into a core competency and costs of ownership decline, many applications will be justified based solely on the intrinsic value of closing the loops and improving the metric, just like traditional single-loop control practice. At the same time, more effective automation of many multivariable control applications will continue to bring the large-scale economic benefits often associated with conventional MPC (FIG. 2) (TABLE 1). HP

FIG. 2. The APC tool gap. Traditional advanced regulatory control (ARC) has limited capabilities, while the high ownership costs of conventional MPC make it suitable mainly for large applications. This has left a large tool gap for industry’s many midsized multivariable control applications, which can be filled by more cost-effective and agile APC tools based on insights from traditional manual multivariable control practices.


  1. Kern, A., “Understanding multi-variable control (and industry’s missing advance process control metric),” 2020 AFPM Summit, Hydrocarbon Processing, August 24, 2020, online:
  2. Mayo, S. M., R. R. Rhinehart and S. V. Madihally, “APC maintenance scheduling—Part 1,” Hydrocarbon Processing, February 2020, online:
  3. Kern, A., “Advanced process control metrics: Closing the loop on APC performance,” Hydrocarbon Processing, November 2018, online:

The Author

Related Articles

From the Archive



{{ error }}
{{ comment.comment.Name }} • {{ comment.timeAgo }}
{{ comment.comment.Text }}