09 August 2001
Control System Reliability: Process Out of Control
by Lane Desborough, Randy Miller, Perry Nordh
Like scientists probing the mysteries of the deep, we've seen things rarely witnessed by human eyes. Control loops oscillating so badly they would be better off in manual, and textbook examples of square-wave sawtooth sticky valves. We've seen facilities where 80% of the control loops are in manual or a third of the loops are cycling at the same frequency.
After 10 years of studying control loops at hundreds of sites, we know process control has a long way to go, despite significant technical advances in performance monitoring.
Today the situation is basically unchanged since the landmark studies by Dave Ender in 1993: One-third of control loops are working OK, one-third are in manual, and one-third are disruptive to the process.
The reason? Process control engineers don't have the tools to allow them to continuously monitor all of their loops.
It's a Tough Process Life
As the facility's resident techies, process control engineers are busy facilitators of plant operation. Multiple interruptions, conflicting priorities, and responsibility for everything associated with the control system make it difficult for them to approach control loop maintenance in a disciplined way.
They are often ill prepared for the myriad activities that today fall on them to perform: fixing operator graphics, explaining processes to junior engineers, supporting business integration, and working with vendors. Control loops often receive very little attention-at least not until the process goes awry.
"We implemented a homegrown control loop performance monitoring system, but it was too hard to configure and maintain" is a common refrain of control engineers.
The role of these systems is not to be a configuration and maintenance burden, nor is it to give the control engineer the job of spending two hours a day deciphering esoteric statistics. Their role is to answer two simple questions for the control engineer: "What has changed, and what should I do about it?"
This role and these answers result from collecting time-series operating data, calculating relevant and actionable metrics, and presenting those metrics in the right context at the right time. Turning data into action is the crux of a real control scheme.
Watching the Stock Market
When we got into the process control business 11 years ago, we had an academic approach to performance monitoring. Even today, academic performance measures receive disproportionate attention in that few focus on measures that are practical.
After hundreds of site assessments totaling more than 50,000 loops, we can safely say people play the leading role in monitoring, with technology as supporting cast. Control loop performance metrics must be economical to compute from available data.
The metrics must provide specific indications of what has gone wrong, and they must drive the right behavior and ultimately result in the proper reaction. A well-designed control loop performance monitoring system is worthless unless one heeds its recommendations (one makes money by the well-informed buying and selling of stocks, not simply by watching the stock market).
Most control loop performance monitoring systems don't provide value in industrial practice for three reasons: First, necessary data is either unavailable or prohibitively expensive to obtain for every control loop in your facility. The foundation of successful performance monitoring is uncompressed data sampled faster than the loop time constant, an articulated business objective, and a determination of how the control loop impacts facility business objectives.
Second, reported metrics often fall short of identifying particular failures. Control loops have very specific failure modes, which demand focused reliable detection. Nonspecific but easy-to-compute toy algorithms blow a lot of smoke but don't show the source of the draft.
Third, performance metrics fall on the control engineer without regard to workflow issues. Does the engineer have to go into the system and spend a bunch of time determining which control loop is the problem, or does the system push a Top 10 list of problems?
More information is not better. We have seen complex, powerful statistical control loop performance reports stacking up on the control engineer's desk, only to be ignored.
Harden Promising Algorithms
Most facilities maintain control loops reactively. By studying how users monitor and maintain their own control loops, we identified how technology could automate the mundane yet critical tasks of data collection, metric calculation, and metric presentation.
By incorporating Internet connectivity to get technology to end users faster and to get data to a centralized server for metric calculation and report generation, the process reaps multiple benefits:
- Common and important control loop faults are identified using a single data warehouse, which is analogous to having one group looking at 50,000 loops rather than 5,000 groups looking at 10 loops each.
- Algorithm development is prioritized to address the most common and important faults first.
- New, focused detection methods can be rapidly prototyped and validated.
- Voluminous performance monitoring data can harden promising algorithms.
- Facility performance can be benchmarked.
- Centrally handling fault-prone aspects of monitoring minimizes on-site configuration and maintenance.
Through data collection automation and the Net, we've seen the diagnostic effort cut from a half-hour/loop to a half-hour/1,000 loops, including installation and configuration. IC
Figures and Graphics
- Forty Alarms in One Hour? Go Manual . . .
- Further Reading
- Keeping Score is the Goal
- Living is Easy and Other Strange Notions
Lane Desborough has multiple degrees in chemical engineering. He has written much about the area of controller performance monitoring and advanced control and has experience in the refining and petrochemical industries.
Randy Miller has a journeyman instrument mechanic license, a B.S. in chemical engineering, and an M.S. in process control. He has published many papers and holds a number of international patents.
Perry Nordh has a B.S. in electrical engineering, with an emphasis on computer engineering. His interests range from data transport and event handling to multivariable control and performance monitoring. All three work in the @sset MAX division of Honeywell Hi-Spec Solutions.