Systems Integration / Optimization

How to Build a Better Operator

Collaborative Research Effort Looks for Keys to Improved Effectiveness

By Aaron Hand

ABB A&PW 2012

If you want to keep your plant operators performing at optimal levels, how many alarms should you try to hold your system to? A commonly quoted number is 10 alarms in 10 minutes. But how do you know whether that's right for your group? What's the best way to present procedures to those operators so they know what to do when they do face an alarm situation? And what's the best way to train them how to follow those procedures?

There's no shortage of anecdotal evidence to provide suggestions for these and other questions related to target alarm rates, worker fatigue, training, display colors and graphics, information hierarchy, and any number of factors that could contribute either to operators' ability to keep a plant running, or the likelihood of them bringing it to its knees, noted Dave Strobhar, principal human factors engineer at Beville Engineering. But hard research can be harder to come by, so several players in the petroleum industry joined forces about five years ago to get the research done.

Strobhar presented several interesting findings from the open industry-academia collaboration, the Center for Operator Performance, at ABB Automation & Power World this week in Houston. The group has found, for example, that the alarm-per-minute average isn't as magic a number as some might believe; that providing procedures that span several units improves performance over per-unit procedures; and that operator error rates do not rise linearly over time as one might believe, but instead double on the ninth day of work after eight days of consistent performance.

Driven by operating companies, which are primarily in petroleum, the center also counts among its members three of the major control system suppliers: ABB, Emerson and Yokogawa. The group was founded at Wright State University in Dayton, Ohio, which has a strong background in human performance research, and is managed by Beville Engineering, which specializes in the analysis of operator performance issues in the refinery and petrochemical industry.

Research is done across a range of performance-shaping factors: interface/information systems, procedures/job aides, selection and training, automation/system demands, job design, and organization and staffing.

An ongoing project with associate professor Sandeep Purao at Penn State University is exploring the best way to present procedures, which are growing ever-more voluminous. Procedures tend to be organized by unit, each with its own set of procedures. "The problem is that in some cases I might have responsibility for more than one unit," Strobhar said. "I've seen console operators that have three sets of procedures in front of them because alarms are going off on three different units at the same time."

Modularizing those procedures instead could help companies tailor procedures to an individual operator instead of a unit, mixing and matching procedures that occur across several units. "If you could modularize them, you could tag them with certain attributes," Strobhar explained. "Then you could recombine them to create procedures for an individual."

Through an algorithm that converts procedures to text files, Purao and his group are able to build a set of heuristics that are parsed into a table, finding the key steps that tend to occur together, and creating a single task module from them.

One thing that came out from this research was the ability to identify gaps in procedures; certain steps that were missing key elements. The research also can help find opportunities are for procedural automation.

With Louisiana State University's Craig Harvey, the center has also researched the ideal alarm frequency within an operating system. As Strobhar notes, the oft-quoted number of 10 alarms in a 10-minute period seems reasonable, but why that particular number?

The group ran university students through a series of controlled experiments—with rates of one, two, five, 10 and 20 alarms per 10-minute period—to see how operator response time would measure up. "For one, two, five and 10 alarms in 10 minutes, performance was flat," Strobhar said. "But at 20 in 10 minutes, you see the alarms starting to queue up."

The first study ran the experiments in 10-minute bursts, so researchers then wanted to see how the numbers would be affected in one-hour time spans. In this case, they looked at 15, 20, 25 and 30 alarms per 10 minutes. Although there was a slight increase across the timeframes, it wasn't until 30 per 10 minutes that alarms started really queuing up more.

They also wanted to see if it made a difference whether the experiment subjects were students or professional operators. At a rate of 10 alarms per 10 minutes, the operators performed just a little better. But at a rate of 20 alarms for the same time period, the operators were twice as fast as the students. "That shows there's an experience effect," Strobhar noted. "You're not going to see a performance difference until stress comes up. That's when experience is really going to show."

The conclusion was that 10 alarms in 10 minutes is in fact a very conservative number. "A lot of companies are beating themselves over their heads trying to reach that number," Strobhar said, noting that with better operators, better displays and other positive factors, that number could be more like 20-25 alarms per 10 minutes.

Of course, everyone wants to know the best way to make a "better operator." In a study with Klein Associates, the center looked into whether the researchers could adapt the military's decision making exercises (DMX) to process plants. What they found was that it could be done easily with very short training sessions—one hour on occasion at the start of a shift, for example.

"You need to practice making decisions," Strobhar emphasized. "It's a skill like anything else." Through relatively simple, low-cost training sessions, companies have been able to keep their operators' skills honed, identify knowledge gaps and lost practices, and help build mental models.