Methods and best practices to map displays to operator decisions
HMIs should be designed to reflect the ways in which operators have to deal with information and solve problems.
The critical nature of display design in process control system operator interfaces is highlighted by the fact that a high proportion of incidents are caused by operator errors. Currently, control operator display design is typically based on the P&ID drawings because this approach is fast and easy. The problem is that displays designed by this method do not necessarily provide the right information to the operator at the right time. A new approach organizes the display to provide information supporting decisions to operate the plant optimally and return the plant from abnormal situations back to normal conditions. The new display design process starts with analyzing the decisions operators have to make, maps available information to these decisions, and allocates information to the different levels of displays. The new approach has the potential to reduce the number of incidents by improving operator decision making.
Display design challenges
There is an increasing awareness of the role that operators play in the incident-free operation of a plant. A recent report by the Chemical Manufacturers’ Association on the causes of incidents attributed 26% of the incidents to operator errors (see Figure 1). Some individual facilities have informally stated that they attribute a much higher proportion—even up to 66% of incidents—to operator errors. One of the most promising paths to reducing the number of operator errors is improving the operator interface, which consists of the set of graphic displays that allow the operators to view the process they are responsible for and take action as needed.
As a starting point, initial operator displays often mimic the process equipment shown on P&IDs. These initial displays include measurements, valves, other final elements, and control elements. They include enough of the process equipment and piping so that the process flow can be followed. Display navigation is added, allowing operators to follow the flow of the process quickly and drill into and out of detail as required. Over time additional related information from upstream and downstream processes is often added to the displays. The overall effectiveness of the system of displays depends greatly on the experience of the designers, the involvement of the operators, and the manner in which the graphic displays are structured.
A new approach to display design
Clearly, the purpose of the display is to enable operators to make more effective decisions and, in particular, to eliminate operator errors as much as possible. To improve the overall effectiveness of operator graphics, the Center for Operator Performance (COP) is investigating a different approach to display design. Instead of following equipment layouts, the display design is based on the decisions that control room operators make. As part of this approach a systematic review and characterization of the decisions made by operators and others in process manufacturing is used. A rating and clustering technique is used to map the available information to the decisions. Information is then allocated to different display levels.
A typical first step is to identify the key operator decisions for each major section of the process. In an initial study, key decisions such as, “Why have I lost hydrogen?” and “Why are my separator levels changing?” were asked. The decisions that were selected require multiple data values from the underlying process model. For example, “Have I lost hydrogen?” requires only checking hydrogen measurement and should be alarmed. On the other hand, “Why have I lost hydrogen?” requires the analysis of multiple data points. Decisions involved in shift handoff, key upset systems, and typical daily instructions were also considered.
In a test case, the following decisions were selected for the hydrocracking unit:
- Why have I lost hydrogen?
- Why am I venting so much?
- Why has the recycle gas changed?
- Am I maximizing preheat?
- Am I operating inefficiently?
- Can I increase charge?
- Is my feed system set up to produce desired product?
- Is my recycle compressor operating near optimum?
- Am I at risk for a temperature runaway?
- Why has the reactor temperature taken off?
- Why don’t I have enough feed?
- Are my reactors set up to produce desired product?
- Why am I not making the desired amount of light product?
- Why is the naphtha off spec?
- Why are my separator levels changing? and
- This should be displayed all of the time (cross-check).
What should be included?
One of the key challenges in designing a display is determining what information should be included in the display and how that information should be organized. There are thousands of data points in a typical unit, far too many to try to include on a display. A hydrocracker can have 3,000 tags, so it is important to define the ones that are most important. The procedure starts by dividing the unit into logical sections and then defining key data elements based on the process requirements and items that have important alarms and similar factors.
In many cases the data points must be combined to provide information that will aid the decision-making process. In this example, the number of tags was reduced from 3,000 to 120, with a total of 194 data elements. For example:
- Front End-High-pressure separator pressure PV
- Back End-Splitter top pressure output
- Back End-Splitter fuel gas pressure PV
- Utilities-Wash water flow PV
- Back End-Splitter top pressure PV
- Front End-Make-up compressor unit pressure mode
- Front End-Make-up compressor unit pressure output
- Front End-High-pressure separator pressure output
- Front End-Charge unit charge flow PV, and
- Front End-Furnace/heater (Both HT/HC RXS) fuel gas pressure PV
The next step is determining which among these data points are most important to the operators’ decision-making process. For each of these decisions, experienced operators were asked to rate the importance of key data elements. The operators rated each data point on a scale of 1 to 5 based on its importance to each decision.
5 – Critical or extremely important
4 – Very Important
3 – Important
2 – Somewhat important
1 – Not at all important, and
0 – Doesn’t exist on my unit
The responses from the operators were analyzed and each data element was assigned an average rating across operators for each decision. The results of the survey were then evaluated using a technique called cluster analysis to determine how the parameters should be organized. Cluster analysis sorts objects into groups where the objects in a group are similar to one another and different from the objects in other groups.
As shown in Figure 2, determining the number of clusters may require a few attempts to find the ideal number. As a general rule a good starting point is four or five clusters. The cluster groups are used to define the display hierarchy needed to move from high level situational awareness across multiple decisions down to individual data points on mimic displays. The following procedure is used to form five clusters:
- Each observation is in a separate group
- Each two observations which are closest together are combined to form new groups
- The distance between the remaining groups is calculated
- The two groups then closest together are combined, and
- This process repeats until only five groups remain.
To help explain cluster analysis, an example will be provided of how this method can be used to group cities based on demographic, economic, and environmental variables. The data set shown in Table 1 was used for this example.
The data set above is shown in Figure 3 as a dendogram, a tree diagram frequently used to illustrate the arrangement of the clusters produced by hierarchical clustering. The bottom row of nodes in the dendogram represents individual observations, and the other nodes are used to define the clusters to which the data belong. The vertical distance from the common nodes to the bottom row is inversely proportional to the similarity of the members of the group. In other words when distance is small, the cities are more similar. For example, the dendogram shows that Boston and Washington are the most similar cities and that the most similar combination of three cities is those two plus Atlanta. The general shape of the dendogram suggests that the cities can be organized into two groups:
- New York, Chicago, Boston, Washington, Atlanta, and San Francisco
- Los Angeles, Houston, Phoenix, and Miami
Since Group #2 contains cities that tend to be located in warmer areas, climate plays an important role in grouping the cities when the farthest neighbor method is used.
Applying cluster analysis to the hydrocracker example
The next step is to apply cluster analysis to the hydrocracker example mentioned earlier. Figure 4 shows the dendogram for the hydrocracker example.
The centroid is the average value of all members of the cluster on a particular variable. In the hydrocracker example, cluster 1 had an average response of 3.1 and cluster 2 had an average response of 1.8 to question 1 – Table 2. It is important to make decisions based on the patterns across the decisions and data instead of relying solely on the highest decision scores. For example, seven of the decisions had high decision scores in clusters 1 and 3, which both should play an important role in developing those decisions.
Cluster analysis separated the low impact data into clusters 2 and 4. The example below from cluster 4 suggests that these decisions can be separated out and handled on different overview or lower level displays.
Front End-Furnace/heater ID/FD fan status 0
Front End-Amine contactor/absorber level SP
Front End-Amine contactor/absorber level PV
Front End-Amine contactor/absorber level mode
Front End-Amine contactor/absorber level output
Back End-Fractionator reboiler steam flow SP
Back End-Fractionator reboiler steam flow PV
Back End-Fractionator reboiler steam flow mode
Back End-Fractionator reboiler steam flow output
Designing the display
Operators can be more effective when all of the related information they need is included on the same display. Data overload can be limited by defining the information that is needed and eliminating everything else. So it’s important to determine what information should be at what level in the display hierarchy. The starting point is to determine what information is needed for high-level situational awareness. Display designers should also define what other displays operators need for direct access to related information. Information should be organized in terms of its value for high level overviews and its relevance in drilling down for more detailed information from the overview screens.
In the hydrocracker study, data was ranked across all decisions to determine the information for different display levels. Criteria were identified for determining which data elements should be included based on their ratio ranking as well as their average rating across all decisions. For example, in the COP study no more than 40 data elements are included in level 1 and all data elements in level 1 must have an average rating across all decisions of at least 3.0.
- Level 1—High-level overviews and alarms
- Level 2—Primary operation (unit-wide operation)
- Level 3—Secondary operation (task-oriented operation), and
- Level 4—Process detail or support graphic.
Once the parameters are organized into appropriate displays, the next step is to consider alternative ways to present the information, for example, graphical, textual, tabular, and auditory. Certain data is best visualized one way to support one task or set of tasks and better visualized a different way to support different tasks. As an example, consider the ways in which telephone numbers can be presented. If someone asks you to remember 1-800-677-4992, you’ll probably have to work at it. However, once the number is situated in long-term memory, dialing the number is a simple task. Contrast that with trying to remember 1-800-MR-PIZZA. That representation of the same sequence of button pushes is easier to remember; however, it is harder to dial.
While low-level P&ID graphics based on plant layout and equipment are often easy to design, they may not provide the support needed for everyday supervisory control and situational awareness. The new method described here can help designers to consider the decisions operators are making, identify the content needed to support those decisions, and provide a systematic method to organize the content. The end result should be that the displays will help operators make better decisions and fewer errors.
Jennie J. Gallimore, PhD, is professor of industrial and human factors engineering at Wright State University. Cindy Scott is DeltaV product manager for Emerson Process Management. Mark Nixon is director of applied research at Emerson Process Management.
- Most HMI designers create operator screens using the P&ID as the main design element. This works, but doesn’t consider what information is most critical to operators.
- Operator decisions have a major effect on preventing upsets or making them worse, so delivering the right information quickly and clearly can minimize potential problems.
- Understanding basic decision making techniques can help an HMI designer create highly effective graphics.
- Find out more about Wright State University at www.wright.edu
- Watch an on-demand webcast about HMI design at www.controleng.com/media-library/webcasts.html
- The research in this article was conducted by the Center for Operator Performance, a diverse group of industry, vendor, and academic representatives addressing human capabilities and limitations with research , collaboration, and human factors engineering. Its mission is to raise the performance of operators and improve health, safety, and environmental awareness. Learn more at http://operatorperformance.org
• Learn more about Emerson Process Management at www.emersonprocess.com
|Search the online Automation Integrator Guide|
Case Study Database
Get more exposure for your case study by uploading it to the Control Engineering case study database, where end-users can identify relevant solutions and explore what the experts are doing to effectively implement a variety of technology and productivity related projects.
These case studies provide examples of how knowledgeable solution providers have used technology, processes and people to create effective and successful implementations in real-world situations. Case studies can be completed by filling out a simple online form where you can outline the project title, abstract, and full story in 1500 words or less; upload photos, videos and a logo.
Click here to visit the Case Study Database and upload your case study.