Detecting and responding to plant disturbances

Article in October 2006 edition of Ergonomics journal
Title 'Human process of control: tracing the goals and strategies of control room team'
By J Patrick, N. James and A. Ahmed

A study from the nuclear industry. Five shift operations teams were evaluated using a process simulator. Each team was made up of two control room operators and a supervisor.

The simulator was a full-scale mimic of operations.

The teams were set about normal operations and then asked to carry out a routine task of changing over boiler feed pumps. Whilst doing this task a small leak was initiated on one of the pumps. And on top of this a spurious fire alarm in an office was activated that required some action from the supervisor, although not the operators.

The simulated leak was considered to be a plant disturbance that would not create an immediate alarm. However, the leak would cause a drop of level in the dearator, which if undetected would cause a low level alarm and eventually a reactor trip.

The time taken to detect the level drop was recorded. The results were
Team A - 9 min 3 s
Team B - 4 min 6 s
Team C - 1 min 57 s
Team D - 6 min 3 s
Team E - 2 min 30 s
Expert judgement on the site that this scenario should be detected within 2 minutes. Therefore, from the results only Team C was successful.

To explain the long time taken to detect the event the researchers looked at how much attention the operators paid to the routine pump changeover task. Apparently, the normal procedure was that only one operator would be involved in the task, leaving the other to monitor the plant and the supervisor would have minimal involvement. The actual results for proportion of time spent on the task were
Team A - Op1 74% Op2 64% Super 12%
Team B - Op1 81% Op2 53% Super 0%
Team C - Op1 26% Op2 00% Super 16%
Team D - Op1 95% Op2 90% Super 44%
Team E - Op1 98% Op2 93% Super 23%
Clearly, with the exception of Team C, both operators were heavily involved in the routine task, and it is no wonder that they took so long to detect the leak because they would have had little time to monitor the rest of the plant. This is further exacerbated by the fact that in some teams the supervisors got quite involved in the task (which should not have been necessary) taking them away from the main supervisory tasks.
Delving deeper the researchers also found that in most teams, not only did they spend relatively little time monitoring the plant, when they did they were fairly poor at it. In other words the monitoring they did was not good enough to detect the leak quickly.
Overall the finding is that operators easily become fixated on procedural tasks at the expense of the wider, continuous task of monitoring. Also, that supervisors do not tend to intervene to re-orientate the operators (in this case making sure only one is involved in the routine task and the other is concentrating on monitoring) and also have a tendency to get involved where it should not be necessary.

Having detected the leak, the teams had to diagnose its cause and take appropriate action. This was also evaluated. The results for diagnosis and control were
Team A - 6 min 11 s
Team B - 3 min 23 s
Team C - 4 min 46 s
Team D - 1 min 58 s
Team E - 4 min 48 s
In trying to diagnose the problem each team developed a number of hypotheses. Interestingly the operators generated most of these, with relatively little input from the supervisor. It had been assumed that supervisors would lead diagnosis, which clearly did not happen.
Unfortunately none of the hypotheses covered the actual cause of the problem and none of the teams diagnosed the cause of the problem correctly from the control room and needed information from the plant (i.e. someone looking where the water was leaking from).
The researchers identified that the teams took significantly different approaches to dealing with this phase of the scenario in the way they spilt their time between problem solving, mitigating the consequences of the leak and keeping an eye on the rest of the plant. They concluded that none of the teams gave problem hypothesis generation and testing a high enough priority. This suggests a need for more training in diagnosis.

Andy Brazier