NEW RESEARCH SUGGESTS ALARMS ARE NOT ALWAYS BENEFICIAL

NEWSLETTER ARTICLE

True or false: Adding alarms always improves an operator's performance. True you say? Guess again. A long held axiom by many people designing control system interfaces is that alarms are always beneficial, regardless of the situation. Could it be possible that this is not always true? Are there conditions in which the benefits of an alarm are outweighed by its costs? Are there cases in which adding an alarm could actually be detrimental to human performance? New research suggests the answer is yes. A startling consequence of the new information is that adding alarms may not make a system safer, but actually more dangerous. Just how much safer is your process unit with all of those alarms that were added on a recommendation from the last HAZOP?

In a recent study, (Elvers, G., Elrif, P. (1997). The Effects of Correlation and Response Bias in Alerted Monitor Displays, Human Factors, 39(4), 570-580), the relationship between the overall system sensitivity in detecting anomalies and the overlap of information monitored by humans versus alarm systems was investigated. In the study, humans worked in conjunction with automated signal detection systems (or alarm systems) to identify abnormalities. The study evaluated the relationship between dependence and independence of information monitored by humans versus alarm systems and the effects on overall system sensitivity (or ability to detect abnormalities). Results of the study suggest that if the operator is already monitoring the information, presenting an alarm on the information indicating an anomaly will be of little benefit and can even be detrimental. The results of the study indicate that the ability to detect abnormalities decreases as the overlap of information monitored by the human and alarm system increases.

In a real life setting, if an operator is having problems with a particular section of a plant and is working to resolve the problem, then continuing to present alarms telling him of the problems in that section will not help his performance. The research suggests the additional alarms are probably degrading his performance since the new alarms present no new information and impose a cost to his mental resources in processing the alarm.

In chemical processing plants this situation is all too familiar for control board operators. Cases where a pump shutting down triggers multiple alarms indicating that the pump has shut down are common. After the operator has diagnosed the pump loss and is working to resolve the failure, the additional alarms provide no new information to the operator but impose the cost of having to process the additional information. It is in these cases where alarm system designers need to use alarming logic to reduce the alarm system burden.

Another issue raised by the investigators is how selection of the alarm's criterion can influence the human's response bias. When an alarm's criteria (similar to the trip point) is set to be liberal and produces a lot of false alarms, the human attempts to compensate by becoming more conservative (and dismiss more of the alarms as being false) with a lower false alarm rate. Conversely, when an alarm's criteria is set to be conservative and produces less false alarms, the human adjusts and becomes more liberal (accepts more of the alarms as being true) and more willing to signal false alarms.

In a chemical processing plant with hundreds of alarms, the alarms with the liberal criteria are usually the ones that operators simply acknowledge without further investigation. The alarms that are conservative are the ones that operators further investigate and try to correct. If your operators are simply acknowledging alarms and not making many follow-up control changes, then their alarm system likely has a liberal bias.

Are there dangers in having a liberal (a lot of false alarms) or conservative (few false alarms) alarm system? The answer is yes. If an alarm system has liberal settings that produce a lot of false alarms, the operator may begin to view the alarms as unreliable. It is quite possible that the operator might silence the alarm without the necessary investigation and miss a true alarm event. False alarms also have the effect of adding to the operator's workload since they still have to be processed.

On the other hand, an alarm that is too conservative poses a danger of missing some of the events that should be alarmed. Another problem of a conservative bias is an alarm that provides the operator with insufficient time to react and correct the alarm condition.

In real life examples, such as a chemical processing plant containing hundreds of individual alarms, there is usually a mixture of alarms that have liberal and conservative biases. In this mixed situation, there is a real danger of liberal alarms creating a masking effect for the conservative alarms. In one instance Beville observed, a slide valve failure resulted in 101 alarms actuating in a one hour period. Of the 101 actuations, 96 were the result of a single alarm cycling in and out of alarm state. The other 5 were unique alarms that had also actuated. In this instance, the operator was working to resolve the slide valve problem and was intimately aware of its condition, and so the cycling alarm only had the effect of increasing his workload and masking the new alarms.

To attain a system with the highest sensitivity, those configuring an alarm system should select the alarm trip point such that the alarm criterion is neutral-to-slightly conservative.

The results of the study highlight why it is so important to judiciously select alarm points. Alarms should not be blindly added to a control system in the hopes they will automatically improve operators' performance. Under a number of circumstances, the additional alarms will actually hurt the system's performance instead of help.

Copyright 1998 Beville Operator Performance Specialists, Inc., All Rights Reserved