This research in this paper considers the evidence on the success of alarm reduction strategies reported in the open literature. Despite strong beliefs to the contrary, the empirical evidence suggests that alarm reduction strategies have not been as successful as initially expected. This seems to be due to the fact that alarm reduction strategies actually deprive process control operators of information. In order to determine the ability of people to sift through alarm information, a study of alarm detection with three ratios of target to non-target alarms was devised (i.e. 2%, 6%, and 10%) and the information was presented at three rates (i.e. 1 second, 4 seconds and 8 seconds). The results show that the ratio of target alarms has no effect on detection performance, but the temporal rate does. Given that process operators are rarely required to acknowledge alarm information in realtime, it is suggested that more emphasis should be placed on initial definition of alarms and better presentation methods, rather than attempts to block the flow of alarms that have already been triggered. KEYWORDS: Alarms, False Alarms, Alarm Reduction, Signal Detection.
PROBLEMS WITH ALARMSThe inappropriate presence of alarms can cause substantial problems for the process operator (Woods, O'Brien & Hanes, 1987). Typical problems are; the avalanche of alarms during a major transient or shift in operating mode, standing alarms, alarm inflation, nuisance alarms, and alarms serving as status messages (Stanton & Baber, 1995; Woods et al, 1987;Hoenig, Umbers & Andow, 1982;Andow & Lees, 1974). In a major incident, alarm presentation rate may be somewhere between 50-300 alarms a minute in a nuclear power station (Hickling, 1992). This may lead to problems for the operator in being able to identify and respond to alarms that are worthy of attention.Certainly, the limited number of actions that arise from alarms might suggest that there is a lot of redundant information present, e.g. Kragt & Bonten (1983) found that only 7% resulted in operator action. The problems appear to stem from the design of alarms based on 'normal' operation (Singleton, 1989) and on a 'one measurement -one indication' philosophy of presenting essentially raw plant data (Goodstein, 1985).However, a change in plant state would mean a change in what could be considered 'normal', i.e. what is 'normal' in start-up, maintenance, and shut-down? In addition, the oscillatory behaviour of a variable that is close to its alarm parameter can lead to distrust of the alarm. Hale & Glendon (1987) propose that a shift in confidence occurs, such that the next time the alarm occurs, the first hypothesis the individual will have is that the alarm is a false one. This lack of trust will grow with the number of false alarms experienced. Sorkin (1989) suggest that individuals are regularly disabling warning systems in locomotive, aircraft and process industries. All the examples are from situations where critical events could arise (e.g. potentially life threatening incidents). Sorkin sugg...