No Alarms and no Surprises: How Qualitative Data Informs Bayesian Inference of Anticipated Alarm Sounds

Abstract

Operators of control rooms may become desensitized when they are exposed to a large number of alarms. Minimizing the number of non-actionable alarms can reduce this safety hazard. However, existing practice fails to address how operators anticipate system events, and discards valuable feedback signals. We observed five operators in three satellite control rooms to assess the functionality of their alarm design. We recorded 140 auditory signals, of which 31 were labeled as alarms. Two of these alarms were actionable. Transitional Journey Maps were used to represent qualitative interview data and observation data. Statement analysis revealed that operators anticipated the majority of the alarms. Bayesian inference related anticipation to response behavior. Case-by-case analysis identified various non-actionable anticipated alarms as feedback signals. Existing practice marks these signals as nuisance, which seems inappropriate. Therefore, this study shows that qualitative data analysis is essential for interpreting quantitative data on alarm responses.