Tracking the complexity of interactions between risk incidents and engineering systems

Large-scale engineering systems increasingly contend with massive databases of incident reports pertaining to risk and security. With appropriate processing and analysis, these databases can be offered as inputs to system risk assessments. This paper develops an automated detection of anomalies in an evolving set of incidents associated with engineering systems. It adopts a definition of a system-to-incident interaction, which associates one or more system components with an incident. An interaction is direct when a specific association of system and incident is clear, e.g., discovered in the text narrative of an incident report. An interaction is indirect when the relationship is not obvious and builds on how components relate to others in the system. Statistical process control (SPC) is adopted to analyze a sequence of incident reports. The variables—(i) interaction entropy and (ii) interaction informativity—are used in control charts to track the evolving complexity of the frequency counts of interactions. An example describes risk identification for an oil distribution system. A probability-based characterization of the strength of interaction is introduced as a direction of future effort. © 2005 Wiley Periodicals, Inc. Syst Eng 8: 262–277, 2005