Organizational warning systems: a probabilistic approach to optimal design

The optimization of warning systems includes the choice of signals to be monitored, the sensitivity of the alert level and the response to these signals. In organizations that design or manage complex, risk-critical systems, the problem is not only the observation of a signal, but also its accurate communication to the appropriate decision maker, and the actions that he or she can take upon receiving the signals given the state of deterioration of the physical system at that time. In this paper, we describe an analytical framework focused on organization performance, based on decision analysis and probability, to design and optimize such a warning system from a management perspective. The probability of failure during the time it takes to observe a signal, decide of its importance, and communicate a warning, and the time elapsed between the occurrence of an accident initiator and the events that follow, including the action taken in response, are computed in parallel, using a dynamic probabilistic risk analysis, assuming a Markov (or semi-Markov) evolution process when appropriate. The analysis then allows optimization of the choice of signals to be observed, the design of the communication chain, the filtering of the information, and the response at the appropriate management level. The questions are thus: What is the value (in terms of risk reduction) of a warning system of given structure and procedures? And what is its optimal configuration? The optimization model presented here reflects the nature of the engineering system to be monitored, the communication structure within the organization that designs or manages it, errors that can occur in the communication phase, the best action(s) of the decision maker once he or she receives a warning signal, and the expected value of the costs, including those of system failure. The schematic example of inspection and maintenance of an airplane is used throughout the paper to illustrate, at a very simplified level, the creation and evaluation of this model.