Safety performance: A control theory perspective on the use of incidents and accidents in measuring safety

Safety, defined as the absence of risk, prompts many decision-makers to assess the performance of high risk systems based on reported events through incidents and accidents reporting processes. Consequently, zero-accident objectives have become widely established standards, both in workplace and process safety. Scientific literature has already highlighted the limitations of such an approach, describing it as not only reactive but also potentially conducive to manipulation of event reporting to meet predefined safety targets, particularly when these targets are linked to financial or institutional incentives. By formalizing safety as a control problem, this paper seeks to provide a novel perspective on the debate by employing the concepts of observability and controllability, introduced by Kalman in the 1960s as part of modern control theory. Defined as central criteria for characterizing dynamic systems, these concepts are now extensively applied to guide the design of control system architectures and performance measurement frameworks in several disciplines: biology, electronics, AI… This study leverages these concepts to evaluate the limitations of using events for designing control schemes in high-risk systems, with the railway system serving as an illustrative example. By integrating theoretical reflection grounded in control theory and empirical demonstration using real-world data, this work aims to draw attention to the need for a clearer definition of the validity domains of performance measurement tools based on event accounting (incidents and accidents), as well as to identify avenues for improving their reliability for policymaking.

Keywords: Safety performance, Safety index, railway safety.