Accidental knowledge: Using accidents and other project failures to inform research in systems engineering

Projects experience cost overruns, late deliveries, quality issues, cancellation, and accidents despite the best efforts of the systems engineering community. There is relatively little research on why systems engineering failures in general happen, but a substantial body of work on accident causation. Here, we investigate whether systems failures in general exhibit the same patterns of causation as accidents. We conducted a review of existing accident models to develop a model that could be applied to all types of project failures. Our model helped us to classify where the factors occur during the system development/system operation phases and which entity was involved in each factor. We analyzed 58 failure case studies. The failure cases span non-accidents, accidents, and dual failures. The sources for each subset had varying depth and scope of investigation. We developed a coding method to compare the factors between failure cases that broke each factor down into an “actor-action-object” structure. We further generalized the actions from the “actor-action-object” strings into control flaws so that we could analyze the failure cases at a high level. We analyzed the control flaws, actions, and actors for each failure case and compared the results for accidents and non-accidents. Of our results that we could not attribute to study biases, we found similarities and differences between project failure causation. We also identified which control flaws, actions, and actors were the most prevalent in the different types of project failures. Of all the actions, “failure to consider factor in system development” contributed most to non-accidents, while “failure to consider step in risk management” contributed the most to accidents. Of all the actors, “company management” contributed the most to non-accidents and accidents