Patterns of Causation in Accidents and Other Systems Engineering Failures

摘要

Project failures occur despite the industry's best systems engineering efforts. Detailed information on project failures is difficult to find, which makes it difficult to study or understand their causes and hence to identify ways to prevent future failures. In contrast, many accidents have extensive investigations with public reports. To determine whether lessons from accidents can also be used to prevent other types of failures, we compared the causes of accidents and project failures. Our sample consisted of 30 accidents and 32 non-accidents, ranging across various industries. We extracted causes from the reports on accidents and project failures, and analyzed these causes by framing them in a common "actor-action-object" structure. We discarded the details particular to each failure (e.g., the particular component that was affected) to allow us to focus on the underlying problem (e.g., components were not properly maintained). Next, we extracted the particular actions involved in each cause (e.g., failed to maintain). Ten, or just under half of the 22 problematic actions we identified were equally likely to appear in accidents or project failures (95% confidence interval). Of the remaining twelve actions, the majority can be reasonably attributed at least in part to differences in investigation style between accidents and non-accidents. Some differences occur because of when accidents and non-accidents tend to happen. Poor maintenance was rarely an issue in non-accidents, because most project failures occur long before maintenance becomes a concern. Finally, actions involving regulations, procedures, or training were cited more often in accidents, reflecting the greater number and extent of regulations, procedures, or training related to safety.