Human Error Data Collection and Data Generation

Abstract

Human error data in the form of human error probabilities should ideally form the corner‐stone of human reliability theory and practice. In the history of human reliability assessment, however, the collection and generation of valid and usable data have been remarkably elusive. In part the problem appears to extend from the requirement for a technique to assemble the data into meaningful assessments. There have been attempts to achieve this, THERP being one workable example of a (quasi) database which enables the data to be used meaningfully. However, in recent years more attention has been focused on the PerformanceShaping Factors (PSF) associated with human reliability. A “database for today” should therefore be developed in terms of PSF, as well as task/ behavioural descriptors, and possibly even psychological error mechanisms. However, this presumes that data on incidents and accidents are collected and categorised in terms of the PSF contributing to the incident, and such classification systems in practice are rare. The collection and generation of a small working database, based on incident records are outlined. This has been possible because the incident‐recording system at BNFL Sellafield does give information on PSF. Furthermore, the data have been integrated into the Human Reliability Management System which is a PSF‐based human reliability assessment system. Some of the data generated are presented, as well as the PSF associated with them, and an outline of the incident collection system is given. Lastly, aspects of human common mode failure or human dependent failures, particularly at the lower human error probability range, are discussed, as these are unlikely to be elicited from data collection studies, yet are important in human reliability assessment. One possible approach to the treatment of human dependent failures, the utilisation of human performance‐limiting values, is described.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 34 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.