At the limit? Using operational data to estimate train driver human reliability

Human reliability analysis plays an important role in the safety assessment and management of rail operations. This paper discusses how the increasing availability of operational data can be used to develop an understanding of train driver reliability. The paper derives human reliability data for two driving tasks, stopping at red signals and controlling speed on approach to buffer stops. In the first of these cases, a tool has been developed that can estimate the number of times a signal is approached at red by trains on the Great Britain (GB) rail network. The tool has been developed using big data techniques and ideas, recording and analysing millions of pieces of data from live operational feeds to update and summarise statistics from thousands of signal locations in GB on a daily basis. The resulting driver reliability data are compared to similar analyses of other train driving tasks. This shows human reliability approaching the currently accepted limits of human performance. It also shows higher error rates amongst freight train drivers than passenger train drivers for these tasks. The paper highlights the importance of understanding the task specific performance limits if further improvements in human reliability are sought. It also provides a practical example of how big data could play an increasingly important role in system error management, whether from the perspective of understanding normal performance and the limits of performance for specific tasks or as the basis for dynamic safety indicators which, if not leading, could at least become closer to real time.