Predicting the unpredictable: Consideration of human and organisational factors in maintenance prognostics

Human performance is a major contributor to system performance and maintenance errors can have a significant influence on system reliability. However, existing reliability modelling approaches lack any methodologies to take account of maintenance actions in predicting system failure probability. The primary objective of this work is the development of a methodological framework to enable the integration of human and organisational factors (HOFs) as quantitative metrics within prognostics maintenance models. Inclusion of significant HOF metrics derived from performance shaping factors (PSFs) should reduce the predictive uncertainty of models developed for system failure probability estimation. This research investigates human error during maintenance activities as one variable contributing to system failure events. The hypothesis is that including HOF metrics derived from the performance shaping factors (PSFs) influencing maintenance tasks can reduce the predictive uncertainty of models developed for system failure probability estimation, provided that the PSFs are found to be significant predictors of the failure event. This research uses a case study from the biopharmaceutical industry to demonstrate the industrial application of the developed methodology. Regarding the case study, current field data is unable to isolate a process variable that can reliably predict sudden component failure. Technician error during installation and system maintenance activities is therefore investigated as a potential significant variable. This applied research explores how human errors can be discovered and accounted for within the reliability modelling process. The use of PSFs in this way forms one part of the development of data driven soft-sensors using a knowledge fusion approach. This soft sensor approach utilises a combination of quantitative and qualitative information in the form of laboratory tests, historical industrial process data, and metrics derived from human factors analysis, the combination of which is unique in the literature.