Reliability prediction through Bayesian inference based on product change

This paper provides a methodology to determine the reliability of a system, when only limited information about the system (in this case a patient table of a medical scanning system) is available. With this methodology, Bayesian inference, different sources of information (expert knowledge and field data) can be combined, and the uncertainty that is inherent to reliability prediction when only little information is available can be depicted. In order to use the information that can be provided by experts, the expert knowledge first has to be quantified, which has also been done in this case study. It is found that it is possible to quantify expert knowledge about reliability performance over time of the subsystem. However, the process to do that (elicitation process) is a difficult process to control, because of the different issues that play a role. Also, in this case, the design of the subsystem under study had already been defined, which could have influenced the ease of elicitation. Furthermore, it is shown that it is possible to predict the reliability performance of the new table over time based either solely expert knowledge and one reliability indicator (without further data), or based on a combination of data and expert knowledge. Also the possibility to depict the uncertainty regarding the prediction and its reduction when more information becomes available is shown.