ENBIS-18 in Nancy
2 – 25 September 2018; Ecoles des Mines, Nancy (France)
Abstract submission: 20 December 2017 – 4 June 2018
Uncertainty Calculations in a System Reliabilty Tool
4 September 2018, 12:00 – 12:20
- Submitted by
- Jan-Willem Bikker
- Jan-Willem Bikker (CQM BV)
- The Dutch consultancy firm CQM (Consultants in Quantitative Methods, ~35 consultants) supports many customer’s projects in R&D (biostatistics, marketing, product development, reliability). One customer is a large electronics manufacturer with whom we co-developed a software tool for reliability prediction of their product portfolio. The tool is widely used within the organisation, with an eye towards warranty and services. The systems are composed of various sorts and types of components from a predefined library. Lifetime of and degradation in the light-source are described using typical survival models with temperature and current as typical regressors. The output consists of various graphs and tables describing survival rates over time according to several criteria and failure mechanisms, as well as breakdown of failure over the criteria or over subcomponents.
The system’s build-up is much more complex than just serial or parallel, and uses Monte Carlo simulation for the calculations. For instance, a system is considered still functional if only a few light-sources are down. A novel feature translates the uncertainty of each subcomponent’s reliability to the output on system level, which helps in the interpretation. The knowledge of a subcomponent’s failure may come from stress tests, leading to a covariance matrix of the model parameters’ estimates, or sometimes from practical educated guesses, in which case the uncertainty takes on the character of a Bayesian prior. As system could have several dozens of different subcomponents, there is a computational challenge to deliver fast system-level uncertainty intervals for the various failure curves. The talk explains how DoE schemes with up to hundreds of factors are used, and highlights some of the development aspects where computer science, mathematics, and statistics meet.
Return to programme