Speaker
Description
Reliability analysis of field systems is essential for maintenance planning. However, lifetime data for field
systems are often incomplete, heavily censored, and affected by changes in data-recording practices. In
addition, failure mechanisms are frequently not documented, making it difficult to identify an appropriate
statistical model.
This work presents a Python-based Weibull Analysis Tool that automates reliability assessment and
failure forecasting for operational systems.
The tool was developed for CERN’s Front-End Computers and integrated into the organisation’s
Confluence environment, enabling regularly updated reliability trends based on field data. It supports
two- and three-parameter Weibull models, as well as mixture and competing-risks models, allowing two
failure mechanisms to be considered even when failure causes are not explicitly recorded. An
automated, cross-validation-based model selection approach identifies the most suitable model for each
data set, while confidence bounds quantify the uncertainty of reliability estimates and failure forecasts.
Validation with synthetic data confirmed that the implemented methods produce reliable confidence
bounds comparable to those of established reliability software. Application to CERN field data
demonstrated that the tool can generate actionable reliability trends and failure forecasts to support
maintenance prioritisation and replacement planning.
The developed workflow is intended not only for CERN's control hardware but also as a reusable
approach for the reliability analysis of other field systems with censored lifetime data.