Don't be misled by what you see: Reliability estimation and maintenance optimization from observational data using causal inference

#reliability-optimization #estimation #maintenanace #observational-data #causal-inference

 In reliability engineering, often we need to rely on observational data - the data directly collected from the operational filed - to establish reliability models and make decisions related to maintenance, As compared to data collected from controlled tests like lifetime or accelerated lifetime testing, observational data are prone to various types of bias, e.g., selection bias, confounding bias. In this talk, we first show through a few examples why traditional statistical estimation methods might lead to misleading decisions when applied to observational data, and analyze why the traditional approaches. Then, we develop a new causal inference based method for reliability modeling from observational data. An Accelerated Lifetime Model with covariates are used as the reliability model, and a parameter estimation method based on propensity score matching is developed to accurately estimate the model parameters under severe selection bias. Finally, we develop a joint maintenance optimization model that integrates age replacement with periodical service that recover the degradation levels, The developed model is applied in a case study of engineering maintenance planning of critical components in electrical transmission networks from our industrial partner RTE.

• Add Event to Calendar
  iCal
  Google Calendar

• No. 37 Xueyuan Road, Beihang University
• Haidian District
• Beijing, Beijing
• China
• Building: Weimin Building | 为民楼
• Room Number: 320

---

• Contact Event Host
• Co-sponsored by School of Reliability and Systems Engineering. Beihang University | 北航可靠性与系统工程学院