2022 Fiscal Year Final Research Report
Non-parametric Bayesian approach to modelling system reliability
Project/Area Number |
18K04621
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 25010:Social systems engineering-related
|
Research Institution | Waseda University |
Principal Investigator |
Hayakawa Yu 早稲田大学, 国際学術院, 教授 (80398916)
|
Project Period (FY) |
2018-04-01 – 2023-03-31
|
Keywords | System reliability / Bayesian non-parametrics / Gamma process / Bathtub hazard rate / Warranty analysis / Geometric-like process / Alpha-Series process |
Outline of Final Research Achievements |
Richard Arnold, Stefanka Chukova and Yu Hayakawa have carried out a work on modelling system from a Bayesian non-parametric perspective. Hazard rate functions of natural and manufactured systems often show a bathtub shaped hazard rate. Parametric modelling of such hazard rate functions can lead to unnecessarily restrictive assumptions on the function shape. We have extended Lo and Weng (1989) approach and specified four non-parametric bathtub hazard rate functions drawn from Gamma Process Priors. We use a gamma-scaled Dirichlet Process prior to construct the Gamma Process Prior, and have implemented simulation and inference for these four models. We and Sarah Marshall also worked on other projects on geometric-like processes, the alternating alpha-series process, nonzero repair times dependent on the failure hazard, mean and variance of an alternating geometric process, and delayed reporting of faults in warranty claims. These projects are complementary to our original goals.
|
Free Research Field |
Reliability Theory
|
Academic Significance and Societal Importance of the Research Achievements |
Based on work done by Lo and Weng (1989), we use Gamma process prior to specify non-parametric hazard rate functions. We have implemented simulation and inference for four bathtub hazard rate models. Our use of the translated Gamma Process prior for the log-convex hazard rate model is novel.
|