Reliability Analysis Using Warranty Data Consisting Only of Failure Information
| Autor: | Md, Mesbahul ALAM |
|---|---|
| Jazyk: | japonština |
| Popis: | Reliability is quality over time. Knowledge about product lifetime obtained fromactual usage (field) data is of great importance to manufacturers to get better informationof the true reliability of their products. Warranty database is a preferredsource of such knowledge which is automatically generated and updated at no additionalcost from repair claims during warranty coverage. For engineering purposes,usage time (e.g., mileage or copy number) is more relevant and lifetime parametersmeasured in usage time is an integral part of reliability analysis using warrantydata. The reliability analysis becomes difficult because warranty data usually consistof only failure-related information, and non-failure information is not included.In literature it is found that effective usage-based estimation requires supplementaryinformation about usage accumulation of non-failure units such as, follow-upstudies, or a usage time distribution which includes both failure and non-failureproducts. However, sometimes they are expensive to obtain and even impossible insome circumstances. Thus, the unavailability of the usage time of censored unitsmakes it difficult to estimate the usage-based lifetime distribution of products. Thisthesis deals with this problem, and discusses the usage-based estimation methodsfor product lifetime distributions from warranty claims data and sales data. Assumingthat failures depend on cumulative usage, two models are considered in thisthesis. The first model consists of one lifetime variable with two cases: (a) an exponentialfailure random variable, which corresponds to random failure mechanism,and (b) a Weibull failure random variable, which corresponds to wear-out failuremechanism. The second model consists of two lifetime variables, which is formulatedas a competing risks model of random failure and wear-out failure. To beginwith, maximum likelihood estimation is discussed for the first model. Then thisis extended to the second model. Since the straightforward maximum likelihoodinference requires large amount of calculation, an alternative hybrid method of estimationfor the second model is developed, called semi-parametric method, which combines maximum likelihood estimation for one failure random variable case witha nonparametric maximum likelihood method. In taking these approaches usageaccumulation or usage rate is treated as a censoring variable that is determined independentlyof failure processes. Throughout this thesis, Gauss-Hermite quadraturemethod is used to evaluate the unobserved part included in the likelihood function.To investigate the properties of the proposed methods, simulation studies are performed,and applications involving actual automobile warranty data are illustrated.This research shows that reasonably good estimates can be obtained by using onlyusage-at-failure data from the warranty database. The proposed methods are applicablefor more than two failure modes. The practical consequence of this thesis isthat supplemental information is not needed to obtain fairly good estimates of thelifetime parameters. 2009 |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|