位置-尺度族的區間設限現場失效樣本的可靠度分析

Reliability Analysis Based on Interval-Censored Field Failure Samples for Location-Scale Family

吳思樺(Sih-Hua Wu)

指導教授 ： 蔡宗儒

淡江大學/商管學院/統計學系碩士班/碩士(2018年)

https://doi.org/10.6846/TKU.2018.00044

摘要

自動化生產中不完整的現場失效資料經常出現在設備系統可靠度的評估上，產品壽命機率模型的不確定性，會影響分析結果的正確性，本論文假設系統壽命服從位置-尺度族，並以最常見的常態分配、最小極值分配和羅吉斯分配為可能的候選分配，根據不完整的區間設限現場失效資料，研究如何透過模型的互相競爭，篩選出最適合目前資料的產品壽命機率分配，配合使用牛頓法根據篩選出來的機率分配進行模型的參數估計，並以蒙地卡羅模擬以及實際案例評估所建議的統計方法的估計成效。通過模擬結果得到本論文提出的模型選擇方法有高的模型選擇正確率，參數估計結果表現穩定。最後，我們使用一個高速電機系統的現場失效資料說明我們所提出的統計方法的應用。

關鍵字

現場失效資料 ； 位置尺度族 ； 設限樣本 ； 擬牛頓估計法 ； 赤池信息量準則 ； 貝氏信息準則 ； 最大概似估計

並列摘要

Incomplete field failure data from automated production often are applied for evaluating the system reliability. The uncertainty of the product lifetime distribution could affect the evaluation on the system reliability. This study assumes that the system lifetimes follow a member in the location-scale family, and then assigns the most likely candidate from the normal distribution, smallest extreme value distribution and logistic distribution as the underlying lifetime distribution. Based on the incomplete interval-censored field failure data, we study how the candidate distributions compete with each other and the practitioners how to select the most suitable candidate distribution as the lifetime distribution of products. The parameters in the candidate models are estimated through using the Newton method. Monte Carlo simulations are conducted to evaluate the estimation performance of the proposed model. Simulation results show that our proposed method can have good estimation performance even the underlying lifetime model is uncertainty. A real interval-censored data set regarding high speed motor is used for illustrating the applications of the proposed method.

並列關鍵字

Field failure data ； Location-Scale Family ； Censored sample ； Quasi-Newton method ； Akaike information criterion(AIC) ； Bayesian Information Criteria(BIC) ； Maximum likelihood estimation

參考文獻

1. K. Suzuki (1985). Estimation of lifetime parameters from incomplete field data. Technometrics 27: 263-271.

2. L. J. Bain, M. Engelhardt, F. T. Wright (1985). Tests for an increasing trend in the intensity of a poisson process: a power study. Journal of the American Statistical Association 80 : 419-422.

3. M.-W. Lu (1998). Automotive reliability prediction based on early field failure warranty data. Quality and Reliability Engineering International 14: 103-108.

4. Y. Wang, Y. Jia, J. Yu and S. Yi (1999). Field failure database of CNC lathes. International Journal of Quality & Reliability Management 16: 330-343.

5. D. W. Coit and K. A. Dey (1999). Analysis of grouped data from field-failure reporting systems. Reliability Engineering and System Safety 65: 95-101.