Logo image
奈米溶膠保存期限之統計推論
Dissertation

奈米溶膠保存期限之統計推論

姚禹丞
Doctor of Philosophy (PHD), 國立清華大學, 統計學研究所
2015

Abstract

奈米溶膠試驗 雙常態混合分布 保存期限估計 pH值加速衰變模型 期望條件最大演算法 模型誤判分析 最佳試驗計畫 Nano-sol mixture of two normal distributions shelf-life prediction pH acceleration model expectation/conditional maximization (ECM) algorithm model mis-specification optimal test plan
Motivated by nano-sol data set, this thesis addresses the shelf-life prediction of a nano-sol product. Instead of using temperature, humidity or voltage to do accelerated life/degradation experiments in the past, this study used pH as accelerating factor to predict the shelf-life of nano-sol product. The observed distribution data of nano-particles in the nano-sol are histogram-valued frequencies. A mixture of two normal distributions was used to describe the particle size distribution. The time evolution of the particle size distribution under different pH values was described by a pH accelerated degradation model. The maximum likelihood estimator (MLE) for unknown parameters in the pH accelerated degradation model can be solved by applying Expectation/Conditional Maximization (ECM) algorithm. The estimated shelf-life under normal-use-condition and corresponding 95% confidence intervals can then be obtained. An optimal test plan for pH accelerated degradation model can be obtained by minimizing the approximate variance of the estimated shelf-life of the nano-sol product under the constraint that the total experimental cost not exceeding a pre-specified budget. The sensitivity analysis reveals that the optimal test plan is quite robust to moderate departures from the model parameters. Finally, the effects of model mis-specification were discussed. Specifically the asymptotic large sample distribution of the shelf-life estimates was derived when the particle size distribution described in the first topic is wrongly fitted with inappropriate model. The result show that the effects on the accuracy and precision of the product’s shelf-life are critical.

Metrics

1 Record Views

Details

Logo image