傳統上對於可靠度的分析討論往往僅以最主要的失效模式作為唯一的失效模式進行分析與改進。然而，由於產品內部各個位置其材料性質與幾何形狀的差異；引發失效的失效模式往往不只一個，而造成失效數據的不完整。尤其在加速壽命試驗中為了迅速引發產品失效而以遠高於正常操作環境的條件進行試驗，若不對其中各失效模式各自進行評估分析，將難以取得足夠的訊息而導致壽命估計發生誤差。 本論文針對在溫度加速試驗中馬達定子絕緣之壽命估計進行研究。首先討論溫度效應下氧化反應對聚合物材料電性絕緣性質退化的影響，並探討其失效中累積損傷性質隨著溫度上升而逐漸趨於隨機性的現象。接續探討可靠度退化過程中失效其累積損傷性質與隨機性質的差異，配合蒙地卡羅模擬法探討強度與應力條件對此特徵的影響。並提出多個失效率具累積損傷性質之模型進行分析比較，選擇其中最適於應用在加速試驗分析之模型。 本研究以最大似然估計法處理恆定溫度加速試驗中馬達定子絕緣其失效數據不完整的各失效模式；並以累積損傷性質的觀點對傳統上描述壽命與溫度關係之加速模型進行修正。結果顯示本研究所提出之方法得以修正原先做法中對試驗數據估計的矛盾之處；並可根據模型參數對累積損傷性質的描述提出加速壽命試驗之最大容許試驗溫度，作為後續試驗規劃的參考。 Traditionally in reliability analysis, failure is associated with one of several failure modes is common that the specimen is exposed to. However, due to the complexity of the product geometry, failure may occur at different locations, each with its own particular features. These are the so-called the failure modes. There is a specific probability for failure at each different position. The lifetime estimation becomes obviously optimistic if only the most frequently occurred failure mode considered. Especially in Accelerated life tests(ALT), which usually carried out by stressing specimens at overstress conditions to explore the defects and estimate the mean- time-to-failure (MTTF) at normal use. The Maximum Likelihood Estimator is taken to identify the characteristics of insulation failure about the motors by considering the accelerated life testing data under censored situation. In this study, a deep understanding of the material degradation with stress has to be addressed. In the present study the motor stator failure due to the deterioration of insulation which is caused by the thermal degradation of insulation materials. Generally to the electrical insulation material, the dielectric strength is degraded with rising temperature. Failure mode becomes moderate in a sense of less capability about the accumulation of insulation damage at higher temperatures. Two different reliability models have been adopted in this study, the AE model and the Weibull distribution model. The numerical results indicate that both two models are well fitted to the data and give more information to the predicted lifetime. Moreover, in this study the maximum test temperature can be suggested according to the parameter of cumulative damage.
