為了在有限的時間內推估高功率LED 晶粒的可靠度, 本論文使用雙應力變數對高功率LED 晶粒進行恆定應力加速退化測試實驗。假設加速退化實驗所收集的元件之累積損害退化資料服從Wiener 隨機過程, 我們建議以逆高斯分配建立了加速退化模型, 進而推導出產品壽命百分位數的點估計值及信賴下界。本文並以高功率LED 晶粒退化資料實例來說明本方法之應用。 It is difficult to evaluate the lifetimes of high reliable products in limited experimental time. In this thesis, statistical methods based on a two-variable constant-stress loading accelerated degradation test are developed to overcome this difficulty. Assuming the cumulative damage information of test units due to degradation has a Wiener process, the inverse Gaussian distribution is used to derive the lifetime percentiles and their low confidence bounds. A real example of LED chips data set is used to demonstrate the application of the proposed method.