工學院: 工業工程學研究所指導教授: 吳文方賴鴻毅Lai, Hung-YiHung-YiLai2017-03-022018-06-292017-03-022018-06-292015http://ntur.lib.ntu.edu.tw//handle/246246/271535本研究根據不同溫度條件下馬達壽命加速測試結果探討馬達的可靠度，而在傳統的馬達加速測試中，溫度與對數化後的馬達壽命之關係往往被視為是線性，馬達壽命的標準差也常被假設為固定，然而在高度重視可靠度的產業界，由於對於馬達加速壽命測試的預估誤差之要求標準越來越高，因此本研究採用以溫度為自變量的非線性迴歸，套用至馬達對數壽命平均值與標準差為應變數的預估迴歸模型之中，分析結果顯示，平均值預估迴歸模型中，二次迴歸模型的預估誤差較線性迴歸模型低；而標準差預估迴歸模型中，指數迴歸模型擬合能力最佳。本研究接著以預估結果最佳的馬達對數壽命之平均值與標準差預估迴歸模型，探討馬達壽命數據在業界絕熱等級中的可靠度評估。本研究亦捨棄傳統加速測試規劃中，在各溫度水準下配置相同樣本數的方式，為了增加預估迴歸模型的準確度，本研究應用費雪資訊矩陣來降低線性迴歸模型中截距與斜率參數的變異程度，藉此推導出如何將有限樣本有效率地分配於各溫度水準中，經分析結果表示，實驗者應該將所有樣本分配於加速測試中最高溫度與最低溫度兩種環境等級之中，且若欲估算的環境溫度越低時，所配置於最高溫度環境的樣本數也應隨之增加，反之亦然。To understand the life distribution and reliability of electrical motors in short time, accelerated life testing (ALT) has been applied widely in industry. In traditional accelerated tests, linear regression tends to be an appropriate method to analyze the relationship between the logarithm life and surrounding temperature of a motor, and the standard deviation of the logarithm life is regarded as a constant in accordance with linear regression and for the sake of convenience. However, due to requirements for decreasing life prediction errors of motor after ALT, one of the research objects of this thesis is to estimate the mean and standard deviation of life of a motor based on nonlinear regression. The result indicates that a quadratic regression model is the most appropriate one than others to estimate the mean of the logarithm life, while an exponential regression model performs optimally to fit the standard deviation of the logarithm life of a certain kind of motor. In order to conduct an accelerated life test accurately and efficiently with constrained resources in practice, sample allocation proportions under different temperatures should be determined carefully at the design stage. These decision variables primarily affect the accuracy of life prediction of motors. To improve the accuracy, this thesis proposes and illustrates a method using real test data of motor to optimize an objective function under a few constrained resources based on Fisher information matrix. The result indicates that there is no need to assign test samples at temperature levels other than the lowest and highest ones. In fact, experimenters should allocate more samples at the highest temperature level.1141350 bytesapplication/pdf論文公開時間: 2020/7/23論文使用權限: 同意有償授權(權利金給回饋學校)可靠度加速測試非線性迴歸非齊一標準差樣本配置reliabilityaccelerated life testingnonlinear regressionnonconstant standard deviationsample allocationthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/271535/1/ntu-104-R01546020-1.pdf