本研究旨在應用類神經網路估測感應電動機之溫升，透過關聯度分析，評估影響感應電動機溫升較大之因子作為輸入變數，來估測感應電動機之溫升。在實務上，溫升試驗要全面測試是有其困難的，通常需歷時1~3小時。本研究藉由一組已知的無載電流、無載功率、無載功率因數、室內溫度、溫升溫度等資料，經由類神經網路中的倒傳遞網路學習與訓練，再由另外一組無載特性所獲取的無載電流、無載功率、無載功率因數、室內溫度作為輸入變數用來估測溫升，藉以縮短溫升測試時間，作為實際溫升量測時之輔助參考，供研發及製造過程中全面及早發現溫升異常現象，期能提高產品品質，節省製造成本。

This study investigates the temperature rise estimation of induction motor by back-propagation neural network (BPN) approach. The correlation analysis method is utilized to decide the input variables from the more influential factors on the temperature rise of induction motor. In practice, it is difficult to complete the temperature rise test because the test will spend 1-3 hours normally. A set of no-load data, such as no-load current, no-load power, no-load power factor, indoor temperature and temperature rise, are used instead to train the BPN. After training the BPN with the preceding data set, another set of no-load data, including no-load current, no-load power, no-load power factor and indoor temperature, are fed into the system to get the estimation of temperature rise. In this way, the measuring time of temperature rise can be reduced which may provide early discovery of the abnormal temperature phenomenon in the processes of manufacturing. Hence, a better quality of products and less manufacturing cost may be obtained.