散熱風扇廣泛使用於不同設備的散熱系統中，而散熱風扇的穩定度將直接影響設備 的性能。有許多國際標準組織訂定標準來評估風扇的效能，且許多風扇製造商為提升散 熱風扇的穩定度及使用壽命，提出不同方式及改善方法，例如改善風扇組裝結構，或利 用電子電路監控風扇的運轉模式。 本研究採用資料探勘技術，並多種監督式學習技術進行風扇壽命預測模型的建立， 包括類神經網路(ANN)、隨機森林(RF)、支援向量機(SVM)以及羅吉斯迴歸(LR)。此外， 由於風扇的組成結構包含許多機械元件及電子元件，該些元件均係影響風扇壽命的因子， 引此本研究納入 26 個自變項進行風扇壽命預測模型的建立，用以提升散熱風扇的壽命 預測準確率。 最後，根據本研究結果顯示針對扇熱風扇在運轉 4360hr、6360hr 及 7360hr 的運轉 後是否會發生 FAIL 之情形進行探討，結果顯示均能有 78%以上的預測準確度，另在上 述四種技術當中，就研究結果顯示較佳的評估結果為支援向量機及隨機森林，此兩個演 算法技術可以幫助風扇壽命預測上，建立較佳的預測模型。;Cooling fans are widely used in the cooling systems of different equipment, and the stability of the cooling fans will directly affect the performance of the equipment. Many international standards organizations have set standards to evaluate the performance of fans, and many fan manufacturers have proposed different methods and improvement methods to improve the stability and using life of cooling fans, such as improving the fan assembly structure, or using electronic circuits to monitor the performance of fans. This paper investigates uses supervised learning techniques, including artificial neural network (ANN), Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR) to construct a cooling fan life prediction model. Since the component of the cooling fan includes many mechanical and electronic parts, these components are affect the life of the fan. Therefore, collects 26 key parameters in this study to establish a fan life prediction model to improve the accuracy of life prediction of cooling fans. According to the results of experimental, it is studies whether FAIL will occur after the cooling fan running for 4360hr, 6360hr and 7360hr. The results can be more than 78% accuracy at each prediction model. In addition, among the above four learning technologies, the research results show that the best evaluation results are SVM and RF. These two algorithm technologies can help to establish the better prediction model for fan life prediction.
