應用資料探勘於預測原廠汽車零件壽命之研究

The Application of Data Mining in Predicting the Life-Span of Original Vehicle Parts

黃珮婷(Pei-Ting Huang)

指導教授 ： 黃信豪

國立虎尾科技大學/管理學院/工業工程與管理研究所/碩士(2013年)

摘要

根據美商全球調查公司J.D.POWER亞太區報導可知，最近幾年來台灣汽車經銷商售後服務顧客滿意度年年降低。至2012年更是創下最近三年來新低，從評比的項目中可知顧客越來越重視售後服務。而在原廠汽車維修廠與私人汽車維修廠間競爭日趨激烈，原廠汽車維修廠想在著手於技術研究上，以贏得顧客專業層面的信賴度。本文應用資料探勘軟體Poly Analyst6.0於台灣原廠汽車維修廠所提供的顧客進廠維修資料。使用單一零件，如：正時皮帶、剎車真空管、電瓶、水泵浦、發電機皮帶等零件，與廠牌、里程數、車齡先進行決策樹分析，再使用個別零件進行類神經網路模型預測，期望在三項因子之間找出汽車零件使用壽命是否存在著可預測模型。以提供給原廠維修廠於顧客關係管理上，結合實質的數據分析與專業的經驗，創造原廠汽車維修廠與車主之間雙贏的局面。
從研究結果可得知：1. 在此筆資料庫中，SA廠牌的汽車裡，「時規惰輪」需在里程數大於91,535公里或者車齡大於10.5年時，才需進行更換。2.應用類神經網路來預測個別零件，在不同的目標屬性與預測因子下是否存在可預測的模型，在資料庫筆數不足的情況下，類神經網路是無法進行預測的。3.應用類神經網路來預測個別零件時，預測因子不足下，分析後數據結果恐無法套用至其他資料庫之相同零件。

關鍵字

資料探勘 ； 汽車維修 ； 類神經網路 ； 決策樹分析 ； 汽車零件壽命

並列摘要

According to the Asia-Pacific Report by J.D.POWER, a US-based global market research company, Taiwan's car dealers have shown a yearly decline in customer satisfaction with after-sale services in recent years and customer satisfaction reached three-year new low 2012. Rating items reveal that customers are paying increasing attention to after-sale services. As competition intensifies between car manufacturer and private auto repair shops, the former should focus on R&D to win customer confidence in their expertise. In this study, the data mining software Poly Analyst 6.0 was used to analyze customer vehicle service data from Taiwan's car manufacturer repair shops. Single parts, such as timing belts, brake vacuum tubes, batteries, water pumps and alternator belts, as well as vehicle brands, mileages and age were first analyzed using decision trees. Then, predictions were made on individual parts using artificial neural networks. The goal was to determine whether a model that can predict the life-span of vehicle parts exists among the three factors, and provide findings that help manufacturer repair shops improve their customer relationship management by incorporating actual statistical analysis and professional experience, thus achieve the win-win outcome for themselves and car owners.
　　Research findings show the following results. 1. In this database, a SA vehicle requires the replacement of its timing idler only when its mileage or age exceeds 91,535 km or 10.5 years. 2. Regarding prediction on individual parts using artificial neural networks to determine whether a predictive model exists among different target attributes and predictive factors, data shortage in the database made it impossible for the artificial neural networks to make predictions. 3. While making predictions on individual parts using artificial neural networks, analytical data may not be applicable to similar parts in other databases due to insufficient predicative factors.
Keywords: Data Mining;Vehicle Maintenance;Neural Networks;Decision Trees;Life-Span of Vehicle Parts.

並列關鍵字

Data Mining ； Vehicle Maintenance ； Neural Networks ； Decision Trees ； Life-Span of Vehicle Parts

參考文獻

40. 黃家祚(2009)，多目標資料採礦軟體中文使用手冊，皮托科技股份有限公司。

12. 李蔚局（2007），台灣原廠汽車維修產業研究與事業經營策略分析，國立中央大學，管理學院高階主管企管碩士班，碩士論文。

58. 魏崢(2010) ，應用資料探勘技術於汽車維修業之研究，國立虎尾科技大學，工業與工程管理研究所，碩士論文。

19. 易圖清（2004），汽車維修制度與顧客滿意度之研究，國立成功大學，管理學院高階管理碩士在職專班(EMBA)，碩士論文。

33. 許韋皓（2008），資料探勘技術於不孕症問題之分析與應用，暨南國際大學，管理學院經營管理碩士學位學程碩士在職專班，碩士論文。