應用資料探勘技術於汽車維修業之研究

The Study of Applying Data Mining Techniques in Automotive Maintenance Business

魏崢(Cheng Wei)

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(2010年)

摘要

台灣汽車零件業具有少量多樣、彈性製造之優勢，在業者不斷投入研發及提升生產技術後，已具國際競爭能力。在受到國內車業黯淡和油價居高的影響，造成汽車業產值在近幾年不斷下滑，自從2006年汽零業產值首度超前後，就一直居於領先的地位，可見在汽車零件這塊市場已漸漸凌駕於汽車業之上，而與汽車零件密不可分的汽車維修業更是積極的提升服務品質以獲得顧客的認同。

本研究之目的有以下幾點：1.找出逾期(一年以上)未回廠維修的顧客；2.找出影響維修項目較為顯著的因子；3.應用資料探勘中的倒傳遞類神經網路來預測可能需要維修之項目；4.應用自我組支應射網路去找出各大維修項目的高風險時期，依據研究結果，提出建議與策略，以供汽車維修業者及維修客戶未來參考運用。

從研究結果可得知：車齡對維修項目的影響最大，其次則是里程和廠牌； 應用倒傳遞類神經網路來進行分類與測試，檢測結果顯示：在A廠牌方面，最高的是煞車總泵的測試樣本，有高達98.33％的準確率；其他廠牌方面，最高的是消音器的測試樣本，有高達96.31％。從自我組織映射網路所分析出的結果可看出各大維修項目的常態維修時期。

關鍵字

汽車維修 ； 倒傳遞類神經網路 ； 自我組織映射網路 ； 資料探勘

並列摘要

In recent years, due to the saturated domestic car market and the not fully recovered economy plus high oil price, sales of vehicles are not so good. On the other hand, the automobile repair market is gradually drawing attention of the public. At the same time, the car-repairing market is also facing more challenges and competitions. However, compared with the international car-repairing market, there is still plenty of room for improvement on the qualities of testing and repairing, qualifications of the maintenance workers, and new ideas of marketing. In order to break through these limits and bottlenecks and create a new stage, deep and further research on the current situations is inevitable.

The aim of this study is to explore how data mining technology can be used in the automotive repair industry. The database used by this study is from a car repair workshop. By using data mining technology, the study performs the following analyses:1. Find those clients who didn’t return for maintenance in one year. 2. Apply neural network to analyze maintenance items and find out those more significant factors which affect them. 3. Use backpropagation neural network to analyze the relationship between the failure of major parts maintainance and car age, mileage and brand in order to build forecast model. 4. Apply self-organizing map to find out the major high-risk period of each maintenance item.

The results of this study show the following: 1. Using “Select” function of the data mining software can sort out the clients who didn’t return for maintenance in one year. 2. The most significant factor is car age, followed by mileage and brand. 3. The predicton results of backpropagation neural network show that in the A brand, the highest is the test samples of the brake master cylinders, with accuracy up to 98.33%, while the lowest is the test samples of the spark plugs, with accuracy of 78.24%; in other brands, the highest is silencer test samples with accuracy up to 96.31% , while the lowest is the transmission shaft test samples, with accuracy of 70.2%. 4. Using the self-organizing map can find out major maintenance items’ normal maintenance periods.

並列關鍵字

vehicle maintenance ； back-propagation neural network ； self organizing

參考文獻

24. 陳世杰（2005），中央空調直接負載控制績效分類與評估系統，中原大學，電機工程學系，博士論文。

19. 曹毓齡（1998），應用資料探勘於顧客分群之研究－以製藥業為例，大同大學，資訊經營學所，碩士論文。

12. 李蔚局（2008），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。

13. 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。

20. 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。