深度學習於車輛零件及保養時間預測之應用研究;Research on the Application of Deep Learning in Vehicle Parts and Maintenance Time Prediction

NCU Institutional Repository > 管理學院 > 資訊管理學系碩士在職專班 > 博碩士論文 > Item 987654321/95494

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/95494

題名:	深度學習於車輛零件及保養時間預測之應用研究;Research on the Application of Deep Learning in Vehicle Parts and Maintenance Time Prediction
作者:	麥柔惠;Jou-Hui, Mai
貢獻者:	資訊管理學系在職專班
關鍵詞:	深度學習;車輛零件;維護時間預測;deep learning;vehicle parts;maintenance time prediction;RUL;BERT;predictive maintenance
日期:	2024-07-11
上傳時間:	2024-10-09 16:54:19 (UTC+8)
出版者:	國立中央大學
摘要:	車輛產業本身為較封閉的產業而商用車更為封閉且少為人知的產業，經銷商會遇到零件庫存的問題導致庫存成本及客戶等待抱怨等問題，這些問題可以延伸到庫存管理、零件成本、客戶營業損失賠償等。在過往歷史上受限於車輛產業的資料封閉大多研究停留在車廠內部，導致相關研究斷斷續續，研究無法延伸也無法使用舊有資料重現比較，經銷商多為傳統產業較少有應用機器學習優化服務流程的能力，現有相關研究大多注重在模型而不是實際應用，本研究使用時下的機器學習技術搭配某S公司實際車輛維修履歷，預測客戶”下次”回廠的時間及”下次”維修的零件。本研究使用注意力機制Transformer輸入車輛的車齡、里程數、車型、保固情況，預測零件更換(多類別標籤的分類任務)及回廠時間(迴歸任務)，並比較了其他時間序列的模型效能，實驗各特徵對準確率的重要性。本研究成果展示了深度學習技術在車輛售後服務領域應用的可行性，期望提高汽車行業的效率、降低成本並提升客戶滿意度。 ;The vehicle industry is relatively closed, and the commercial vehicle industry is even more closed and less well-known. Dealers often face problems with parts inventory, leading to inventory costs and customer waiting complaints. These issues can extend to inventory management, parts costs, and compensation for customer business losses. In the past, due to the closed nature of data in the vehicle industry, most research was limited to internal studies within vehicle manufacturers, resulting in intermittent research that could not be extended or replicated using existing data. Dealers, often traditional industries, have limited ability to apply machine learning to optimize service processes. Existing related research mostly focuses on models rather than practical applications. This study uses current machine learning techniques combined with actual vehicle maintenance records from Company S to predict the time of a customer′s "next" visit to the service center and the parts to be repaired during the "next" visit. This study employs the attention mechanism Transformer, inputting vehicle age, mileage, model, and warranty status to predict parts replacement (a multi-class classification task) and the time of the next service visit (a regression task). The performance of other time series models is compared, and the importance of each feature for accuracy is experimented. The results of this study demonstrate the feasibility of applying deep learning techniques in the field of vehicle after-sales service, with the expectation of improving efficiency, reducing costs, and enhancing customer satisfaction in the automotive industry.
顯示於類別:	[資訊管理學系碩士在職專班 ] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	57	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....