基於機器學習的消費性電子產品維修預測模型研究

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/98324

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98324

Title:	基於機器學習的消費性電子產品維修預測模型研究
Authors:	林建宏;Lin, Chien-Hung
Contributors:	資訊管理學系
Keywords:	機器學習;預測模型;故障分析;Machine learning;Fault analysis;Predictive Clients
Date:	2025-07-22
Issue Date:	2025-10-17 12:37:51 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著消費性電子產品日益普及，相關維修需求顯著提升。為提升維修服務效率與顧客滿意度，本研究提出一種基於機器學習之故障維修位置預測模型的驗證方法。透過分析歷史維修資料、設備使用模式及生產工序等因素，預測電路板之實際故障位置。研究中採用多種機器學習演算法，包括隨機梯度下降分類器（Stochastic Gradient Descent Classifier）、決策樹（Decision Tree）、神經網絡（Neural Network）、自適應增強（AdaBoost）及隨機森林（Random Forest），進行資料訓練與驗證，並從資料結構面探討最佳預測資料建構方式，以避免因資料結構不當導致預測準確率下降。此外，本研究透過模擬實際運作情境，驗證訓練資料更新頻率之最佳化策略，使模型能適用於量產產品與新開發產品之不同樣本數，進而選擇最適預測模型進行演算。研究結果顯示，透過準確預測故障位置，能有效提升維修資源配置效率、降低營運成本，並增進顧客滿意度。本研究亦探討消費性電子產品生產過程中常見故障之檢修流程，包括目視檢查、故障分析、焊接修復與功能測試，並介紹常用之故障判斷方法，如比較法、觸摸法與手壓法。此等技術不僅有助於快速定位故障位置，亦可提升產品之可靠性與穩定性，對電子產品製造與維護具重要意義。未來研究可進一步結合深度學習技術與即時感測資料，提升模型的自我學習能力與預測精度，並探索跨產品類型的模型泛化能力，以建立更具彈性與可擴展性的智慧維修系統，促進電子產業的數位轉型與智慧製造發展。;With the rising popularity of consumer electronics, the demand for repair services has significantly increased. To improve repair efficiency and customer satisfaction, this study proposes a machine learning-based fault location prediction model. By analyzing historical repair data, usage patterns, and production processes, the model predicts actual fault locations on PCB. Multiple machine learning algorithms are used, including Stochastic Gradient Descent, Decision Tree, Neural Network, AdaBoost, and Random Forest, for training and validation. The study also examines optimal data structuring to prevent reduced prediction accuracy from poor data organization and emphasizes the importance of clean, well-structured data. Simulated real-world scenarios help validate strategies for updating training data, ensuring the model works for both mass-produced and newly developed products. Results show accurate fault prediction improves repair efficiency, lowers costs, and enhances customer satisfaction by providing more targeted repair solutions. The study also covers common repair processes—visual inspection, fault analysis, soldering, and functional testing—and introduces diagnosis techniques like comparison, touch, and pressure methods. These aid in quick fault detection and improve product reliability. Future work may integrate deep learning and real-time sensor data to enhance learning and generalization across product types, supporting a flexible, scalable smart repair system for digital transformation and smart manufacturing.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	31	View/Open

社群 sharing

Loading...