| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 李睿恩 | zh_TW |
| DC.creator | Juei-En Lee | en_US |
| dc.date.accessioned | 2018-6-8T07:39:07Z | |
| dc.date.available | 2018-6-8T07:39:07Z | |
| dc.date.issued | 2018 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=105522050 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 現今全球製造業致力於將工廠藉由工業物聯網、大數據分析、虛實融合系統(Cyber Physical System, CPS)等技術用以實現工業4.0智慧工廠(Smart Factory)，而預診斷及健康管理(Prognostic and Health Management, PHM)是智慧工廠中一項重要的核心系統，透過大數據的蒐集與分析，此系統能讓我們快速的掌握機械的運作情形，提早做出因應的措施。本篇論文著重於發展預診斷及健康管理中的機器剩餘可用壽命(Remaining useful life, RUL)預測技術，利用深度神經網路(Deep Neural Network, DNN)模型預估機械元件的剩餘可用壽命，可避免因元件突然損壞使機器瞬間停止運作而造成重大損失。
本論文提出了時間序列多通道卷積神經網路(Time Series Multi-Channel Convolutional Neural Network, TSMC-CNN)架構對機械設備進行剩餘可用壽命之評估，TSMC-CNN與傳統CNN不同之處在於，傳統CNN主要應用於圖片辨識或影像處理上，而TSMC-CNN將時序性的資料透過多重折疊的疊加處理，讓神經網路能夠提取出長時間序列資料變化的有效特徵，準確的預估機械設備剩餘可用壽命。
本論文以法國研究機構FEMTO-ST在PRONOSTIA實驗平台蒐集的軸承運行資料來驗證我們所提出的TSMC-CNN預測軸承剩餘可用壽命的精確度(accuracy)，且和文獻中所提出的DNN、GBDT、SVM、BP、Gaussian regression、Bayesian Ridge方法做比較，實驗結果顯示，我們提出之TSMC-CNN架構無論是均方根差(RMSE)，或是平均絕對誤差(MAE)結果都是最佳的。 | zh_TW |
| dc.description.abstract | Today′s global manufacturing industry is committed to transforming traditional factories into industrial 4.0 smart factories through technologies such as Industrial Internet of Things (IIoT), big data analysis, and Cyber Physical System (CPS). The Prognostics and Health Management (PHM) system is one of important systems of the smart factory. Through the collection and analysis of big data, the system can allow users to monitor machinery operation states and health condition in a timely manner so that proper countermeasures can be taken as soon as possible to mitigate potential problems. This study focuses on developing the Remaining Useful Life (RUL) estimation method for the smart factory PHM system. The method can be used to avoid sudden component/machine failures, which may lead to a huge loss.
In this study, we propose a deep learning method using the Time Series Multi-Channel Convolutional Neural Network (TSMC-CNN) architecture for the RUL estimation. Unlike the traditional CNN architecture that is mainly used for image recognition or image processing, the TSMC-CNN architecture divides time-series data into multiple folds and superimpose them altogether to extract relationship between data pieces that are far apart for accurately predicting the RUL of machine/component. The bearing operation data collected by the French research institute FEMTO-ST on the PRONOSTIA experimental platform is used to evaluate the accuracy of the proposed method. The evaluation results are compared with those of the DNN, GBDT, SVM, BP, Gaussian regression, and Bayesian Ridge methods proposed in the literature. The comparisons show that the proposed method is the best in the aspects of both the root mean squared error (RMSE) and the mean absolute error (MAE). | en_US |
| DC.subject | 工業4.0 | zh_TW |
| DC.subject | 智慧工廠 | zh_TW |
| DC.subject | 虛實融合系統 | zh_TW |
| DC.subject | 預診斷及健康管理 | zh_TW |
| DC.subject | 剩餘可用壽命 | zh_TW |
| DC.subject | 卷積神經網路 | zh_TW |
| DC.subject | 深度學習 | zh_TW |
| DC.subject | 時間序列 | zh_TW |
| DC.subject | Industry 4.0 | en_US |
| DC.subject | smart factory | en_US |
| DC.subject | Cyber Physical System | en_US |
| DC.subject | prognostics and health management | en_US |
| DC.subject | remaining useful life | en_US |
| DC.subject | convolutional neural network | en_US |
| DC.subject | deep learning | en_US |
| DC.subject | time series | en_US |
| DC.title | 時間序列多通道卷積神經網路用於軸承剩餘可用壽命預估 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Time Series Multi-Channel Convolutional Neural Network for Bearing Remaining Useful Life Estimation | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |