隨著台灣製造業的轉型，企業在設備維護上的重視程度日益增加。傳統的反應式維護已無法滿足需求，計畫性維護雖然改善部分問題，但仍存在資源浪費的風險。因此，預測性維護逐漸成為企業的重點，透過即時監控設備狀態來偵測故障的出現，得以降低成本及提高生產效率。大型企業所使用的高階設備本身多有提供監控參數以進行分析做預測，對資源有限的中小型企業而言，設備往往沒有太多的監控參數資料可供使用。而最快速便捷能取得，且與設備關聯度較高的資料就是設備的振動訊號。目前振動訊號的監控在旋轉類設備與CNC加工類設備上已有初步研究，但尚未有應用於切腳刀具的研究。故本論文研究切腳類設備的切腳刀具進行分析，期望在切腳刀具磨耗上能導入早期異常偵測，使設備與刀具壽命得以延長，並降低無預警的停機風險。本研究探討如何由振動訊號為基礎進行刀具磨耗的早期異常偵測。首先透過感測器取得振動訊號資料，在進行資料前處理與特徵提取後，進行早期異常偵測的模型訓練。其中分為先透過變化點分析 (Change Point Detection) 篩選資料，以及直接使用資料進行訓練兩種方法。實驗結果顯示，先透過變化點分析篩選資料的訓練結果較佳，並以GRU加上Attention的模型在早期異常偵測表現最佳。

With the transformation of Taiwan's manufacturing industry, enterprises have increasingly emphasized equipment maintenance. Traditional reactive maintenance can no longer meet current demands, and while preventive maintenance addresses some issues, it still poses the risk of resource waste. As a result, predictive maintenance has gradually become a key strategy for enterprises. By monitoring equipment conditions in real time to detect the onset of faults, companies can reduce costs and enhance production efficiency.While large enterprises often rely on advanced equipment with built-in sensors that provide abundant parameters for predictive maintenance, small and medium-sized enterprises usually operate older machines with limited data availability. Among the most accessible and highly correlated alternatives, vibration signals have been applied in monitoring rotating machinery and CNC equipment. However, their application to trimming tools in lead-cutting processes remains unexplored. This thesis proposes vibration signal analysis for trimming tools to enable early anomaly detection of tool wear, thereby extending equipment and tool lifespan while reducing unexpected downtime.This thesis explores how to perform early anomaly detection for tool wear based on vibration signals. Vibration data are first collected through sensors, followed by data preprocessing and feature extraction, and finally, model training for early anomaly detection. Two approaches are compared: one using data directly for training, and another incorporating change point detection (CPD) to pre-filter the data. Experimental results indicate that pre-filtering data through CPD leads to better model performance, with the GRU model combined with an attention mechanism achieving the best results in early anomaly detection.

[1] Alexandros Bousdekis, Dimitris Apostolou and Gregoris Mentzas, Predictive Maintenance in the 4th Industrial Revolution: Benefits, Business Opportunities, and Managerial Implications, IEEE Engineering Management Review, Volume 48, pp 57-62, 2020.
[2] Mounia Achouch, Mariya Dimitrova, Khaled Ziane, Sasan Sattarpanah Karganroudi, Rizck Dhouib, Hussein Ibrahim and Mehdi Adda, On Predictive Maintenance in Industry 4.0: Overview, Models, and Challenges, Applied Sciences, Volume 12, 2022.
[3] Meghdad Khazaee, Ahmad Banakar, Barat Ghobadian, Mostafa Mirsalim, Saeid Minaei, Mohamad Jafari, and Peyman Sharghi, Fault Detection of Engine Timing Belt Based on Vibration Signals Using Data-Mining Techniques and a Novel Data Fusion Procedure, Structural Health Monitoring, Volume 15, pp 583-598, 2016.
[4] Muhammad Ali Khan, Muhammad Atayyab Shahid, Syed Adil Ahmed, Sohaib Zia Khan, Kamran Ahmed Khan, Syed Asad Ali and Muhammad Tariq, Gear Misalignment Diagnosis using Statistical Features of Vibration and Airborne Sound Spectrums, Measurement, Volume 145, pp 419-435, 2019.
[5] Kilian Vos, Zhongxiao Peng, Christopher Jenkins, Muhammad Rifat Shahriar, Pietro Borghesani and Wenyi Wang, Vibration-based Anomaly Detection using LSTM/SVM Approaches, Mechanical Systems and Signal Processing, Volume 169, 2022.
[6] Paulo Antonio Delgado-Arredondo, Daniel Morinigo-Sotelo, Roque Alfredo Osornio-Rios, Juan Gabriel Avina-Cervantes, Horacio Rostro-Gonzalez and Rene de Jesus Romero-Troncoso, Methodology for Fault Detection in Induction Motors via Sound and Vibration Signals, Mechanical Systems and Signal Processing, Volume 83, pp 568-589, 2017.
[7] Charles Truong, Laurent Oudre and Nicolas Vayatis, Selective Review of Offline Change Point Detection Methods, Signal Processing, Volume 167, 2020.
[8] Corinna Cortes and Vladimir Vapnik, Support-Vector Networks, Mach Learn, Volume 20, pp 273-297, 1995.
[9] Kyunghyun Cho, Bart van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk and Yoshua Bengio, Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation, Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp 1724-1734, 2014.
[10] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár, Focal Loss for Dense Object Detection, Pattern Analysis and Machine Intelligence, Volume 42, pp. 318-327, 2018.
[11] Dzmitry Bahdanau, Kyunghyun Cho and Yoshua Bengio, Neural Machine Translation by Jointly Learning to Align and Translate, arXiv preprint arXiv:1409.0473, https://arxiv.org/abs/1409.0473, 2014.
[12] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser and Illia Polosukhin, Attention is All You Need, Advances in Neural Information Processing Systems (NeurIPS), pp 5998-6008, 2017.
[13] Laurent Oudre and Nicolas Vayatis, Ruptures: Change Point Detection in Python, arXiv:1801.00826, https://arxiv.org/abs/1801.00826, 2018.