船舶在航行時易受到外在環境影響，如波浪拍打於船體上造成船舶振動，亦會因儀器設備的損壞導致異常振動。倘若為劇烈振動使得船舶的振動幅度過大，不僅讓乘客感到不適，同時船舶的各種儀器設備更會因振動造成損壞而增加船舶的危險性。因此，本研究試圖經由觀察設備的振動訊號，探討是否能有效監控船舶設備的健康狀態，來提升船舶安全性及避免財產損失。
本研究取得一艘船舶主機振動訊號資料，經過資料預處理及時頻分析後，接著運用統計方法、相關係數矩陣分析及共線性診斷後的結果，以萃取振動資料之訊號特徵。並應用K-Means集群分析方法來定義訊號狀態，同時建立5種訊號狀態分級，後續再將此5種訊號狀態分級，利用類神經網路技術建構訊號偵測模型來確認其一致性及有效性。
使用類神經網路訊號偵測K-Means集群模型的一致性比率達99.95%，且將新訊號資料輸入模型內驗證之一致性比率也高達97.58%。建立的模型能判斷訊號資料處於正常範圍或超出閾值，從而為船舶狀態監控提供方法，達成提早發現設備異常及維護，並提升船舶安全效能、降低意外風險。

Ships are susceptible to external environment, such as when waves hit the hull during navigation may cause vibrations. If the vibration is severe, the vibration amplitude of the ship will be too high for passengers to feel uncomfortable, and the various instruments and equipment of the ship will be damaged due to vibration, which will increase the risk. Thus, by observing the vibration signal of the equipment, the health status of the ship's equipment can be effectively monitored to improve the safety of the ship and avoid property losses.
This research obtained the vibration data of one ship’s main engine. After data preprocessing and time-frequency analysis, statistics, correlation matrix and collinearity test were employed to extract the characteristics of variables in the vibration data and K-Means clustering was applied to define and establish 5 signal states. In the follow-up, 5 signal states were used to construct a signal detection model using neural network in order to verify the consistency and validity.
The consistency ratio of the neural network signal detection model for K-Means classes is 99.95%, and the consistency ratio of inputting the new signal data into the model for verification is as high as 97.58%. The model can determine that the signal data is within the normal range or exceeds the threshold, so as to provide a method for ship condition monitoring, achieve the early detection of equipment abnormalities and maintenance, improve ship safety, and avoid accidents.

[1] 黃迎春、彭公武(2015)。《船艇振動與噪聲》。哈爾濱市：哈爾濱工程大學出版社。
[2] Niu W.Y., Zhou X.C., Li Q. (2015), Theoretical research on ship shafting torsional vibration based on power flow theory. The Third International Conference on Transportation Information and Safety, Wuhan, P. R. China.
[3] He Y.- N., He L., Lu Z.-Q., Shu L.-H. (2006), Radiation noise analysis using the results of power flow finite element method. Journal of Ship Mechanics. Volume 10, pp.150-154.
[4] Lyon, R.H., Maidanik, G. (1962), Power flow between linearly coupled oscillators. The Journal of the Acoustical Society of America. Volume 34, pp. 623-639.
[5] ITREAD01(2018)。快速傅立葉變換在訊號處理中的應用。檢自https://www.itread01.com/content/1544330530.html (March 15, 2021)
[6] Badri N. M., Rashimita K.M. (2017), FFT and spare FFT techniques and applications. The Fourteenth International Conference on Wireless and Optical Communication Networks (WOCN), Mumbai, India.
[7] Yang H., Yuan Y., Wu J., Huang Y., Yang J. (2012), NUFFT applied to motion compensation in the Near-Space SAR imaging. The 2012 International Geoscience and Remote Sensing Symposium, Munich, Germany.
[8] Nicolas B., Fabien L., Anatole L. (2011), Comparative study of band-power extraction techniques for motor imagery classification. The 2011 IEEE Symposium on Computational Intelligence, Cognitive Algorithms, Mind, and Brain (CCMB), Paris, France.
[9] Nicolas B., Fabien L., Anatole L. (2011), Comparative study of band-power extraction techniques for motor imagery classification. The 2011 IEEE Symposium on Computational Intelligence, Cognitive Algorithms, Mind, and Brain (CCMB), Paris, France.
[10] Wil van der Aalst, Ernesto D. (2015), Processes meet big data: Connecting data science with process science. IEEE Transactions on Services Computing. Volume 8, pp.810-819.
[11] Yang L.H., Deng M.L. (2010), Based on k-Means and fuzzy k-Means algorithm classification of precipitation. The 2010 International Symposium on Computational Intelligence and Design, Hangzhou, China.
[12] Li R., Xi O-J, Pang B., Shen J., Ren C-L (2009), Network application identification based on wavelet transform and k-means algorithm. The 2009 International Conference on Intelligent Computing and Intelligent Systems, Shanghai, China.
[13] Jeffrey E., Martin A., Anirban M. (2006), Traffic classification using clustering algorithms. The SIGCOMM’06 Workshops, Pisa, Italy.
[14] Ding X.Z. (2020), Research on camera calibration technology based on deep neural network in mine environment. The 2020 International Conference on Computer Vision, Image and Deep Learning (CVIDL), Chongqing, China.
[15] 張志禹、榮柱(2007)。基於非均勻FFT的距離徒動算法。計算機工程與應用，43卷17期，32-34頁。
[16] 陳鴻烈、蔡大偉(2007)。不同優養水質共線性分析及模式選擇之研究。水土保持學報，39卷3期，229-246頁。
[17] 周建張、苟榮華、蔡朝祿、俞惠麟(2019)。《船藝學》。高雄市：翠柏林企業。