本研究目的為發展進給系統健康診斷系統。在進給系統中最容易發生故障的部分為滾珠螺桿之預壓力消失，其次為滾動軸承，而軸承故障主要原因為裝配不當及元件損毀，因此本研究主要針對滾動軸承之組裝與元件進行健康狀態分析，共分為三個部分。第一部分發展頻帶能量特徵方法，利用軸承振動訊號，經解調後取得其包絡譜，再引用頻帶切割與掃頻的概念提取有效的頻域特徵，結合取樣熵(Sample Entropy, SE)、方均根(Root mean square, RMS)及奇異譜熵(Singular Spectrum Entropy, SSE)等時域特徵對軸承進行健康診斷。第二部分使用全息譜觀察滾動軸承運轉時軌跡之變化，並提取能代表軸承旋轉情形之振動特徵，進而針對組裝之機械故障進行狀態監測，避免因裝配問題造成軸承使用壽命減短，可及時調整組裝方式並解決問題。第三部分將軸承振動訊號提取之特徵以自組織映射圖(Self-Organizing map, SOM)建立健康模型，再透過計算待測資料與健康模型之最小量化誤差(Minimum quantization error, MQE)來量化軸承及組裝狀態，並結合移動窗與馬式距離(Mahalanobis distance)的概念對軸承進行長期監控，觀察軸承之全生命週期狀態與預防突發之異常狀況。最後，透過SOM結合時、頻域特徵與全息譜同步對軸承組裝及滾動軸承狀態進行健康診斷，並引用2012PHM集NASA軸承疲勞實驗數據集進行驗證，由實驗結果證明，此進給系統健康診斷技術能有效提升狀態估測準確度。 The purpose of this research is to develop a diagnosis system for ball screw feed drive systems. The most prone to failure component in the feed drive system is the preload loss, followed by the rolling bearings. Furthermore, the main cause of the bearing failure is the installation error and/or component damage. The thesis consists of three parts. In the first part, several features extracted from the vibration measurements in frequency domain and time domain such as Sample entropy (SE), Root mean square (RMS), Singular spectrum entropy (SSE) and band energy are compared for their effectiveness in diagnosing the ball bearing faults. In the second part, features which could represent the assembly condition of the ball screw feed drive systems such as misalignment using the holographic spectrum are proposed and their effectiveness are also investigated. Finally, all the features aiming to detecting faults of ball screw feed drive systems due to bearing and/or assembly extracted from the vibration measurements are merged using Self-organizing map (SOM) . Whether the ball screw feed drive systems is healthy can be determined by calculating the Mahalanobis distance according to the Minimum quantization error (MQE) from the SOM. The proposed diagnosis technique is validated using the data sets from PHM and NASA. Experimental results show that the ball screw feed drive system can be diagnosed with a reasonable accuracy by this system.