本文主要探討在齒輪發生損壞與無損壞的狀況下，齒輪組之動態響應，並以快速傅立葉分析與倒頻譜分析找出無損壞齒輪組的訊號與損壞齒輪組訊號之不同。研究方法為利用有限元素軟體ANSYS建立無損壞齒輪模型，並分析齒輪之嚙合剛度，再將嚙合剛度代入以集中質量法建立的齒輪動態系統模型。以Matlab程式計算齒輪動態系統模型，動態系統模型尚且包括了齒輪發生偏擺與沒有發生偏擺的狀況，之後以訊號處理方法對時域訊號進行分析。最後以實驗驗證模擬之訊號特性，並探討模擬訊號與實驗訊號之異同。;The goal of this research was to diagnose the gear faults through dynamic signals. A finite element software, ANSYS, was utilized to calculate the gear mesh stiffness of normal and damaged gears. A finite element software, ANSYS, was used to calculate the gear mesh stiffness of normal and damaged gears. The dynamic, lumped-parameter model of gears was developed and solved. The model also considered not only the normal and damaged gears but also the unbalanced gears. Signal processing methods, fast Fourier analysis and cepstrum analysis, were then introduced to determine the difference between dynamic signals from the normal gear mesh and from the faulty gear mesh. An experiment was also conducted to obtain data for comparison. This study concludes that the damage of a gear can result in small components in the frequency domain with the interval equal to the rotating speed of the particular gear. This characteristic can further be clearly identified in the cepstrum domain.