Gradually induced faults may occur while machinery is running for a long time. This kind of fault may cause serious damage if the potential faults aren't remedied in time. To avoid a serious breakdown, prognostic maintenance is commonly adopted to treat this kind of problem. In this paper, the general types of faults of mechanical systems are first introduced. The predicted model of performance degradation is proposed based on the trend of the over-all values of vibration. The model was constructed using a regression analysis method in cooperation with measured vibration signals. The techniques of vibration-based diagnosing are also reported for identifying the fault sources when machinery is at the degraded condition. Fault identification is done by comparing time-domain signal characteristics with the set of fault patterns. A rotational test device is designed to implement signal measuring, performance degradation prediction and fault diagnosis. The experimental results show that the exponential model and the measured data have a high correlation, indicating that they can be used to predict the performance of a machine. Moreover, the studied results can be used as the basis for developing prognostic maintenance for a system.