旋转机械在启动过程中由于某种故障无法达到额定转速的情况下，数据采集的不完整造成故障诊断系统无法正常工作。提出用分割截取的方法自动提取旋转机械三维谱图特征形成数据网码，用于神经网络的训练和识别，以使故障诊断在非常状态下得以完成，用于２００ＭＷ汽轮发电机组证明该方法是可行的 Rotating machinery in the start-up process due to some kind of fault can not reach the rated speed of the case, the incomplete data collection caused the fault diagnosis system can not work properly. The method of segmentation and interception is proposed to automatically extract the features of rotating machinery 3D spectrum to form the data network code for training and identification of neural network so that the fault diagnosis can be completed under abnormal conditions. The method is applied to 200MW turbine generator set to prove that the method is feasible