转子轴心轨迹故障诊断特征识别方法研究

A New Coding Method of Rotor Whirl Orbit

转子轴心轨迹形状反映转子系统的工作状态，用神经网络方法可识别转子轴心轨迹形状，但神经网络的训练速度和稳定性与网络输入数据编码方式有关，提出一种使轴心轨迹图形编码得到较大压缩的平面图形可变等长度压缩编码方法，从而减少了轴心轨迹神经网络识别系统的输入变元数，使训练后的神经网络的联想能力得到较大提高，也加快了网络的训练速度及稳定性．将该图形形状识别系统加入到故障诊断专家系统中，提高了故障诊断专家系统的自动诊断水平．

The figures of rotor whirl orbit represent the working condition of rotor dynamic system. Neural networkshas been used to observe, recognize and judge the rotor system working condition and has been proved to be feasible. But the traning speed and the stability of the neural network depends on the input coding data. In this paper,a new method of compressing coding plane graphics is presented acconling to the characteristics of the rotor whirl orbit. By means of this method, the graphic code of the whirl orbit is compressed largely, so that Ihe input data of theneural networks recognizing system of rotor whirl orbits are reduced. The training speed of the network is accelerated, the stability of the network is imprnved and the autornatic level of a fault diagnsis expert system with this systemis risen.