转子轴心轨迹作为转子故障的典型特征之一，可以提供更具代表性的故障特征信息。对转子轴心轨迹形状进行准确识别是构建转子故障特征征兆的基础。为提高转子轴心轨迹形状识别的泛化能力，提出一种基于深度卷积神经网络的转子轴心轨迹成像及形状识别方法(DimShapeNet)。将转子轴心轨迹映射到二维数字图像中，利用反灰度化预处理方法，去除二维数字图像中多余的颜色信息；将预处理后的转子轴心轨迹数字图像输入深度卷积神经网络中进行训练。结果表明：经过反灰度化预处理的转子轴心轨迹数字图像在深度卷积神经网络的训练中更有优势；相比于传统的转子轴心轨迹形状识别方法，基于深度卷积神经网络的转子轴心轨迹形状识别方法具有更高的准确性和更好的鲁棒性。

As one of the typical characteristics of rotor faults, the rotor axis trajectory can provide more representative information of rotor faults. Accurate identification of the shape of the rotor axis trajectory is the basis of constructing the characteristic symptoms of rotor faults. In order to improve the generalization ability of the shape identification of rotor axis trajectory, this paper proposes a rotor axis trajectory imaging and its shape identification algorithm based on deep convolutional neural network called DimShapeNet. The analytic expression of rotor axis trajectory is mapped to twodimensional digital image. By using antigrayscale method to preprocess the rotor axis trajectory image, the redundant color information of twodimensional digital image is removed. Then, the preprocessed digital images of rotor axis trajectory are input into deep convolutional neural network for training. The experimental results show that the rotor axis trajectory digital image after antigrayscale preprocessing has more advantages in the training of deep convolutional neural network. Compared with the traditional shape identification method of rotor axis trajectory, the shape identification method of rotor axis trajectory based on deep convolutional neural network has higher accuracy and better robustness.

TP183

上海市2019年度“科技创新行动计划”高新技术领域项目(19511103700)