中文摘要：

在非线性时间序列预测研究的基础上，提出非线性预测效果的特征提取方法。首先对采集到的足够长轴承数据采用小波变换进行消噪处理及边界延拓，使其满足预测需要的无限长、无噪声的条件，这样延迟时间取任意值均能重构原系统相空间；然后采用基于可预测性的选取嵌入维数的方法确定轴承各种状态信号的嵌入维数，进行相空间重构。应用实验结果表明：该方法提取的特征值能明显地区分轴承各种状态信号，且对数据分段长度的稳定性好，可以作为识别轴承故障的一种新途径。

英文摘要：

Based on nonlinear time series prediction, a feature extraction method with nonlinear prediction effect is proposed to describe bearing fault quantitatively. First, wavelet transform method is used to denoise the bearing signal and make boundary extension to meet the nonlinear prediction requirement. In infinite long and noise free condition, any value of delay time can be adopted to reconstruct phase space of the original system ; an approach of determining the embedding dimension based on nonlinear predictability is then introduced to determine the embedding dimension of the bearing signals in all status and reconstruct phase space. Application experiment results show that the features extracted using this method can recognize all kinds of status signals of bearing clearly. The method has good stability to data segmentation length and can be used as a new approach to descript bearing faults.