中文摘要：

针对机械系统故障预测中微弱信号特征提取困难和预测精度不高的问题，引入滤波效果良好的小波相关滤波法和对信号微弱变化特征敏感的排列熵算法，以及具有较强的动态过程时序模式分类能力的隐马尔可夫模型，提出一种新的小波相关排列熵特征提取方法和基于隐马尔可夫模型的故障预测方法。对采集到的设备振动信号进行小波相关滤波处理，得到信噪比较高的各层小波系数，在此基础上计算小波系数的排列熵复杂度，构造信号沿各小波分解层分布的小波相关排列熵特征矢量，并据此构建相应的隐马尔可夫模型进行退化状态的识别和故障发生概率的预测。通过对滚动轴承全寿命数据的分析，验证了这种方法的有效性和优越性。

英文摘要：

To overcome the difficulty of feature extraction for weak signal and lower precision in fault prediction, wavelet trans- formation correlation filter method （WTCF） which is good at filtering, Permutation Entropy （PE） algorithm which is sensitive to weak variation of the signal, and hidden markov model （HMM） which has preferable classification capability for time series modes in dynamical process are introduced. A new method named wavelet correlation permutation entropy （WCPE） for feature extraction and HMM for fault prediction are proposed. The gathered vibration signal of equipment is processed by WTCF to obtain high signal-to-noise wavelet coefficients for each layer. Then their PE complexities are calculated to construct the WCPE feature vectors, which are employed to construct the HMM for degradation state recognition and fault probability prediction. The analytical results for full lifetime datasets of a certain bearing demonstrate the validity and advantage of the method.