基于改进冗余提升方案的汽轮机组振动故障特征提取

故障特征提取是大型机械设备状态监测和故障诊断领域的核心问题。传统的振动故障特征提取方法主要是基于频谱分析的方法，小波变换的出现则为该领域提供了新的工具。文中提出并构造了一种改进的冗余提升小波变换算法来提取振动信号的时域特征。算法以第2代小波为基础，设计了冗余提升小波变换的算法，不进行分裂，直接利用构造的算子进行预测和更新，各层分量和原始信号的数据长度相同，从而保留了更多的时域信息。研究了提升小波和冗余提升小波算法中存在的频率混叠问题，阐述了产生频率混叠的原因。通过对冗余提升小波分解得到的近似信号和细节信号采用傅里叶变换的方法消除了与其对应频带无关的频率成分，以突出相应频带信号的时域特征。对仿真信号和实际汽轮发电机组振动故障信号进行了分析，结果表明，改进的冗余提升小波变换算法能够较理想地提取出故障特征，有效地解决了提升小波算法中存在的频率混叠问题。

Turbine Vibration Fault Feature Extraction Based on Improved Redundant Lifting Scheme

The extraction of fault feature is the key to the condition monitoring and fault diagnosis of large mechanical equipment. The traditional feature extraction methods for vibration fault are mainly based on spectrum analysis, while the development of wavelet transform brings a new tool in this field. An improved redundant lifting wavelet transform was utilized to extract the feature in time domain of fault vibration. This algorithm was based on the second generation wavelet transform and a redundant lifting algorithm is designed. The splitting operation was unnecessary in this algorithm and the signal was predicted and updated directly, so the approximation and detail signal at each level has the same length as the original signal and retains more characteristics in time domain. The causes of frequency aliasing inhere in both lifting and redundant lifting wavelet transform was discussed and also an anti-aliasing algorithm for redundant lifting wavelet transform was presented. A typical simulated signal and real-life vibration fault signal measured from a turbine generator unit were used to test this method. It was shown that the presented algorithm is quite effective for avoiding frequent aliasing and the typical feature of impact-rub and misalignment of turbine generator in time domain are desirably extracted.