车辆传动系统变参小波流形融合故障诊断方法

应用流形学习方法非线性融合信号在不同小波参数下中央尺度对应的小波包络，研究了强背景噪声下车辆传动系统振动信号故障瞬态脉冲包络的有效提取问题，并与传统信号时频分解方法进行了对比研究；采用不同小波参数对振动信号进行连续小波变换，提取了每组参数下中央尺度上的小波包络；采用基尼指数选择若干包含故障瞬态脉冲信息的小波包络，构造了高维小波包络矩阵；采用局部切空间排列算法对高维小波包络进行流形融合，获得了反映故障瞬态脉冲包络本质结构的小波包络流形；为了验证所提方法的有效性和优越性，采用不同方法对轨道车辆轮对轴承和汽车变速齿轮箱的故障振动信号进行了对比分析。研究结果表明:在分析轴承外圈故障信号时，所提方法基尼指数比传统信号时频分解方法提高27.32%以上；在分析齿轮磨损故障信号时，所提方法基尼指数比传统信号时频分解方法提高26.74%以上。可见，所提方法通过综合具有不同形态的变参小波包络，可以在无需优化小波参数情况下，对车辆传动系统中的不同关键部件故障振动信号具有较好的自适应性，提取的故障脉冲包络中的带内噪声少，故障脉冲特性明显，容易识别其频谱中的故障特征频率，是检测车辆传动系统故障的一种有效方法。 

Fault diagnosis method of vehicle transmission system based on manifold fusion of parameter-varying wavelet

The manifold learning method was utilized to nonlinearly fuse the wavelet envelopes corresponding to the central scale under different wavelet parameters. The problem of effective extraction of the fault transient impulse envelopes from the vibration signals in vehicle transmission systems under heavy background noise was studied, and a comparative study with the traditional time-frequency methods for signal decomposition was carried out. Different wavelet parameters were adopted for the vibration signals to perform continuous wavelet transform (CWT), and the wavelet envelope corresponding to the central scale under each group of wavelet parameters was extracted. Some wavelet envelopes containing the information of the fault transient impulses were selected by Gini index, and the high-dimensional matrix of wavelet envelopes was constructed. The high-dimensional wavelet envelopes were fused based on manifold by using the local tangent space alignment (LTSA) algorithm, and the wavelet envelope manifold reflecting the intrinsic structure of fault transient impulse envelopes was obtained. In order to verify the effectiveness and superiority of the proposed method, the fault vibration signals of a railway-vehicle-wheelset bearing and an automobile change-speed gearbox were analyzed comparatively by different methods. Research results indicate that, compared to the traditional time-frequency methods for signal decomposition, the proposed method can improve the Gini index by over 27.32% in the case of bearing signal with an outer-race fault, and by over 26.74% in the case of gearbox signal with a wearing fault. It can be seen that, by synthesizing the parameter-varying wavelet envelopes with different patterns, the proposed method is well adaptive to the vibration signals for the faults of different key parts in the vehicle transmission systems with no need for wavelet parameter optimization. The extracted envelopes of the fault transient impulses have less in-band noise and distinct fault impulsive features. These advantages are beneficial to the identification of the fault characteristic frequencies in the spectra of the extracted envelopes. Therefore, the proposed method is an effective method for the fault detection of vehicle transmission systems.