中文摘要：

针对齿轮振动信号具有非线性和非平稳性的特点，采用去趋势波动分析对振动信号特征进行提取。应用该方法对采自实验台的齿轮振动信号进行分析，获得对数尺度-波动函数图。发现齿轮振动信号在不同时间尺度上具有不同的幂率关系，信号具有双标度性。分析了双标度产生的原因，提出将去趋势波动分析双对数图中小尺度下的尺度指数与截距组成齿轮振动信号的特征向量。应用高斯混合模型对100组不同故障模式的齿轮振动信号进行特征描述，然后采用最大贝叶斯分类器对50组齿轮测试信号进行分类，结果表明：应用该特征提取方法可获得较高的故障识别率。

英文摘要：

The nonlinearity and non-stationarity are two kinds of characteristics of gear fault vibration signals.The detrended fluctuation analysis （DFA）is introduced to extract the characteristics of vibration signals.With this method,the gear vibration signals acquired from experimental bench are analyzed.The logarithm scale-fluctuation function maps of DFA show that the signals are double-scaling which mean that there are different power-law correlated between scale and fluctuation amplitude in different time scale.The reason of double-scaling was discussed and a feature vector of gear vibration signal which consisted of the small scale exponent and the intercept was suggested.With proposed feature vector,100 groups of training signals corre-sponding to different fault conditions are described by Gaussian mixture model and 50 groups of test signals are classified by maximum Bayes classification.The results verify that higher failure recognition rate can be got by proposed feature extraction method.