拟合函数系数特征的SVDD行星齿轮故障诊断方法

SVDD Fault Diagnosis Method of Planetary Gears based on Fitting Function Coefficient

行星齿轮振动信号复杂多变,离散型故障信号需进行降噪、降维等复杂的处理过程,且信号处理过程中易造成信息缺失等。针对以上存在的问题,提出将函数型数据分析(Functional data analysis,FDA)与改进的支持向量数据描述(Support vector domain description,SVDD)相结合的故障诊断方法。根据齿轮不同故障类型建立不同的故障拟合基函数,将训练集数据与傅里叶基函数进行拟合,根据拟合得到的函数系数特征建立SVDD模型,并以ROC(Receiver operating characteristic)的评价函数为优化目标使用模拟退火算法对SVDD模型中的核参数σ和惩罚因子c进行优化;将不同的测试样本带入SVDD模型中,通过计算测试样本到超球体球心的相对距离来识别故障种类,进而完成行星齿轮的故障诊断。实验结果对比表明,本文中提出的方法能够解决离散型故障信号处理复杂、信息丢失等问题,准确地识别行星齿轮故障种类。

For the problems of complex and variable vibration signals of planetary gears, feature extraction distortion and small sample size, a new fault diagnosis method combining function data analysis with improved support vector domain description(SVDD) is proposed. According to different fault types of gears, different fault fitting basis functions are established, the training set data is fitted with the Fourier basis function, and the SVDD model is established according to the function coefficient features obtained by fitting, and in a ROC(receiver operating characteristic) of the evaluation function optimization objective simulated annealing algorithm kernel parameter model SVDD penalty factors σ/c and optimized;different test samples are taken into the SVDD model, and the fault type is determined by calculating the relative distance between the test sample and the center of the ball, thereby completing the planetary gear fault diagnosis. The comparison of experimental results shows that the proposed method can accurately identify the types of planetary gear faults.