基于改进粒子群算法的提取齿轮磨损特征方法

A Feature Extraction Method Based on Advanced Particle Swarm Optimization for Recognizing Wear Fault of Gear

摘要:

针对轻微齿轮磨损故障信号在啮合频率和边频带上的幅值特征难以将其与正常信号区分的问题,提出用Laplace小波提取信号的粘滞阻尼比作为一种冲击特征,同时提取对冲击信号敏感的时域和频域峭度指标作为另外两种特征;又由于传统Laplace小波匹配方法存在计算耗时和精度不高的缺点,采用一种基于概率模型的改进粒子群算法以快速找到最佳匹配结果,从而得到其阻尼参数;将上述3种冲击特征用于农用拖拉机变速箱的齿轮磨损故障识别,其结果表明提出的方法在错分率上比基于啮合频率和边频带幅值特征的方法降低了12%。

Abstract:

Due to the amplitude features of the meshing frequency and the side frequency useless to distinguish the normal status and the slight wear fault, three shock features are extracted in this paper. One of the features called viscous damping ratio is extracted by matching the Laplace wavelet with the acquired signals, while another two features are the kurtosis values in the time-domain and the frequency-domain. Since the traditional matching method of Laplace wavelet is time-consumed and accuracy-low, a new method is proposed based on an advanced particle swarm optimization algorithm, where a probability model is adopted. Experiments on using the three proposed features to recognize the wear fault of gear in a farm tractor show the proposed features obtain 0.12 lower in error rate than the amplitude features of the meshing frequency and the side frequency.