基于混合高斯密度HMM的轴承故障诊断技术研究

Faults Diagnosis for Bearing Based on HMMs with Gaussian Mixtures

从轴承正常和故障条件下的实际机械系统中测试的振动信号入手,利用LPC系数进行特征提取,并把提取的特征矢量用于建立HMM隐状态下的高斯概率混合器模型,以便于引入到HMM的训练中,形成一个新的HMM类型-混合高斯密度HMM(GMD-HMM)。通过选取输出最高概率的HMM进行各种轴承故障类型的决策。通过异步电机系统的驱动端轴承的测试信号验证了该故障诊断方法的精确性。

Feature vectors based on the coefficients of LPC filters extracted from vibrations signals from both normal and faulty bearings were used to model Gaussian mixtures to represent observation probability densities of HMMs. So HMMs of various bearing conditions were trained. The method allows for diagnosis of the type of bearing fault by selecting the HMM with the highest probability. The method was tested with experimental data collected from an accelerometer measuring the vibration from the drive-end ball bearing of an induction motor mechanical system and has proven to be very accurate.