基于AR-连续HMM的故障诊断模型及应用

在状态监测与故障诊断中,被测设备的状态一般不能直接观察到,要通过测量被测设备的表现来感知,这和隐马尔可夫模型(HMM)在本质是相通的。因此可以利用连续高斯密度混合HMM分析被测设备的振动信号,首先以AR模型系数为特征,研究不同状态数与不同混合高斯数对HMM模型分类的影响,再利用较优的状态数与混合高斯数HMM模型进行状态监测和故障诊断,诊断与对比实验结果表明该方法能利用少量样本进行训练和有效诊断。

A Diagnosis Model Based on AR-Continuous HMM and Its Application

In condition monitoring and fault diagnosis, because the state of the unit under test(UUT) cannot be observed directly, it should be judged by its behavior. This is similar to Hidden Markov Model(HMM) in nature, so continuous Gaussian mixture HMM is adopted here to analyze the vibration signals of UUT. First through the features based on the reflection coefficients of AR model extracted from vibration signals, the influence of different number of states and Gauss numbers on HMM are investigated, then the HMM with better number of state and Gauss number is used to monitor and diagnose the rolling-bearing′s conditions. The result shows that the proposed method is effective for diagnosis problem with small training samples.