基于几乎周期时变AR模型的故障早期预报

针对旋转机械振动信号具有循环平稳性的特点,提出一种基于几乎周期时变AR模型的故障早期预报方法,推导了基于循环统计量的辨识算法,其具有抑制加性平稳噪声的优点;并利用该模型对转子早期碰摩试验数据进行了分析,首先根据正常状态数据辨识模型参数,然后对故障状态时的残差信号进行峭度分析,结果表明该方法可以检测早期故障并预报未知故障。

EARLY FAULT FORECASTING BASED ON ALMOST PERIODIC TIME- VARYING AUTOREGRESSIVE MODEL

According to the characteristics of vibration signal in rotating machinery, one novel method of early fault forecasting based on almost periodic time varying autoregressive (APTV-AR) model is presented and the algorithm of identifying parameters based on higher order cyclic-statistics (HOCS) is proposed, which has the advantage of suppress additive stationary noise. In the end vibration signals from rotors with early rub-impact are analyzed with the APTV-AR model. At first, model parameters are identified under normal condition and then each kurtosis of residual signal under faulty conditions is calculated. The results demonstrate that the proposed method can detect early faults and forecast unknown faults.