基于关联积分的非线性降噪和故障诊断技术

Technique for Nonlinear Denoise and Fault Diagnosis Based on Cross-Correlation Integral

基于交叉关联积分算法,利用重叠滑动窗口对非平稳信号进行分段比较,给出了一种故障诊断和预报的方法.对滑动窗口中的一维非平稳测量信号,首先采用正向过阈值变换的方法,将原始测量信号转变为完全包含系统固有信息的间期时间序列,并用延时嵌入的相空间重构方法,对间期时间序列进行相空间重构,从而实现系统固有动力学特性的再现;与此同时对测量信号进行非线性复合消噪处理;最后通过检测各滑动窗口相对基准窗口的关联积分的变化,对系统故障进行诊断和预报.实例结果表明,此故障诊断方法具有准确率高、实时性等优点,可望用于大型、复杂系统的故障诊断.

A technique is proposed for diagnosing and predicting faults based on the algorithm of cross-correlation integral and the comparison of the unstable signals by using overlapping sliding windows. For each sliding window, the signal is firstly transformed to pure phase information by taking the sequence of time intervals unidirectionally across the threshold. Furthermore, the time intervals are used for reconstructing the system's original phase space dynamical information by way of the delay embedding reconstruction. Finally, with these steps to reduce noise influence, the system faults are diagnosed and predicted by testing the variation of values of cross-correlation integral. The computational results show that the technique is reliable and real-time, and it could be expected to be applied to the diagnosis of large complex equipment.