基于SAX的关联规则挖掘方法在故障诊断中的应用

Application of SAX-based Association Rule Mining on Fault Diagnosis

在利用关联规则进行故障信息挖掘时,需要将连续型数据离散化和区间化。离散化效果决定了关联规则挖掘的效果,传统的均匀区段划分法忽略了数据的分布特点,加权划分法和模糊指数法均存在权值选择问题。鉴于此,提出基于符号聚合近似(SAX)的关联规则挖掘方法。首先对振动信号进行特征提取,然后利用SAX方法自适应对特征值数据离散化,从而实现关联规则挖掘,进行故障分析和信息提取,最后利用挖掘结果进行故障诊断。转子故障模拟试验分析结果表明:与等宽度和等密度离散化方法相比,该方法可以更好地进行数据离散化,实现故障信息挖掘和诊断。基于SAX的关联规则挖掘方法对试验数据和真实设备之间的数据通用具有良好的鲁棒性,便于进行实际应用。

When mining fault information using association rule,consecutive data need to be discretized and regionalized. The result of association rule mining is determined by the effect of discretization. Traditional uniform partitioning approach neglects the distribution characteristics of data,and both weighted partitioning methods and fuzzy index method have the problem of choosing weight values. To address the issue,association rule mining method based on SAX is proposed. Firstly,feature extraction on vibration signals is conducted,and then the characteristic value of the data are discretized adaptively by SAX,thus,association rule mining can be realized to conduct fault analysis and information extraction. Fault diagnosis could be done using mining results. The analysis results of rotor fault simulation experiment showed that,compared with equal density and equal width discretized approach,the proposed method could carry out a better data discretization and realize mining and diagnosis of fault information. The SAX method has good robustness to the experiment data and the real equipment data,and is convenient for practical application.