数字工厂研究室知识库

针对现代工业过程数据的高维性和分布复杂性等问题,提出了一种基于IJB-PCA-ICA(Improved Jarque-Bera-Principal component analysis-Independent component analysis)的故障检测方法。首先采用改进的Jarque-Bera检测方法(JBtest)对原始数据划分高斯与非高斯核心部分,并对其中的高斯性与非高斯性均不明显的变量划分半高斯部分。将半高斯部分通过高斯分布置信概率加权与高斯核心部分和非高斯核心部分分别建立高斯子空间和分高斯子空间,然后对高斯子空间进行相关性划分后并采用PCA方法得到高斯子空间的统计量;对非高斯子空间进行主元投影划分后并采用ICA方法得到非高斯子空间的统计量,接着通过贝叶斯推断得到的联合统计量进行故障检测。最后通过田纳西-伊斯曼(TE)仿真实验,有效验证了所提出方法的有效性。

In view of the high dimensionality and the distribution complexity of modern industrial data, a fault detection method based on IJB-PCA-ICA (Improved Jarque-Bera-Principal component analysis-Independent component analysis) is proposed in this paper. Through the method of Jarque-Bera test (J-B test), the original data are divided into Gaussian part non-Gaussian part and semi-Gaussian part. The semi-Gaussian part divided from those variables with not obvious Gaussianity or non-Gaussianity is weighted to participate into Gaussian subspace and non-Gaussian subspace by the Gaussian confidence probability. After the partition by the correlation and principal component projection, the statistics of the Gaussian and non-Gaussian subspaces are obtained by PCA and ICA, respectively. Then the Bayesian inference is applied to obtain the comprehensive statistics for fault detection.