基于NND-PCASVDD的多模态工业过程故障检测

Multi Modal Industrial Process Fault Detection Based on NND-PCASVDD

现代工业生产过程存在多种运行模态,有效的故障检测技术能够保证生产的正常进行.作为一种单模态故障检测算法的主元分析支持向量数据描述(Principal Component Analysis Support Vector Data Description,PCASVDD)对多模态故障的检测存在局限性.为解决这一问题,提出一种基于近邻差分(Nearest Neighbors Difference,NND)和PCASVDD算法结合的多模态过程故障检测算法(NND-PCASVDD).首先,应用NND剔除数据多模态结构;接下来,对差分后的数据应用PCA算法进行数据降维;最后,SVDD作为一种故障检测器对降维之后的主元空间数据进行检测.对于多模态过程故障检测,NND-PCASVDD不要求过程知识和多模型建模,只需要单独的一个模型,符合PCASVDD单模态故障检测要求.通过一个多模态数值例子和半导体生产过程对该方法的有效性进行验证.实验结果表明该方法优于PCASVDD.

There are many operating modes in modern industrial production process.Effective fault detection technology can ensure the normal operation of production.As a single modal fault detection algorithm,Principal Component Support Vector Data Description(PCASVDD)has limitations in the detection of multimodal processes.In order to solve this problem,we propose a new multimodal process fault detection algorithm(NND-PCASVDD)based on the combination of Nearest Neighbors Difference(NND)and PCASVDD algorithm.First of all,NND is used to eliminate the multi-modal structure of data.Then,PCA algorithm is applied to reduce the dimension of differential data.Finally,SVDD is used as a fault detector to detect the dimension-reduced principal component space data.For multi modal process fault detection,NND-PCASVDD does not require process knowledge and multi model modeling,but only need a single model,which meets the requirements of PCASVDD single modal fault detection.The validity of the method is verified by a multimodal numerical example and a semiconductor production process.The experimental results show that the method is superior to PCASVDD.