• 基于差分的动态加权SVDD在多模态过程故障检测中的应用

    Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-10-11 Cooperative journals: 《计算机应用研究》

    Abstract: There are many operating modes in modern industrial production processes, and there is a strong correlation between data sequences. Traditional SVDD as a single-mode static fault detection algorithm, it is difficult to ensure the accuracy and real-time performance of multi-mode dynamic process fault detection. In order to solve this problem, this paper propose a weighted dynamic SVDD monitoring method (NND-DWSVDD) base on nearest neighbor difference . First, use NND to eliminate the data multimodal structure and ensure that the process data obeys the unimodal distribution; then, introduce the dynamic method for the differentially processed data and add weights to highlight useful information. Finally, establish a monitoring model by using the SVDD method to achieve online monitoring. NND-DWSVDD improves the multi-modal dynamic process fault detection rate. For multimodal dynamic process fault detection, NND-DWSVDD does not require multi-model modeling, and only need a single model. It meet single-modal fault detection requirements. Through multi-modal numerical example and semiconductor production process data to validate the effectiveness of the method.