基于NNDSVDD的多模态工业过程故障检测

Fault Detection in Multimode Industrial Processes Based on NNDSVDD

现代工业存在多模式运行环境,传统的支持向量数据描述(support vector data description,SVDD)作为一种单模态故障检测算法对多模态过程故障的检测存在局限性.针对这一问题,提出一种基于近邻差分(nearest neighbors difference,NND)算法和SVDD算法相结合的多模态工业过程故障检测方法(NNDSVDD).首先,应用NND预处理多模态数据;然后,在差分数据集上应用SVDD确定统计量的控制限;最后,计算测试数据的统计值将其与控制限比较确定测试数据的状态.近邻差分算法剔除数据的多模态结构为SVDD提供良好的数据基础;在差分数据集的基础上应用SVDD,提高了传统SVDD对多模态过程故障的检测能力.将NNDSVDD应用于数值模拟例子和半导体生产过程进行仿真测试,与传统SVDD相比,NNDSVDD的故障检测率均能达到100%,验证了本文方法的有效性.

Modern industrial manufacturing usually involves multimode processes but the traditional support vector data description （SVDD） algorithm is a single-mode fault detection method and is therefore unsuitable for multimodal processes. To address the above shortcoming, we propose a multimode process monitoring method that combines the nearest neighbors difference （NND） and SVDD （NNDSVDD） algorithms. First, we use NND method to eliminate the muhimode structure of an original data set and to obtain a difference data set. Then, we implement the SVDD algorithm in the difference data set to determine the statistical control limit. Lastly, we calculate the statistical values of the test data and compare them with the control limit to determine the test data as being in either a normal or fault state. NND eliminates the multimodal data structure and provides a good data foundation for SVDD. By applying SVDD to difference data sets, we improve the fault detection ability of SVDD in multimodal processes. We applied the NNDSVDD in numerical simulation exampies and a semiconductor manufacturing process to verify the performance of our proposed method. The fault detection rate of the NNDSVDD reached 100% in each case and the simulation results indicate that the proposed method outperforms the conventional SV DD.