基于FCM建模的DKICA间歇过程故障诊断

DKICA Fault Diagnosis of Batch Process Based on FCM Clustering Modeling

由于对间歇过程应用KICA算法进行在线监控时需要预测从当前时刻到反应结束时刻的测量数据,同时间歇过程具有多个生产模态,应用KICA算法进行独立建模时,模型结构复杂,计算量大。针对间歇过程数据预估不准确和多模态计算量大的问题,提出了一种基于FCM聚类建模的双核独立元分析(DKICA)间歇过程故障诊断算法,该算法首先对间歇过程数据进行批次展开与变量展开,构建二维建模数据单元,然后将展开后的高维数据按FCM聚类进行分类,再对分类后的每个聚类模块应用DKICA算法提取独立元,建立每个聚类模块的诊断模型,并计算I^2和SPE统计量及相应的控制限,跟踪过程的运行状态,及时检测到故障的发生。将该方法应用于DuPont间歇聚合反应过程中,仿真结果验证了所提出方法的可行性和有效性,并显示出比KICA更好的诊断效果。

Measured data are forecasted from the current moment to the end moment of reaction when KICA algorithm is applied to monitor on-line in batch process. At the same time, batch process has multi production modality. So model structure is complex and calcu- lated amount is very large. Aiming at the problems about inaccuracy of data estimate and large calculated amount of multimode, a DKI- CA Fault diagnosis algorithm of batch process based on FCM cluster modeling is proposed in this paper. First this algorithm can process batch expansion and variable expansion for data of batch process and structure data units of two dimension model, then classify dimen- sional expansion data according to FCM cluster. Every clustering diagnostic model is set up by abstracting independent components by DKICA algorithm. This can be used to calculate statistics and control limits of 12 and SPE, follow running status and detect process faults timely. Simulation results of DuPont batch polymerization process have confirmed feasibility and effectiveness of the proposed algorithm which shows better diagnosis effect than KICA.