动态局部邻域主多项式分析故障检测研究

针对复杂工业过程中存在的动态性、多模态及非线性等特征,提出一种动态局部邻域主多项式分析(DNSPPA)的故障检测算法。首先,设置一定长度的时间窗来描述样本点之间的时序相关关系;其次,寻找时间窗内样本在空间方向上的局部近邻集,利用近邻集对数据样本进行标准化处理;最后,在标准化处理后的数据上建立PPA模型,计算统计量并确立控制限进行故障检测。DNSPPA方法能解决复杂工业过程中的动态时序问题和多模态数据中心漂移的问题,从而降低多模态结构对PPA检测性能的影响。使用具有动态特征的多模态非线性数值例子和青霉素数据对DNSPPA方法进行仿真测试,并与主元分析法(PCA)、主多项式分析法(PPA)进行对比,仿真结果表明,DNSPPA方法能更加及时地检测到故障,且故障检测率较高。

Fault Detection of Industrial Process Based on Dynamic Nearest Neighborhood Standardization of Principal Polynomial Analysis

Aiming at the dynamic, multi-modal and nonlinear characteristics of complex industrial processes, a fault detection algorithm based on dynamic nearest neighborhood ntandardization and principal components analysis(DNSPPA) is proposed in this paper. Firstly, a certain length of time window is set to describe the temporal correlation between the sample points. Secondly, the local nearest neighbor set of the sample in the spatial direction in the time window is found, and the nearest neighbor set is used to standardize the data sample.Finally, the PPA model is established based on the standardized data to calculate the statistics and establish the control limit for fault detection. DNSPPA method can solve the problem of dynamic time sequence and multimodal data center drift in complex industrial processes, so as to reduce the impact of multi-modal structure on PPA detection performance. To demonstrate its effectiveness and superiority, the proposed DNSPPA method is tested by multi-modal nonlinear numerical examples with dynamic characteristics and penicillin data. Compared with principal component analysis(PCA) and principal polynomial analysis(PPA), the proposed method in this paper can effectively detect the faults and improve the detection rate.