Process Monitoring Method Of Running State In The Zinc Roasting Process Based On Distributed LOF-PCA

Zinc roasting process(ZRP)is a large-scale production process with many subsystems,which has the characteristics of multivariable,strong coupling,large time delay,non-Gaussian distribution,etc.The complex and changeable raw materials and harsh production environment lead to the frequent fluctuations of working conditions and many faults of ZRP.Timely and accurate detection of abnormal working conditions can ensure stability,and improve production efficiency of the ZRP.In this paper,focusing on the features of the ZRP,the research on the working condition monitoring method of the non-Gaussian process under the distributed framework is carried out.The main research work and innovations are as follows:(1)To solve the problem of the non-Gaussian distribution in ZRP,a local outlier factor-based PCA(LOF-PCA)monitoring method is proposed.When determining the confidence region of the T~2 and SPE statistics with the traditional principal component analysis(PCA)monitoring model,the process data needs to follow the Gaussian distribution.Therefore,in this paper,the density-based LOF algorithm is used to construct new monitoring statistics.Then,the kernel density estimation method is used to obtain the corresponding monitoring threshold,which can effectively reduce the impact of data distribution and outliers on the monitoring performance.Finally,the case studies of the Tennessee Eastman(TE)process and the ZRP show that the false alarm rate(FAR)and non-detection rate(NDR)are significantly reduced based on the proposed method.(2)To solve the problem of large lag,data coupling and wide distribution in the ZRP,a distributed LOF-PCA process monitoring method is proposed.First,because the traditional single monitoring model is difficult to mine all the key information from the complex and coupled process variables,this paper divides the process variables in different temporal and spatial dimensions into several sub-blocks according to the correlation and redundancy of variables.Then,the LOF-PCA method is used to establish a monitoring model for each sub-block,and the Bayesian inference method is used to fuse all the monitoring results of the sub-blocks into a global monitoring statistic to achieve the fast detection of abnormal working states.On this foundation,a fault diagnosis method based on weighted contribution plot is proposed to realize the rapid location of fault related variables.Finally,the case studies of the TE process and the ZRP are used to verify the effectiveness and superiority of the proposed process monitoring method.(3)Based on the actual production demands in the ZRP,the working conditions monitoring system of the ZRP based on the distributed LOF-PCA is developed,which realizes the functions of real-time monitoring of key parameters,working conditions monitoring and regulation,historical data query,report curve printing,providing effective operation guidance for operators.