| The ethylene cracking furnace is one of the most important production units,which operational safety and stability has received tons of attention.Therefore,it is meaningful to realize its process monitoring.The naphtha cracking ethylene process is a continuous operation process with multiple operational conditions.In this paper,the multi-scale analysis of its process monitoring method is realized based on real industrial data.Firstly,data theoretical characteristic(1 Hz)is analyzed,including data auto-correlation and cross-correlation.By considering data auto-correlation and cross-correlation comprehensively,an inner and outer process monitoring model based on PLS-PCA method is proposed.For proposed PLS-PCA method,its outer monitoring model considers data auto-correlation in time-scale and its inner monitoring model considers cross-correlation among different variables in spatial-scale based on latent structure model.The PLS-PCA-based inner and outer process monitoring model has a good performance for process monitoring.Thus,the preliminary theoretical exploration for real industrial production processes data monitoring method is realized.However,in practical applications,the above proposed method is difficult to meet the actual requirements for online process monitoring.Therefore,by analyzing the multimodality of actual production data,a condition recognizer based on process operational variables is designed to identify active adjustments of operators.By combining its results,a monitoring algorithm system based on PCA method is established,which can realize the multiple operational condition process monitoring for actual industrial data with 98%TPR and 1.56%FAR.In addition,for industrial system with automatic control response,the accumulated T2 contribution value method is proposed to realize the fault diagnosis.By applying the above method online,the industrial application for continuous process monitoring and fault diagnosis is achieved with good performance. |
PDF Full Text Request