Multivariate Statistical Methods In A Confined Application In Fault Diagnosis Of Blast Furnace

In Imperial Smelt Furnace, smelting Pb-Zn process is very complex, some problems can exist in this process, such as huge technical structure, control parameter with a large number of data, and large delay of Imperial Smelt Furnace. Because traditional Fault Diagnosis methodology needs to establish accurate mathematical model, it faced many difficulties in real application which cannot be solved. As Fault Diagnosis methodology based on multivariable statistical analysis has some characteristics independent of process models and easy implementation, multivariable statistical control method was adopted in this paper, which combined Principal Component Analysis and multivariate statistical control charts to monitor fault diagnose in the whole process, consequently improve security and reliability of smelting process.First of all, some common methods of multivariable statistical fault diagnosis were studied in this paper. Through selecting a section of data from large number of real data of Imperial Smelt Furnace under usual situation, statistical and monitoring model were established for Principal Component Analysis process. On the one hand, based on different possibility occurred in multivariable statistical control charts in different working situations, the relationship between T~2 and SPE statistical variables and production process was analyzed. Moreover, through the analysis of real charts under usual and default conditions, the relationship between T~2, SPE statistical variables and defaults in production process was further illustrated. On the other hand, combining Principal Component score charts and Principal Component loading charts, defaults were analyzed. Then default resources were achieved. Experiment results illustrated that Principal Component Analysis method can fast and effectively reflect the changes in production process. Production results proved that this method can improve real-time monitoring capacities for Imperial Smelt Furnace production situation and production efficiencies.Secondly, in order to solve large delay of Imperial Smelt Furnace, Dynamic Principal Component Analysis method was adopted in this paper. Through calculating real data of Imperial Smelt Furnace, length of time lag was gained in the process, and then statistical and monitoring model was built for Dynamic Principal Component Analysis process. Also, by comparing with traditional Principal Component Analysis method, it shows that Dynamic Principal Component Analysis method is more accurate. Furthermore, the simulation of statistical control charts was given under different interferential signals, and the results proved that this method is suitable for changes in working situations, reduce misdiagnosis and improve sensitivities of detections.