BLAST FURNACE THERMAL CONTROL AND AERODYNAMIC MONITORING USING ADVANCED SIGNAL PROCESSING METHODS, IN CORUS UK

摘要

On line expert systems are employed in the control and monitoring of Redcar's 14m blast furnace and Scunthorpe's Queens Anne and Victoria, both of which are 9m blast furnaces. The expert systems at both plants are based on the standard G2 platform from Gensym. The objective of the installed systems is to both standardise and improve the quality of metallurgical process control decision-making. Most aspects of furnace operation are monitored including hearth liquid levels, raw materials quality, wall accretion monitoring and cooling water leakage. Trend analysis is applied to both top gas hydrogen and under-tuyere temperatures to determine whether cooling water leakage is entering the furnace. Thermal control and aerodynamic monitoring however form the central core of the systems and it is these aspects that are considered in this paper. Thermal control: Both plants adjust the injectant rate set point at regular intervals to compensate for variations between the achieved injectant rate and the set point. These variations are caused by changes in specific oxygen consumption due to direct reduction variability. At Redcar, changes are also made in response to significant drift in direct reduction by comparing the long and short-term trend. This has prevented significant reductions in hot metal quality. At Scunthorpe, a statistical process control approach is applied to enable or disable further changes to the injectant rate based on actual and predicted metal quality. The reduction in the frequency of changes, and more appropriate changes, has yielded a reduction in hot metal silicon variability of around 10%. Aerodynamic monitoring: Results from the recent 5th Framework CHEM project are described. Multivariate SPC models, generated using principal component analysis, are used to monitor process stability. Stack pressure differentials and top gas analysis signals are the key variables. A G2 module has been developed to run such models on line and undertake contribution analysis when warning or action limits are exceeded. The module allows identification of the signals that have changed most significantly. A second module has also been developed in G2 to classify the trend in each of the two principal components by generating a sequence of discrete events, which can then be analysed. Using both of these techniques it was possible to predict the most significant slips and channels over a 2-year period. Management and implementation issues are discussed, in particular the stepwise introduction of new procedures, and the style and content of the advice given. Clear simple action messages are required with supporting information displayed on supplementary workspaces.