The importance of models for intelligent data analysis in metallurgy

Dr. Sander Arnout, Dr. Els Nagels Michel Bridy, Dr. Mark W. Kennedy, Christian Landaas, Paul Hellinckx

Business intelligence tools translate data into information, and can be especially valuable in metallurgical plants. Indeed, a continuous stream of process data is often available, without conclusive information being led to all relevant personnel. Besides analysing past behaviour, real intelligence also requires the ability to make decisions, based on the prediction of expected performance. This can be achieved by integrating data management with process and flow sheet modelling, based on physics and chemistry. The use of models ensures a closed mass and energy balance, as well as explicit assumptions, which can be calibrated to real results. Furthermore, virtual instruments can be created, even in places where measurements are cumbersome or impossible. Thermodynamics have an important role in these process models, because of their predictive response to changes, even outside the normal operating window. This gives the process owner a large additional flexibility. The possibility to pre-assess materials before treating them, and to see the effect of settings changes before applying them, is important to keep the process in stable conditions and is therefore also a matter of safety. As a result, chemical equilibria, melting behaviour, etc. are highly important to select optimal process settings, as well as to determine potential margins when treating new materials. In this paper, we will illustrate how the use of models can enlighten relations between otherwise noisy data. A first example is the use of the composition dependent melting point as a predictor for furnace tap temperature. A second is the use of smelting results to predict time and resources needed downstream in refinery or converting. This is also important to select materials for which the margin of the whole production process is the highest. A third is the use of production data and theoretical relationships on fumed product quality for design of a new process. These examples show that data management and simple or advanced statistics are many times more powerful, when process understanding is integrated.