Mathematical top models control the blast furnace
We intensively use mathematical models to control and optimize the blast furnace process. They give us an image of the process in the blast furnace at any time and allow us to make changes if necessary. The specialised ArcelorMittal Gent Systems and Models department develops and maintains these models in close cooperation with the production departments.
A blast furnace is a large chemical reactor that converts iron ore to liquid hot metal. This takes place at a high temperature, under high pressure and in an environment containing much CO gas. We can therefore take few measurements in the blast furnace and greatly depend on mathematical models.
There are three kinds of models:
Models for the thermal situation
- Models that allow us to follow the thermal situation of the blast furnace in the short term (15 minutes). These models must particularly react quickly to changes in the situation because they help the operator maintain the required thermal situation.
- Models that give a more detailed image of the internal process in the blast furnace. These enable us to keep the blast furnace working optimally in the longer term.
- Classification algorithms that enable us to follow and interpret the available measurements.
To ensure the good operation of the blast furnace we keep the temperature of the hot metal constant. This may seem easy by measuring the temperature and adjusting the power supply to the blast furnace. We must, however, take account of the great slowness of the blast furnace process. Because of the great mass of material that must be heated, several hours sometimes pass between the cooling down of the blast furnace and the time at which the hot metal temperature also drops.
We can eliminate this delay thanks to the mathematical models. For example, we measure the composition of the gas that leaves the blast furnace at the top. Closely monitoring small changes in this composition allows us to calculate which chemical reactions are taking place at any time in the blast furnace. The blast furnace operator can always supply the appropriate amount of energy, because each reaction consumes a known quantity of energy.
Figure 1: the operator controls the blast furnace process from the control room.
In this way the operator in the control room can perfectly control the thermal situation of the blast furnace. This thermal model is however not the only one he has available. There are numerous models and measurements constantly providing him with information about the situation of the blast furnace, or even predicting the evolution in the near future. All this provides so much information that even experienced operators can use assistance now and again when interpreting the information. For this we work with an expert system. This is an automated system that collects all information about the blast furnace situation, then based on certain rules makes a diagnosis of the current status and also suggests action. The rules with which the system works are based on the experience of the operators. Thanks to the expert system the operators always obtain advice about action to be taken, but they ultimately make the decisions themselves.
Models for optimization in the longer term
Figure 2: an expert system advises the operator based on measurements and model results. The knowledge of the expert system originates from the operator.
It is insufficient to merely keep the blast furnace thermally under control. Due to the ever-changing environment (such as fluctuations in the quality of raw materials), other parameters must also be changed regularly in order to maintain optimal operation. The process engineers use all sorts of models for this. They have a thermal-chemical model of the blast furnace that models the whole gas flow in the blast furnace and the most important reactions. This gives a fairly complete picture of the inside of the blast furnace, only based on a number of measurements at the edge. Another example is the model for charge distribution that allows us to control the gas flow.
Classification algorithms for the following up of the measurements
Figure 3: using mathematical models we obtain a picture of the situation inside the blast furnace.
In some cases, the number of available measuring points is so high that we use models to divide the measurements into classes and present them in a simple way. An example of such a technique is the Self Organising Map from the world of neural networks. This can be used to follow the evolution of multidimensional measurements over a longer period.
We measure the temperature in different places in the wall of the blast furnace. This gives a temperature profile of the height of the blast furnace (figure. 4)
Figure 4: we measure the evolution of the temperature in the wall of the blast furnace.
To follow the evolution in time of this profile, we have the network calculate 16 typical patterns (classes in jargon) from profiles of a longer period (e.g. a year), and have them placed on a card: see figure 2.
From now on, we can calculate every day which of the 16 classes best matches the temperature profile of the day and mark this on the card (a point on the card). The best matching classes of the past days are also displayed (line on the card).
The blast furnace engineer can now see in one glance how the temperature profile has evolved in the past days. In this way, the information on a number of days and from different measuring points is reduced to a single line on the card.
Figure 5: with a neural network we divide the temperature profiles into 16 classes. In one glance you can follow the evolution of the past days.