Modeling and identification of a steelmaking process

TAB (Tout au Barillet) coking ovens are highly disturbed units. Starting with coal and firing at high temperature, they have to produce steelmaking coke and coke oven gas, which, along with the furnaces' charging and discharging cycles, are supplied at highly variable rates and in sharp bursts. These gases, coming from a battery of furnaces, are collected in large pipes known as "barrels", which flow into a large manifold; the latter carries the production to a washing and purification unit.

The quest for improved performance and environmental constraints make it necessary to improve barrel pressure regulation.

Indeed, if this pressure is lower than atmospheric pressure, air inlets risk creating internal combustions, and if the pressure is too high, leaks, which are inevitably always present in this type of installation, risk releasing noxious gases into the atmosphere. To strictly meet these objectives, high-performance control is required.

The controllers traditionally used on this type of process by ordinary steelworks are of the standard PID type, but this control technique is at the limit of its potential if the performance targets are increased in the current drive for competitiveness.

Given the multivariable, non-linear and highly disturbed nature of the process, a predictive control system is required, capable of taking the measured disturbances as a trend a priori, as is the case here. The advantages of this type of control are well known [3] : robustness, performance, ease of adjustment, constraints taken into account, etc., which is why it is rapidly spreading to all industrial sectors, both slow (e.g. furnaces) and fast (e.g. rolling mills). However, it also has its drawbacks: it requires a specific computer, and is based on the use of a predictive model which, in this case, needs to be developed and identified.

This text follows on from article [R 7 140] (see archive) "Process modeling and identification", of which it is an example of application. Readers should therefore first familiarize themselves with this text.