STEELING A MARCH on plant control

British Steel operates nine blast furnaces on integrated sites at Redcar, Scunthorpe, Port Talbot and Llanwern. Together these furnaces produced 12.5mt of iron in 1996, and range in output from 2600 to 9200 t/day.

During the late 1980s and early 1990s the plants were rebuilt or relined. Process control equipment was upgraded to include many new sensors, leading to further improvements.

The performance of a blast furnace can be judged by several factors:

* Reductant rate: the weight of coke and tuyere injectant, coal or oil, needed to produce 1t of iron (typically 500kg/ t).

* Product quality, measured in terms of temperature and silicon content.

* Plant availability: avoiding unplanned maintenance.

* Campaign life: the quantity of iron produced before the furnace needs relining or rebuilding; a capital investment in the order of £80m.

A saving of 1 kg/t on reductant rate (0.2%) on a unit producing 2m t/a is worth about £150,000.

British Steel's increasing numbers of sensors, reductions in shift manning and development of new data assessment techniques make the process an ideal candidate for an operator guidance system. For example, the Redcar blast furnace has more than 50 pressure tappings and 250 thermocouples amongst its systems which monitor at least 1000 parameters.

British Steel's Teesside Technology Centre believed that the company's blast furnaces could benefit from a knowledge based expert system and selected Gensym's G2 system. The initial development took place at Llanwern's No 3 blast furnace and control systems supplier Gensym visited the plant in order to establish, in conjunction with BSTTC staff, the basic system's infrastructure.

The blast furnace process takes place in a water-cooled refractory lined pressure vessel in which prepared iron ore (known as ferrous burden) is reduced to liquid iron and slag.

The ferrous burden is charged (under control) at the furnace top with alternate batches of coke. Hot blast at a constant temperature of 1000-1150oC is blown into the furnace at 3-4 bar through copper water-cooled tuyeres (ports) around the lower part.

Also injected through the tuyere is a secondary reductant, the 'injectant', either coal or oil. Ascending gases chemically reduce and heat the descending ferrous burden, which melts to form liquid iron and slag. These liquids are periodically tapped from the furnace.

Immediate thermal controls available are adjustment of rates of injectant, coke or steam. Control for pressure instability is via the blast volume. Longer term control is to adjust the distribution of coke and ferrous burden at the furnace top, which controls the radial gas distribution within the furnace stack. Moreover, precise control of casting is crucial to a smooth operation. Diameter and length of taphole can be preset.

Many novel techniques have been developed to aid the shift controller by combining and analysing the large volume of data available: predictive control, neural network pattern recognition and fuzzy logic are all either in use or under investigation. Other dynamic models have been developed, such as temperature prediction, thermocouple monitoring for accretion, descent rate monitoring and melt level calculation.

Llanwern Development

Expert systems had already demonstrated their ability to improve process performance. To make further progress in achieving improved blast furnace performance and availability, application of advanced software technology was seen to be appropriate.

In 1995, the Strip Products business of British Steel invested in a full expert system shell for the 2m t/a No 3 Blast Furnace.

The work was led by BSTTC, British Steel's central process technology development establishment. A number of expert system shells were assessed by the Computing Applications Section and Gensym's G2 was deemed to be the most appropriate package.

There was a requirement on the plant for long term data storage, to be used by plant technical staff, and British Steel decided that a database would be installed, for its on-line data storage and for process analysis. DEC RDB was selected as the database management system. It is installed on a dedicated DEC Alpha machine running under the VMS operating system and linked to the existing plant VAX.

The purpose of the system is to improve product consistency and avoid instability by standardising the furnace operating procedures. Since its installation, operator concurrence with the system's advice has been high, and a number of incidents have been spotted which would previously have remained undetected. The system has proven of particular benefit to more recently recruited staff.

Development of the rule base by the process technologist has proven the norm, thus reducing dependence on the ever-rarer software engineer.

The prime measure of product consistency is the variability of hot metal silicon content. The minimum weekly value achieved for this parameter has been reduced by 12 per cent since system implementation, as expected.

Redcar development

During the final commissioning at Llanwern, a decision was made by the Sections, Plates and Commercial Steels (SP&CS) business to install a similar installation at Redcar, the company's largest blast furnace. The main justification for this was to extend campaign life by maximising operational stability.

Here, the system was more complex as it had to interface to a 12 year old PDP and three standalone PCs. Redcar has four tapholes, compared to the two on most other British Steel furnaces. It also has a greater number of sensors. Work started in October 1996 and it was ready for delivery by April 1997.

The database logs more than 1000 parameters at frequencies from 1min to 1 day, as well as event-based data. Following the Llanwern development, British Steel took the opportunity to redesign the knowledge base, including data access. The result has been the creation of a generic shell which can be applied to any blast furnace. Key elements have been retrospectively applied to Llanwern No 3's knowledge base.

Data are read in via another G2-RDB bridge. New data required by the knowledge base must first be stored in an RDB table. Views are used to simplify the G2 coding. To combine the temperature data from the pyrometer and from the thermocouple dip so that G2 only receives new data for both at the same time, a view is used.

Furnace operational data are represented in G2 by a collection of objects, with data being assigned to a particular object. Combining all of the objects together creates a model of the plant. Current values for any attribute can be seen on a table for that particular object which is displayed by clicking next to that object on the furnace mimic. There are currently 145 G2 objects at Redcar which hold 567 data items, or attributes.

All of the rules for the Redcar system were entered by a process technologist using the shell developed by British Steel's software engineers. The process technologist had attended eight days of formal training at Gensym. This also enabled him to contribute to the design of the G2 object hierarchy. Use of the process technologist to develop the rulebase enabled him to include his specialist knowledge, and let the software engineer concentrate on developing a generic system.

The developments have been undertaken using the Technology Centre's accredited software QA system. This has allowed the developers to efficiently design a generic system, control on-going development in both systems and has aided configuration management.

The initial rule base, models and system have functioned reliably since installation at both sites. The management have eagerly claimed ownership.

The system was successfully ported from Llanwern to Redcar. Development was undertaken by British Steel during the transfer to make the system generic and applicable to all of the company's blast furnaces.

The database is frequently being used by technical personnel both on the plant and remotely at the BSTTC.

The modular design and the G2 package have allowed rapid development and implementation of the advisory system.

The article was adapted from a paper presented by Stephen Harvey and Peter Warren of BSTTC at Expert Systems 97.