The Development of a Monitoring and Control System for Pulverised Coal Flames Using Neural Networks

This paper presents the results obtained from a series of experiments that have been conducted on a 150kW pf burner rig based at Casella CRE Ltd. in the United Kingdom. These experiments systematically varied the burner swirl number and the secondary air flow rate over a significant range for two different coals so that both satisfactory and ‘poor’ combustion conditions were obtained. The infra-red emissions from the flame and the combustion noise generated in the furnace chamber were measured with appropriate sensors as were the fuel and air flow rates and pollutant emissions. The signals from the sensors were analysed using signal processing techniques to yield a number of features. These in turn were employed to train a neural network to accurately estimate the gaseous emissions from the rig, such as NOx and CO. In a separate set of experiments, where the combustion process was placed in a poor condition, the sensors were coupled with the neural models and incorporated into an intelligent control system, which was able to alter the excess air level to improve the process. In this fashion simultaneous low Nox and CO levels were achieved with both coal types. This method thus uses a combination of relatively low cost sensors and artificial intelligence techniques to control the combustion of the pulverised fuel burner. It is envisaged as particularly attractive for multiple burner installations that are fed from a common manifold, where individual burner performance is not known.