An intelligent monitoring system for the detection of slag deposition on a pulverized coal fired burner

Abstract

The objective of this paper is to describe the further development of a monitoring system to detect the presence of so-called burner eyebrows, i.e. relatively large deposits of slag around the burner quarl in pulverized coal fired boilers. Experiments were undertaken with a range of coals and with various artificial eyebrows constructed from cast refractory inserts. The system uses a microphone to detect combustion noise and an infrared sensor which measures flame radiation, and the signals from these cheap, easily installed sensors were analyzed by a hybrid neural network. In tests with two coals, the system was able to distinguish the different eyebrows with a high degree of accuracy if representative data were used to train the network for each particular coal. In further tests with a range of six different coals, the system was able to distinguish between a clean burner and one fitted with a particular sized eyebrow. In this case, it proved to be possible to use only the features from three of the coals in the training process and the data from the remaining fuels for validation. The monitoring system, therefore, appears to be relatively independent of changes to the coal fired by the burner if trained with a representative range of coals. Finally, this paper presents a possible method to detect burner eyebrows via the evaluation of so-called ‘eyebrow indices’ using a self-organizing map which is trained solely using clean burner sensor patterns.

Development of an intelligent flame monitoring system for steel reheating burners

“…The topological neighbours of the BMU are updated in a similar manner so that they are also moved closer to the input pattern. More background information on this widely used model can also be found in Kohonen, 14 Vesanto et al 15 and Tan et al 16 The SOM implemented consisted of a grid of 12 Â 12 neurons with the inputs to the SOM being the same RMS values of the three photodiode signals gathered from the full-scale industrial furnaces at Tata and Arcelor-Mittal. The data was divided so that half was used for training data and the remainder to test the network performance.…”

“…The topological neighbours of the BMU are updated in a similar manner so that they are also moved closer to the input pattern. More background information on this widely used model can also be found in Kohonen, 14 Vesanto et al 15 and Tan et al 16…”