Neural Network Predictions of Slagging and Fouling in Pulverized Coal-Fired Utility Boilers
Feed-forward back-propagation neural networks were trained to relate the occurrence and characteristics of troublesome slagging and fouling deposits in utility boilers to coal properties, boiler design features, and boiler operating conditions. The data used in this effort were from a survey of utility boilers conducted by Battelle Columbus Laboratories in an Electric Power Research Institute project. This data base is the largest known collection of public information on slagging and fouling in utility boilers. Although the data base was large, the actual number of data sets with adequate information to use as inputs into the networks was relatively small. This situation limits the usefulness of the algorithms developed in this study. However, the data were adequate to demonstrate that neural networks can be used to reliably predict slagging and fouling in utility boilers using simple, readily available input parameters. This suggests that others can use a similar approach to develop either a general tool that is applicable to all boilers or a specialized tool that is applicable to a limited set of boilers, such as those of one utility or boiler manufacturer.
Two networks were developed in this study, one for slagging and one for fouling, to predict ash deposition in various types of boilers (wall-, opposed wall-, tangentially, and cyclone-fired) that fire bituminous and sub-bituminous coals. Both networks predicted the frequency of deposition problems, physical nature (or state) of the deposit, and the thickness of the deposit. Since deposit characteristics vary with boiler location and operating conditions, the worst documented cases of ash deposition were used to train the neural networks. Comparison of actual and predicted deposition showed very good agreement in general. The relative importance of some of the input variables on the predicted deposit characteristics were assessed in a sensitivity analysis. Also, the slagging and fouling characteristics of a blend of two coals with significant different deposition characteristics were predicted to demonstrate a practical application of developed neural networks.
KeywordsSteam Flow Deposit Thickness Electric Power Research Institute Coal Property Utility Boiler
Unable to display preview. Download preview PDF.
- Bryers, R. and Vorres, K. (Ed), Proceedings of the Engineering Foundation Conference on Mineral Matter and Ash in Coal, Hemisphere Publishing, New York, NY, 1988.Google Scholar
- EPRI, Final Report No. CS-5523, “Slagging and Fouling in Pulverized Coal-Fired Utility Boilers,” 1987a.Google Scholar
- EPRI, Research Project 2256, Proceedings of the EPRI Conference on Effects of Coal Quality on Power Plants, Atlanta, GA, 1987b.Google Scholar
- Hampshire, J.B., and Waibel, A., “Connectionist Architectures for Multi-speaker Phoneme Recognition,” Advances in Neural Information Processing Systems 2, 1990.Google Scholar
- Hoskins, J.C., Kaliyur, K.M., and Himmelblau, D.M., “Incipient Fault Detection and Diagnosis Using Neural Networks,” Proceedings of the International Joint Conference on Neural Networks, 1990.Google Scholar
- Hough, D.C, D.C., “ASME Ash Fusion Research Project,” prepared for the American Society of Mechanical Engineers Research Committee on Corrosion and Deposits from Combustion Gases, Babcock Power Fuel Consultancy Report No. FT/88/01, 1988.Google Scholar
- Marko, K.A., Feldkamp, L.A., and Puskorius, G.V., “Automotive Diagnostics Using Trainable Classifiers: Statistical Testing and Paradigm Selection,” Proceedings of the International Joint Conference on Neural Networks, 1990.Google Scholar
- Naser, J, J., “Exploring Neural Network Technology,” EPRI Journal, December, 1992.Google Scholar
- Niebur, D, D., “Artificial Neural Networks for Power Systems - A Literature Survey,” Proceedings of the EPRI Conference on Expert System Applications for the Electric Power Industry, Phoenix, AZ, 1993.Google Scholar
- Parker, D.B., Optimal Algorithms for Adaptive Networks: Second Order Back Propagation, Second Order Direct Propagation, Second Order Hebbian Learning,“ Proceedings of IEEE First International Conference on Neural Networks, 1987.Google Scholar
- Raask, E., Mineral Impurities in Coal Combustion Hemisphere Publishing Corporation, New York, NY, 1985.Google Scholar
- Rumelhart, D.E., Hinton, G., and Williams, R.J, R.J., “Learning Internal Representations by Error Propagation,” Parallel Distributed Processing, Vol. 1, MIT Press, 1986.Google Scholar
- Smouse, S.M., Wildman, D.J., McIlvried, T.S., and Harding, N.S., “Estimation of NO), Emissions from Pulverized Coal-Fired Utility Boilers,” EPRI/EPA Joint Symposium on Stationary Combustion NO), Control, Bal Harbour, FL, 1993.Google Scholar
- Troudet, T., and Werrill, W., “A Real Time Neural Net. Estimator of Fatigue Life,” Proceedings of the International Joint Conference on Neural Networks, 1990.Google Scholar
- Werbos, P.J., “Beyond Regression: New Tools for Prediction and Analysis in Behavior Sciences,” PhD Thesis, Harvard University, Cambridge MA, 1974.Google Scholar
- Weingartner, E.C., “Coal Fouling and Slagging Parameters,” ASME Research Committee on Corrosion and Deposits from Combustion Gases, 1974.Google Scholar
- Wildman, D.J. and Smouse, S.M, S.M., “Estimation of NQ Emissions from Pulverized Coal-Fired Utility Boilers,” EPRI/EPA Joint Symposium on Stationary Combustion NOX Control, Kansas City, MO, 1995.Google Scholar
- Wildman, D.J., Ekmann, J.M., and Smouse, S.M., “Prediction of Pilot-Scale Coal Ash Deposition: Comparison of Neural Network and Multiple Linear Regression Techniques,” Engineering Foundation Conference on the Impact of Ash Deposition on Coal-Fired Plants, Solihull, England, 1993.Google Scholar