Neural Network Based Industrial Processes Monitoring
This industrial processes monitoring based on a neural network presents low run-time, and it useful for critical time tasks with periodic processing. This method allows the time prediction in which a variable will arrive to abnormal or important values. The data of each variable are used to estimate the parameters of a continuous mathematical model. At this moment, four models are used: first-order or second-order in three types (critically damped, overdamped or underdamped). An optimization algorithm is used for estimating the model parameters for a dynamic response to step input function, because this is the most frequent disturbance. Before performing the estimation, the most appropriate model is determined by means of a feed-forward neural network.
KeywordsOrder System Circular Buffer Intelligent Data Analysis Transitory Response Continuous Mathematical Model
Unable to display preview. Download preview PDF.
- 1.Nimmo, I.: Adequately Address Abnormal Operations. Chem. Eng. Progr. 91, 36–45 (1995)Google Scholar
- 2.Juricek, B., Larimore, W., Seborg, D.: Early Detection of Alarm Situations Using Model Predictions. In: Proc. IFAC Workshop on On-Line Fault Detection, Lyon, France (1998)Google Scholar
- 4.Ogata, K.: Modern Control Engineering, 4th edn. Prentice Hall, NY (2001)Google Scholar
- 5.Boyer, S.A.: SCADA: Supervisory Control and Data Acquisition, 3rd edn. Book News, Inc., Portland (2004)Google Scholar
- 6.Södertröm, T., Stoica, P.: System Identification. Prentice-Hall, Englewood Cliffs (1989)Google Scholar
- 7.Edgar, T., Himmelblau, D.: Optimization of Chemical Processes. MacGraw-Hill, NY (1988)Google Scholar
- 9.Oppenheim, A., Schafer, R., Buck, J.: Discrete-Time Signal Processing, 2nd edn. Prentice-Hall Int. Editions, Englewood Cliffs (1999)Google Scholar