Mold breakout prediction in slab continuous casting based on combined method of GA-BP neural network and logic rules

• Published in: International journal of advanced manufacturing technology Vol. 95; no. 9-12; pp. 4081 - 4089
• Main Authors: He, Fei, Zhang, Lingying
• Format: Journal Article
• Language: English
• Published: London Springer London 01.04.2018; Springer Nature B.V
• Subjects: Alarm systems; Back propagation; Back propagation algorithms; Back propagation networks; CAE) and Design; Change detection; Computer simulation; Computer-Aided Engineering (CAD; Continuous casting; Economic models; Engineering; False alarms; Field tests; Genetic algorithms; Industrial and Production Engineering; Logic; Mechanical Engineering; Media Management; Molds; Neural networks; Original Article; Pattern recognition; Simulation; Slab casting; Test procedures; Thermocouples; Time series; Two dimensional models; Waveforms; Continuous casting; Breakout prediction; Neural network; Genetic algorithm; Mold; Logic rule
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-017-1517-1

Given advantages of artificial intelligent technology, the genetic algorithm (GA) and back propagation (BP) neural network is used to construct time series model for recognizing temperature change waveform of single thermocouple in mold breakout process. Based on breakout mechanism, the logic rules are used to construct spatial model of multi-thermocouples for identifying two-dimensional (2D) propagation behavior of the sticker. Time series model based on GA-BP neural network and spatial model based on logic rules form a new breakout prediction method. And the simulation and field test of this method are carried out for validating its performance. Simulation results of time series model show that the GA-BP neural network has better recognition precision than BP neural network for sticking temperature pattern of single thermocouple. And simulation results of spatial model show that it can predict all stickers accurately and timely, with no missed alarm and false alarm. Furthermore, field test results show that this breakout prediction method has detection ratio of 100% and a lower false alarm frequency (0.1365% times/heat), which is better than actual breakout prediction system used in continuous casting production. So the combined method of GA-BP neural network and logic rules is feasible and effective in breakout prediction and can be used in more intelligently industrial process.