E Oznergiz C Ozsoy I Delice and A Kural Jed Goodell September 9 th 2009 Introduction A fast reliable and accurate mathematical model is needed to predict the rolling force torque and exit temperature in the rolling process ID: 673501
Download Presentation The PPT/PDF document "Comparison of empirical and neural netwo..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Comparison of empirical and neural network hot-rolling process models
E
Oznergiz
, C
Ozsoy
I
Delice
, and A
Kural
Jed Goodell
September 9
th
,2009Slide2
Introduction
A fast, reliable, and accurate mathematical model is needed to predict the rolling force, torque and exit temperature in the rolling process.
Function of Paper: To propose an adaptable neural network model for a rolling mill
Why important? Slide3
Neural Network?
An Artificial Neural Network is a computer model designed to simulate the behavior of biological neural networks, as in pattern recognition, language processing, and problem solving, with the goal of self-directed information processing.Slide4
Introduction - References
1
Sims, R. B. The calculation of roll force and torque in
hot rolling mills. Proc.
Instn
Mech.
Engrs
, 1954, 168(6),
191–200.2 Orowan, E. The calculation of roll pressure in hot andcold flat rolling. Proc. Instn. Mech. Engrs, 1943, 150(4),140–167.3 Hitchcock, J. H. Elastic deformation of roll duringcold rolling. Report of Special Research Committee onRoll Neck Bearings, 1935, pp. 33–41 (ASME ResearchPublication, New York).4 Ford, H. and Alexander, J. M. Simplified hot rollingcalculations. J. Inst. Met., 1964, 92, 397–404.5 Barnett, M. R. and Jonas, J. J. Influence of ferrite rollingtemperature on grain size and texture in annealed lowcarbon steels. ISIJ Int., 1997, 37(7), 706–714.6 Kirihata, A., Siciliano, Jr, F., Maccagno, T. M.,and Jonas, J. J. Mathematical modelling of rolling ofmultiply-alloyed mean flow stress during mediumcarbon steels. ISIJ Int., 1998, 38(2), 187–195.7 Kwak, W. J., Kim, Y. H., Park, H. D., Lee, J. H., andHwang, S. M. Fe-based on-line model for the predictionof roll force and roll power in hot strip rolling. ISIJ Int.,2000, 40(10), 1013–1018.8 Sungzoon, C., Cho, Y., and Yoon, S. Reliable roll forceprediction in cold mill using multiple neural networks.IEEE Trans. Neural Netw., 1997, 8, 874–882.9 Hagan, M. T. and Menhaj, M. Training feed forwardnetworks with the Marquardt algorithm. IEEE Trans.Neural Netw., 1994, 5(6), 989–993.10 Lee, D. M. and Lee, Y. Application of neural-networkfor improving accuracy of roll force model in hot-rolling mill. Control Engng Pract., 2002, 10(2), 473–478.
11 Lu, C., Wang, X., Liu, X., Wang, G., Zhao, K., and Yuan, J.Application of ANN in combination with mathematicalmodels in prediction of rolling load of the finishingstands in hsm. In Proceedings of the seventh InternationalConference on Steel Rolling, Chiba, Japan, 1998,206–209.12 Nishino, S., Narazaki, H., Kitamura, A., Morimoto, Y.,and Ohe, K. An adaptive approach to improve theaccuracy of a rolling load prediction model for a platerolling process. ISIJ Int., 2000, 40(12), 1216–1222. 13 Takahashi, R. State of the art in hot rolling processcontrol: review. Control Engng Pract., 2001, 9, 987–993.14 Gorni, A. A. Application of artificial neural networks inthe modeling of plate mill processes. JOM-e, 49(4), April1997, 252–260.15 Poliak, E. I., Shim, M. K., Kim, G. S., and Choo, W. Y.Application of linear regression analysis in accuracyassessment of rolling force calculations. Met. Mater.,1998, 4, 1047–1056.16 Portmann, N. F., Lindhoff, D., Sorgel, G., andGramckow, O. Application of neural networks in rollingmill automation. Iron Steel Engr., 1995, 72(2), 33–36.17 Lee, D. M. and Choi, S. G. Application of on-lineadaptable neural network for the rolling force set-up ofa plate mill. Engng Appl. Artif. Intell., 2004, 17, 557–565.18 Son, J. S., Lee, D. M., Kim, I. S., and Choi, S. G. A studyon on-line learning neural network for prediction forrolling force in hot-rolling mill. J. Mater. Process.Technol., 2005, 164–165, 1612–1617.19 Pichler, R. and Pffaffermayr, M. Neural networks foron-line optimisation of the rolling process. Iron SteelRev., August 1996, 45–56.
20
Duemmler
, A.,
Nitsche
, H. J., and
Sesselmann
, R. Not
much artificial about artificial intelligence – artificial
intelligence in flat product mini steel mills increases
productivity and product quality. Siemens
Newslet
.
Metal., Mining More, 03/1997, 1–6.
21
O¨
zsoy
, C.,
Ruddle
, E. D., and Crawley, A. F. Optimum
scheduling of a hot rolling process by nonlinear
programming. Can. Metall. Q., 1992, 31(3), 217–224.
22
Tarokh
, M. and
Seredynski
, F. Roll force estimation in
plate rolling. J. Iron Steel Inst., 1970, 208, 694.
23
Schultz, R. G. and Smith, A. W. Determination of a
mathematical model for rolling mill control. Iron Steel
Engr., 1965, 80, 127–133.
24
Lopresti
, P. V. and Patton, T. N. An optimal closed
loop control of a rolling mill. In Proceedings of the
Joint Automatic Control Conference, New York 1967,
pp. 767–777.
25
Cybenko
, G. Approximation by superposition of a
sigmoidal
function. Math. Control, Signals Syst., 1989, 2,
492–499.
26
Babuska
, R. Fuzzy modeling for control, 1998 (
Kluwer
, Boston, MA).
27
Arahal
, M. R.,
Berenguel
, M., and Camacho, E. F.
Neural identification applied to predictive control of a
solar plant. Control
Engng
Pract
., 1998, 6, 333–344.
28
Gomm
, J. B., Evans, J. T., and Williams, D. Development and performance of a neural-network predictive controller. Control
Engng
Pract
., 1997, 5(1), 49–59.Slide5
Relevance to Course
The paper shows
an effective
way to compute the needed
rolling force
, torque and temperature needed for hot rollingSlide6
Design Principles
Empirical Model
Lookup tables
Neural
Network
Empirical
vs
NNSlide7
Design parameters
Outputs
:
Rolling force
Torque
Exit TemperatureSlide8
Design principles: Empirical modelSlide9
Design Principles: Neural Network
MISO System– Multi Input Single
Output
Back Propagation Algorithm
To find Force and Torque:
Inputs: Roll radius, number of revolutions, entry slab temperature,
entry and exit thickness.
Output: Force and Torque
To find Exit TemperatureInputs: Energy required, exit thickness, radius, number of revolutions, entry slab temperature, slab width, slab volume. Output: Exit TemperatureSlide10
Machines
Hot rolling mill at
Eregli
Iron and Steel Factory in Turkey.
The equipment:
Slab furnace
Pre-rolling mill
Reversible mill
Seven strip rolling standsCooling systemShearing SystemData Acquisition and Computer control systemSlide11
Experimental Equipment
Dimensions monitored during each pass by an X-ray
Temperature monitored with pyrometer
Roll force and
torque
monitored using four load cells placed along the millSlide12
Empirical ResultsSlide13
Neural Network ResultsSlide14
Results between models
NN model was 22 % better predictor for force, 24% better for torque, and 14 % better for exit temperature
Errors
decreased by 85% for force, 97% for torque, and 92% for temperatureSlide15
Conclusions
Practical use – faster rolling, reduction in energy , more flatness control
Simple learning method
vs
Adaptable NN
Offline
vs
Online – weight update
Industries most impacted – any industry using sheet metal