TY - JOUR

T1 - Thermal and Thermomechanical Responses Prediction of a Steel Ladle Using a Back-Propagation Artificial Neural Network Combining Multiple Orthogonal Arrays

AU - Hou, Aidong

AU - Jin, Shengli

AU - Harmuth, Harald

AU - Gruber, Dietmar

PY - 2019/4/29

Y1 - 2019/4/29

N2 - To facilitate industrial vessel lining design for various material properties andlining configurations, a method, being composed of the back-propagationartificial neural network (BP-ANN) with multiple orthogonal arrays, isdeveloped, and a steel ladle from secondary steel metallurgy is chosen for acase study. Ten geometrical and material property variations of this steelladle lining are selected as inputs for the BP-ANN model. A total of 160lining configurations nearly evenly distributed within the ten variations spaceare designed for finite element (FE) simulations in terms of five orthogonalarrays. Leave-One-Out cross validation within various combinations oforthogonal arrays determines 7 nodes in the hidden layer, a minimum ratioof 16 between dataset size and number of input nodes, and a Bayesianregularization training algorithm as the optimal definitions for the BP-ANNmodel. The thermal and thermomechanical responses of two optimal liningconcepts from a previous study using the Taguchi method are predicted withacceptable accuracy.

AB - To facilitate industrial vessel lining design for various material properties andlining configurations, a method, being composed of the back-propagationartificial neural network (BP-ANN) with multiple orthogonal arrays, isdeveloped, and a steel ladle from secondary steel metallurgy is chosen for acase study. Ten geometrical and material property variations of this steelladle lining are selected as inputs for the BP-ANN model. A total of 160lining configurations nearly evenly distributed within the ten variations spaceare designed for finite element (FE) simulations in terms of five orthogonalarrays. Leave-One-Out cross validation within various combinations oforthogonal arrays determines 7 nodes in the hidden layer, a minimum ratioof 16 between dataset size and number of input nodes, and a Bayesianregularization training algorithm as the optimal definitions for the BP-ANNmodel. The thermal and thermomechanical responses of two optimal liningconcepts from a previous study using the Taguchi method are predicted withacceptable accuracy.

UR - http://www.scopus.com/inward/record.url?scp=85065156004&partnerID=8YFLogxK

U2 - 10.1002/srin.201900116

DO - 10.1002/srin.201900116

M3 - Article

VL - 2019

JO - Steel research international

JF - Steel research international

SN - 0177-4832

M1 - 1900116

ER -