Prediction of structural dynamics of the Electric Arc Furnace (EAF) is rather difficult, because of a number of phenomena occurring during the scrap melting process. Three large electrodes, corresponding to each phase of a AC circuit, are lowered by the main mast towards the scrap to activate the melting process, induced by the electric arc. Electric current fed to each electrode produces a strong magnetic field and applies an electromechanical force on the other electrodes. Arc voltage looks irregular upon time, even because of the scrap motion within the vessel and temperature growth. The vertical position of the mast is controlled by an hydraulic actuator. Nevertheless, a heavy vibration of the structures affects the regularity of the melting process. A fully coupled model of the whole system is herein proposed, through a multi-physics approach. A first analytical approach, describing the electric circuit of the whole system, is implemented into a Multi Body Dynamics (MBD) model of the EAF, while a reduced Finite Element Method (FEM) model of the flexible structures is used for a preliminary optimization of the design parameters. Electromechanical forces due to the mutual induction among the electrodes are computed and the dynamic response of the system is investigated. Proposed model allows a first refinement of the EAF design, although a complete experimental validation on the real machine has to be performed, in spite of problems due the extremely difficult accessibility of structures during the melting process.

The objective of the present study is to develop an Artificial Neural Network (ANN) in order to predict the bevel angle (response variable) during CNC plasma-arc cutting of St37 mild steel plates. The four (4) input parameters (plate thickness, cutting speed, arc ampere, and torch standoff distance) of the ANN was selected following the results (relative importance) of the Analysis Of Variance (ANOVA) performed based on seven (7) factors (plate thickness, cutting speed, arc ampere, arc voltage, air pressure, pierce height, and torch standoff distance) in a previous study. A multi-parameter optimization was carried out using the robust design. An L 18 (2 1 × 3 7 ) Taguchi orthogonal array experiment was conducted and the right bevel angle was measured, aiming at the investigation of the influence of plasma-arc cut process parameters on right side bevel angle of St37 mild steel cut surface. The selection of quality characteristics, material, plate thickness and other process parameter levels and experimental limits was based on the experience and current needs of the Greek machining industry. A feed-forward backpropagation (FFBP) neural network was fitted on the experimental data. The results show that accurate predictions of the bevel angle can be achieved inside the experimental region, through the trained FFBP-ANN. The developed ANN model could be further used for the optimization of the cutting parameters during CNC plasma-arc cutting.