International Scientific Journals
of Scientific Technical Union of Mechanical Engineering "Industry 4.0"

The electron beam technological processes like electron beam welding, electron beam additive technologies, etc. depend strongly on the characteristics of the electron beam, generated by the electron gun. In this work the estimation of the 3D radial current density distribution using training, testing and validation of different artificial neural networks is considered. The model estimation is based on experimental measurements of the electron beam current distribution in three cross-sections of the beam at different distances from the magnetic lens of the electron gun. The estimated neural models with different structures are compared. Graphical user interface for the evaluation of the radial electron beam distribution in any cross-sections of the beam is developed.

This paper considers the process electron beam welding of stainless steel type 1H18NT in vacuum. Based on experimental data, the influence of the variations of the following process parameters: electron beam power, welding velocity, the distances from the magnetic lens of the electron gun to the beam focus and to the surface of the treated sample is investigated.

Neural and regression models for the geometry characteristics of the welded joints: surface of the weld cross-sections, weld depths and mean weld widths of the samples are estimated, as well as models for defining the areas of the process parameters, where the appearance of defects is or is not expected. The obtained models are used for developing the graphical user interface aiming investigation and prediction of the electron beam welding characteristics and process parameter optimization. This software can be implemented for supporting the operator’s choice of appropriate work regimes, obtaining the required welds quality standards, for education and investigations.

This paper considers the process electron beam welding in vacuum of stainless steel 1H18NT. Neural network based models are developed and used for the description of the defectiveness, depending on the process parameters – electron beam power, welding velocity, the distance between the main surface of the magnetic lens of the electron gun and the beam focusing plane and the distance between the main surface of the magnetic lens of the electron gun and the sample surface. Neural network (NN) models, based on a multi-layered feedforward neural network, trained with Levenberg-Marquardt error backpropagation algorithm are compared with NN models, based on Pattern recording neural network, trained with Conjugate Gradient Algorithm. The neural networks are trained, verified and tested using a set of experimental data. The obtained models are implemented to predict areas of process parameters, where the appearance of defects is most probable and the location of welding regimes that should be avoided.