The research considers conducting orthogonal experiment (OT) as one of the stages in developing a new discrete element method (DEM) parameters calibration approach. The measured responses in experiment are the parameters obtained by DEM animation processing using machine vision system (MVS). The variable factors in experiment are DEM parameters. A brief overview of an existing calibration approaches given in the article. The choice of OT as a design of experiment tool among other mathematical tools discussed. Experiments conducted using specially developed rig where bulk material's flow captured as DEM animation. DEM animation converted to video and then processed using MVS that allow register the values of such parameters as angle of repose or expiration time (measured responses). The results of the OT show that it is possible to identify four measured responses with the most valuable correlation coefficient. DEM parameters with the biggest influence on the measured responses identified for each of the obtained regression. Obtained results are useful in learning or iterative algorithms development for DEM parameters calibration.

The paper presents a methodology for training neural networks for vision tasks on synthe-sized data on the example of steel defect recognition in automated production control systems. The article describes the process of dataset procedural generation of steel slab defects with a symmetrical distribution. The results of training two neural networks Unet and Xception on a generated data grid and testing them on real data are presented. The performance of these neural networks was assessed using real data from the Severstal: Steel Defect Detection set. In both cases, the neural networks showed good results in the classification and segmentation of surface defects of steel workpieces in the image. Dice score on synthetic data reaches 0.62, and accuracy—0.81.

This paper presents a numerical simulation of powder sintering. The numerical model presented in this paper uses the discrete element method, which suggests that the material can be modeled by a large set of discrete elements (particles) of a spherical shape that interact with each other. A methodology has been developed to determine the DEM parameters of bulk materials based on machine vision and a neural network algorithm. The approach is suitable for obtaining the exact values of the DEM parameters of the investigated bulk material by comparing the visual images of the material’s behavior at the experimental stand in reality and in the model. Simulation of sintering requires an introduction of cohesive interaction between particles representing interparticle sintering forces. Numerical sintering studies were supplemented with experimental studies that provided data for calibration and model validation. The experimental results have shown a significant capability of the designed numerical model in modeling sintering processes. Evolution of microstructure and density during sintering have been studied under the laboratory conditions.