There is a growing demand and desire for reliable, efficient, and numerical models to reflect complex interactions between particles so important in wide rage of industrial applications. For the moment, some of the most frequently used numerical models are already available Hybrid Euler-Lagrange and Discrete Element Method approaches for modelling multiphase flows, for instance fluidization process. Within the HEL, the interactions between the particles are calculated on the basis of the Kinetic Theory of Granular Flow, where collisions are determined from solid volume fraction taken from computational cell. Such approach, in spite of the speed of the results obtained and their relative accuracy, is associated with problems in the determination of the closure terms (i.e. granular temperature), which is used to determine the force resulting from the collision of particles in the numerical cell, which may cause also instabilities in the calculations. Therefore, there is empty spacy that can be filed by the combination of the advantages of both models to create a new approach, called Hybrid-Euler Lagrange Surrogate Collision Model, by applying surrogate model for prediction particles interactions. For that purpose a machine learning algorithms were utilized, mainly to developed new idea that can be used for particle collision tracking. The constantly increasing popularity of machine learning technique, ubiquitous availability, speed of operation as well as high accuracy of the obtained results (provided that, having adequate data) makes it currently the best option to use in order to build a new approach. The geometry used for the validation of simulations corresponds to the geometry of the experimental test-rig. The test-rig consists of transparent main core, with two side inlets attachedrnded up with paricle containers. To transport the particle, air is supplied to the container and the main air path. In front of the experimental stand, the high-speed camera is situated to record the images from measurements and observation of the particles streams behavior. The collected measurement images were used to validate the simulation results. To compare the results, 10 cross sections along the length of the main channel were defined, where the particle velocity profiles, calculated by time averaging were determined. For determining the velocity of particles in experiment, a special in-house algorithm was developed in lab-view combined with the python script.