Mlazne pumpe, odnosno ejektori predstavljaju vrstu pumpi koje rade na principu Venturijevog učinka. Eduktori su vrsta mlaznih pumpi koji kao primarni i sekundarni fluid koriste kapljevine ili mješavinu kapljevine i krutih čestica. Budući da eduktori imaju široku primjenu, njihova iskoristivost posebno je važna u velikim postrojenjima koja rade bez prekida. U ovom radu proveden je proces optimizacije eduktora, odnosno traženja geometrije eduktora koja daje maksimalnu iskoristivost uz konstantne uvjete rada primjenom računalne mehanike fluida, umjetne neuronske mreže te genetskog algoritma. Važno je napomenuti da su odabrana tri geometrijska parametra eduktora dok se ostale veličine ne mijenjaju. Varijacijom polumjera komore miješanja, duljine komore miješanja te duljine konvergentnog dijela u definiranom rasponu odabrano je 80 kombinacija za koje se provode numeričke simulacije pomoću programskog paketa Ansys/Fluent. Na temelju numeričkih rezultata izračunate su vrijednosti iskoristivosti eduktora. Numerički određena ovisnost iskoristivosti o geometrijskim parametrima za konačan broj kombinacija parametara korištena je za treniranje umjetne neuronske mreže koja je generirana koristeći programski jezik Python, odnosno Anaconda distribuciju Python-a i Jupyter Notebook kao vizualni alat i uređivač koda. Uz pomoć DEAP genetskog algoritma pronađena je optimalna geometrija eduktora koristeći umjetnu neuronsku mrežu kao zamjenski model koji definira ovisnost iskoristivosti o geometrijskim parametrima eduktora. Na kraju je provedena analiza osjetljivosti optimizirane geometrije eduktora kako bi se vidio utjecaj masenog protoka primarnog fluida na iskoristivost i stupanj ejekcije eduktora. Jet pumps, also known as ejectors, are a type of pump that works on the principle of the Venturi effect. Eductors are a type of jet pump that uses liquids or a mixture of liquid and solid particles as the primary and secondary fluids. Since eductors have a wide range of applications, their efficiency is especially important in large plants that operate non-stop. In this master's thesis, the eductor optimization process was performed, searching for the eductor geometry that gives maximum efficiency under constant operating conditions using computational fluid dynamics, an artificial neural network and a genetic algorithm. It is important to note that three geometric parameters of the eductor were selected, while the other dimensions were not changed. By varying the radius of the mixing chamber, the length of the mixing chamber and the length of the convergent part in a defined range, 80 combinations were selected for which numerical simulations were performed using Ansys/Fluent software package. Based on the numerical results, the eductor efficiency values were calculated. The numerically determined dependence of efficiency on geometric parameters for a finite number of parameter combinations was used to train an artificial neural network that was generated using Python programming language, Anaconda as a Python distribution and Jupyter Notebook as a visual tool and code editor. With the support of the genetic algorithm implemented in DEAP, the optimal geometry of the eductor was found using an artificial neural network as a substitute model, which defines the dependence of efficiency on the geometric parameters of the eductor. At the end, a sensitivity analysis of the optimized geometry of the eductor was performed in order to see the influence of the mass flow rate of the primary fluid on the efficiency and entrainment ratio of the eductor.
