Prediction rotational speed of DC external gear pump caused by hydraulic circuit using feed forward neural network

This study examines the performance prediction of an external gear pump (EGP) under varying hydraulic resistance (Rh) in commercial home applications. Accurate flow rate control is essential for devices such as automatic ice makers, where precise water dispensing is required. Traditional models assume constant pump speed, but in practice, speed varies with Rh, influencing flow rate and pressure. This study integrates experiments and simulations to analyze the relationship between Rh, flow rate, and pump speed. Experimental data were obtained from a pump module operating at constant voltage, while simulations visualized flow fields and pressure distributions. Without accounting for speed reduction, simulation errors reached 48%. To improve accuracy, a feedforward neural network (FFNN) was trained using experimental data to predict Rh and pump performance under various conditions. Incorporating speed variation due to Rh into simulations significantly enhances flow rate predictions, with the FFNN achieving an error margin below 2%. This approach facilitates the development of derivative pump models and design modifications for existing systems.