Optimized PID Controller for Atomizer Geometry: Artificial Intelligence Versus Traditional Methods

Optimized PID Controller for Atomizer Geometry: Artificial Intelligence Versus Traditional Methods

3MT - Three Minute Thesis

This research compares two types of proportional-integral-derivative (PID) tuning methods, traditional and machine learning, to enable atomizer geometry modulation for effective breakup of non-Newtonian slurry. Non-Newtonian slurries are not easily disintegrated due to their nature of having a high viscosity property that keeps changing during atomization. During effective atomization, consistent droplets are produced, and these are useful in many applications such as rocket propulsion where the breakup of gel propellant leads to better combustion that causes efficient propulsion. A baseline PID controller aims to enable the adequate adjustment of the atomizer nozzle diameter to compensate for steam pressure drop during atomization, leading to the production of consistent droplets. Steam is the heating gas that facilitates slurry breakup. We tested the baseline PID controller with transonic airblast nozzle model in Computational Fluid Dynamics (CFD) to compare the proposed machine learning algorithm with other PID tuning methods. The fuzzy logic, artificial neural network (ANN) and Zeigler Nichols (ZN) tuning methods were examined to determine the best method for the transonic airblast nozzle model. Results showed that the ANN method optimized the PID controller best. Identifying the PID performance with the ANN method is useful for industrial applications where PID controllers are greatly used.