Designing of ultrasonic reactor using machine learning

Abstract

Ultrasonic reactors consist of transducers that emit high-frequency acoustic waves, with designs varying based on their intended application. One such application is extraction, achieved through the cavitation effect generated by the pressure gradients produced by the ultrasound field within the medium. The cavitation effect depends on several physical parameters of the ultrasonic reactor, such as the driving frequency, acoustic energy, liquid medium, transducer attachment locations, and size and shape of the reactor. The proper design of an ultrasonic reactor is complex due to the numerous parameters that affect the distribution of ultrasonic energy, which must be concentrated and distributed in a specific way to extract effectively. To simplify the design process, this study utilizes an artificial neural network (ANN) to predict the physical dimensions and transducer locations of a square duct-shaped ultrasonic reactor with transducers attached to its sides based on the desired spatial energy distribution. The algorithm was developed using data from 1365 reactor configurations, which were numerically generated using COMSOL Multiphysics. The study tested two main models of ANNs, one using intensity profile as input and the other using power profile as input, and the results showed that each ANN had its advantages.