Future of Ultrasonic Transducers: How Machine Learning Is Driving Innovation

Excellent capabilities for low-frequency devices are exhibited by magnetostrictive materials such as terfenol. However, their utilization in high-frequency ultrasonic transducers requires further advancements. In this study, a novel approach is introduced, utilizing experimental data to establish the relationship between the output amplitude of a magnetostrictive transducer and various design parameters. These parameters include frequency, current, core lamination thickness, core length, core diameter, and lengths of the primary and secondary horn steps. Several machine learning methods, including the radial basis function (RBF) neural network, support vector machine (SVM), multilayer perceptron neural network (MLP), and Gaussian process regression (GPR), were employed for analyzing the experimental data. The analysis revealed that the RBF model demonstrated the best predictive performance with an RMSE of 0.12. Through sensitivity analysis, the study identified frequency, current, the length of the secondary horn step, core lamination thickness, core length, core diameter, and length of the primary horn step as the most influential design parameters for optimizing the output amplitude of magnetostrictive ultrasonic transducers with a terfenol core. This study proposes the utilization of machine learning in the optimization of the magnetostrictive ultrasonic transducer design. It presents a novel method that integrates experimental data and machine learning techniques for design optimization. The findings emphasize the potential of machine learning in enhancing the efficiency and reliability of transducers for various applications.