Finite element analysis coupled artificial neural network approach to design the longitudinal-torsional mode ultrasonic welding horn

Abstract

The longitudinal-torsional mode horn has gained popularity recently for ultrasonic welding (USW) because it is more efficient than a conventional longitudinal horn. Addition of slanting grooves to the front mass to achieve torsionality is not a novel approach, and few works have already addressed this issue, but comparative studies about the effect of different groove parameters such as length, depth, angle, width, and distance upon the torsionality and resonance frequency are very rare. In the present work, only one parameter was varied at a time while others were kept constant to see their effect on horn attributes, i.e., torsionality and resonance frequency. The torsionality was maximized while keeping the value of resonance frequency as close to working frequency (i.e., 20 kHz) as possible. Depth is considered to be the most important parameter since its effect on torsionality was higher than the other four parameters. Multi-layer perceptron neural network was trained using the input features (i.e., groove parameters) which has the potential of transfer learning and would ease the process of finding the optimum parameters for torsionality maximization in later projects.