Ultrasonic Metal Welding (USMW) is a type of friction welding which offers, among its other advantages, a high potential for automatization. USMW can be used with a variety of metals, including copper, aluminum and gold, and the workpieces can have a range of geometries. Although it is often used in the industry, it still suffers from fluctuations in the strength of the welds produced, due to difficulties in finding the best welding parameters. Furthermore, the choice of these parameters is based on the experience of the machine operator and on previous trial welds, leading to a potentially time-consuming process to find appropriate parameters, one of which is the duration of welding. For these reasons, USMW would be a good candidate for optimisation with machine learning.Based on measurements of the vibrations of the horn and anvil and of the airborne sound during welding, this work investigates the use of machine learning in Ultrasonic Metal Welding to find a good stop time to the welding process. The workpieces were copper sheets, and the only varying parameter was the welding time. Based on the measurement data, an appropriate welding parameter is chosen, and several machine learning algorithms are investigated.