Abschlussarbeit

Neural Network classification of signals for welding optimisation

Ultrasonic Metal Welding (USMW) is particularly suitable for connecting electrotechnical components. The concept is simple: two metal plates are pressed and rubbed against each other at 20 kHz. Due to friction, the plates are welded together. However, the quality of the weld fluctuates between each iteration. During experiments, vibrations of different parts of the welding machine are measured using laser vibrometers, and the airborne sound is measured using a microphone, up to 100 kHz. The goal of this project is to design a neural network classification algorithm that would classify the welds as good or bad welds based on the measured signals. This includes choosing acceptable inputs, trying different combinations, and testing different architectures of neural networks.