Ultrasonic metal welding (USMW) has received significant attention in recent years. It is particularly suitable for connecting electrotechnical components and has been widely used in various industrial applications. However, the bond quality in USMW fluctuates due to a large number of influencing variables which may affect the joint strength. The welding process should thus be analysed in more detail by measuring the tool vibrations that occur during welding with laser vibrometers and recording the emitted airborne noise with a microphone.
This thesis aims to develop a classification algorithm in machine learning, which can classify the bond quality into three strength categories labelled as “Underweld”, “Basicweld” and “Overweld” based on the measured signals. The normalised wavelet packet energy (NWPE) features of these signals are extracted based on the wavelet packet transform (WPT) method. Backpropagation neural network (BPNN), support vector machine (SVM) and linear discriminant analysis (LDA) models are developed and optimised for pattern recognition.
The classification results of all three models in this study present an acceptable recognition performance. The use of NWPE features has thus the potential to be implemented into process control for welding systems. Such a new approach provides insights into the monitoring of joint quality during USMW process.

Melden Sie sich Hier an um die Einladung zu dem Vortrag per E-Mail zu erhalten.
Register here to receive the invitation to this talk via E-Mail.

Zoom-Meeting-ID: 917 0358 9106
Passwort: 058632