Degree Name

Master of Computer Science (Hons)


School of Information Technology and Computer Science - Faculty of Informatics


Lack of fusion is a commonly occurring defect in Pulsed Transfer Gas Metal Arc Welding. Under optimum process parameter settings, such defects are mainly caused by positional error of the torch in relation to the joint profile. This project investigated relationships between seam tracking error and welding process parameters in order to detect lack of fusion. Using digital signal processing techniques, it was found that there existed consistent linear relationships between torch standoff and process variables in both the time and frequency domains. In the time domain, when the torch standoff increases, the mean values of both current power decrease, conversely the mean values of voltage and resistance increase. Whilst in the frequency-domain, it was also discovered that the pulse frequency of power supply varies during welding in accordance to changes in the torch standoff linearly. This curious finding was confirmed by the manufacturer as a result of the characteristics of the specific power supply used. Additionally, when torch oscillation is considered, it was established that torch offset produces distinct patterns in current signal waveforms which is affected by welding setup, such as the joint profile and the oscillation frequency and amplitude of the torch. For girth welding of symmetric joints, the following classification rule was derived in order to successfully monitor weld quality (lack of fusion): The signature of a good weld has twice the weave frequency. The signature of a poor weld has the same frequency as the weave frequency. Due to its independent property of the power supply, this rule enables us to monitor weld quality (detecting lack of fusion) efficiently and robustly for any power supply type. Finally, several potential solutions for flexible on-line weld quality monitoring were also proposed.