Year: 2026
Type: dissertation
Abstract: The keyhole stability is a significant challenge in keyhole laser welding. An unstable keyhole is the main contributor to welding defects, such as pores and spatter. The welding defects compromise the weld integrity, leading to unreliable joints. Traditional non-destructive testing (NDT) methods can be used to detect welding defects, but require extra work. Welding process monitoring could possibly eliminate the need for NDT by directly monitoring the keyhole. This thesis proposes a novel method of monitoring keyhole stability: the keyhole contours are extracted from coaxial footage, and the standard deviation is calculated from the data to represent the magnitude of change in the shape of the keyhole. A convolutional neural network (CNN) with the U-net architecture was trained to semantically segment the keyhole from the welding process images. The effect of power on the standard deviation was studied; using this method, it was found that the keyhole is more stable when welding near the optimal parameters.
Source: UTUPub (University of Turku)
Author: Ranto, Olli
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7220820854
