Real-time image processing of keyhole puddle in variable polarity plasma arc welding

In a previous study, a simple and practical image-sensing system comprising a narrow band filter and neutral filters was developed to acquire front-side keyhole puddle images in variable polarity plasma are welding (VPPAW) of aluminium alloys. However, during actual welding, oxides existing on the puddle surface and variations in image quality with respect to keyhole puddle size complicate the processing of images. In this paper, an image-processing method based on an alpha-cutting-average-median filter, minimum filter, direction enhancement operators and local statistical characteristics enhancement is developed to tackle the various shapes and appearances of the keyhole puddle. Also, an effective edge-tracing method is developed to detect the edge of the visible keyhole. The geometric sizes of the visible keyhole, such as area, width and height, are acquired with sufficient accuracy, which can be used to calculate the keyhole diameter during the welding process. Experiments show that satisfying results can always be obtained with the developed image-processing method regardless of the original image quality. Processing one image frame takes less than 220 ms.