Search results

This paper aims to develop a feasible visual weld detection method to solve the problems in multi-layer welding detection (e.g. cover pass welding detection) for seam tracking and non-destructive testing. It seeks for an adaptive and accurate way to determine the edge between the seam and the base metal in the grayscale image of weld automatically. This paper tries to contribute to next-generation real-time robotic welding systems for multi-layer welding.

This paper opted for invariant moments to characterize the seam and the base metal for classification purposes. The properties of invariant moments, such as high degree of self-similarity and separation, affine invariance and repetition invariance, were discussed to verify the adaptability of the invariant moment in weld detection. Then, a weld detection method based on invariant moments was proposed to extract the edge between the seam and the base metal, including image division, invariant moment features extraction, K-Means adaptive thresholding, maximum connected domain detection and edge position extraction.

This paper highlights the significance of high degree of self-similarity and separation, affine invariance and repetition invariance of the invariant moment for weld detection. An adaptive, effective and accurate method is proposed to detect the edge between the seam and the base metal based on invariant moments.

It is necessary to verify the applicability of the proposed method in variable welding conditions further. Further works will focus on the establishment of a real-time seam tracking system during the whole multi-layer/multi-pass welding process based on such adaptive visual features.

This paper includes the implications for development of an adaptive and real-time weld detection method, which is expected to be applied to online seam tracking in multi-layer welding.

This paper presents an accurate weld detection method in multi-layer welding, overcoming difficulties in effectiveness, adaptability and efficiency of existing weld detection methods.