A visual weld edge recognition method based on light and shadow feature construction using directional lighting

For the efficient and effective detection in automatic non-destructive testing (NDT) of weld defects, the NDT probe must stay aligned with the welds. It is essential to detect the deviation between the seam and the probe by means of weld recognition, in order to adjust the relative position between the workpiece and the probe. The traditional weld recognition method is suitable for real-time seam tracking when the geometric feature of the groove is distinct; however, it may fail when applied to the formed weld recognition. Other methods proposed by researchers are unsatisfactory in the signal-to-noise ratio of imaging, separation degree of visual features and efficiency of the processing algorithm. To solve these problems, this paper proposes a novel visual weld recognition method using two directional lights. The directional lights are projected onto the edges of the seam to produce the distinct man-made “light and shadow” (LS) features. In this paper, we first study the principles and sensing requirements of the LS features; then a stroboscopic circuit is established to lighten the two directional lights alternatively and meanwhile synchronizes the camera to capture images when each light is on. Afterwards, we propose an image processing algorithm based on thresholding and edge extraction to calculate the accurate edges of the seam. Finally, a weld detection platform is established and experiments on weld recognition and deviation correction are conducted. The experiment results show that the proposed detection method can recognize the edges of the seam effectively in 0.023 mm accuracy. The time cost of the image processing algorithm is not more than 13 ms. The proposed method is suitable for weld detection where the reinforcements of the seams are as low as 1 mm. It expected to be applied to the automatic vision-guided NDT of weld defects.