Abstract:
During the welding process, heat input can cause grain coarsening in the welding area, resulting in uneven gray distribution in the weld seam area. Under interference superposition, it is difficult to effectively enhance the detail expression of the weld seam area to accurately extract the weld seam area, resulting in poor segmentation effect of the weld seam and affecting the accuracy of detecting abnormal welding failures on the heating surface. Therefore, an intelligent detection method for abnormal welding failure images of the heating surface of a 1000MW ultra supercritical boiler is proposed. The Lab space color correction algorithm is used to correct the welding color difference of the heating surface of a 1000MW ultra supercritical boiler, and the grayscale value is normalized to solve the problem of uneven grayscale distribution. Based on this, a dehazing method based on saturation estimation transmittance is applied to enhance image detail information, remove interference effects, and achieve image contrast enhancement. After enhancing the image contrast, in order to accurately extract the weld seam area, the RANSAC algorithm is used for rough localization of the weld seam. At the same time, a curvature threshold adaptive calculation method is used to perform precise segmentation of the weld seam after rough positioning. Introducing Euclidean distance feature data into the regional growth method, and fusing it with curvature and normal vectors to construct a comprehensive feature vector for growth judgment. Finally, comparing the growth results with critical dimensions, the detection of abnormal welding failure images on the heating surface of a 1000MW ultra supercritical boiler is completed. The experimental results show that the proposed method can accurately and effectively detect welding abnormal failures, effectively ensuring the normal operation of the boiler.