Regular inspections of steam generator tubes (SGTs) to evaluate structure degradation and sludge deposition are essential for maintaining safety and efficiency of power plants. However, with the interference of complex… Click to show full abstract
Regular inspections of steam generator tubes (SGTs) to evaluate structure degradation and sludge deposition are essential for maintaining safety and efficiency of power plants. However, with the interference of complex structures such as tube support plates (TSPs), simultaneously detecting multiple objects is still a challenging problem. This article proposes a novel integrated versatile probe for multiobject inspection. The probe consists of a low-frequency operating region (LFOR) with tunnel magnetoresistance (TMR) array sensors and a high-frequency operating region (HFOR) that uses array coils as the pickup sensors. After frequency optimization and image calibration, the signals of three types of objects are concentrated in three different modalities of the probe’s output. Then, features for position of TSP, profile, and location of sludges are extracted from the image of LFOR. A feature extraction algorithm based on convolution and pattern search is developed to extract defects’ features from the quadrature component image of HFOR. Finally, a hierarchical independent decision-level data fusion scheme is adopted to integrate and visualize the recognition and discrimination of multiple objects in an RGB image. Combinations of machined defects, sludges, and TSP mockup on an SGT are experimentally tested by the probe. Experimental results show that the TSP and sludges are detected accurately. Most of the defects, including the defect embedded under the TSP, can be identified and localized with a high intersection of union (IoU) of 0.92. The TSP, sludges, and defects are clearly visualized in the fused image. For complex scenarios with multiple objects superimposed, the fused image shows each type of objects clearly.
               
Click one of the above tabs to view related content.