A weld line detection robot based on structure light for automatic NDT
- PDF / 12,178,440 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 47 Downloads / 133 Views
ORIGINAL ARTICLE
A weld line detection robot based on structure light for automatic NDT Zhaoxuan Dong 1 & Zhiheng Mai 1 & Shiqi Yin 1 & Jie Wang 1 & Jie Yuan 1 & Yuenong Fei 1 Received: 25 October 2019 / Accepted: 17 August 2020 # Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract In automatic non-destructive testing (NDT), weld bead tracking is usually performed outside. However, irregular weld boundaries, unconstrained illumination, and rough metal surfaces can cause noise, which increases the difficulty of seam tracking. In this paper, a method of parallel structured light (PSL) sensing based on deep learning and information fusion is proposed to detect weld lines. First, a camera is used to capture the laser stripe image projected by the PSL on the weld bead. Then, a MobileNetSSD deep learning model is trained to extract the regions of interest (ROIs) to de-noise the laser stripe image. Finally, the weld line is obtained by fusing information from multiple weld boundaries. Keywords Structured light sensor . Weld line detection . MobileNet-SSD
1 Introduction Non-destructive testing (NDT) of weld seams is important in guaranteeing the safe operation of industrial equipment including oil pipelines, storage tanks, and wind turbine towers [1]. In recent years, the rapid development of robot and imageprocessing technology has inspired widespread study of automatic NDT systems. Such systems automatically track weld lines to complete testing, thereby significantly improving the quality and efficiency of NDT. Unlike indoor welding, automatic NDT is usually performed in the field, with moving
* Yuenong Fei [email protected] Zhaoxuan Dong [email protected] Zhiheng Mai [email protected] Shiqi Yin [email protected] Jie Wang [email protected] Jie Yuan [email protected] 1
Shenzhen University, Shenzhen, China
platforms, unplanned routes, and unconstrained illumination. Therefore, the biggest challenge in developing automatic NDT systems is the quick and accurate detection and tracking of weld lines under multiple interfering conditions. Concerning weld line navigation, some approaches have been proposed based on the weld visual features, artificial tracks, and structured light sensors. Visual feature–based methods often use grayscale gradients and texture features of the weld to detect weld boundaries. For example, Du et al. found that Haralick texture features, such as “energy” and “entropy,” between the base material and the weld bead differ significantly and can thus be used for weld boundary detection [2]. Zou et al. proposed a confidence-weighted method to combine grayscale gradients and structural-light shape features to accurately detect weld boundaries [3]. Krämer et al. proposed a machine learningbased method to detect the different texture features of two workpieces for welding and obtained satisfactory detection results [4]. The studies mentioned above performed well in the welding process or under ideal conditions. However, NDT working environme
Data Loading...
