【Technical Introduction】
According to the national standard for Visual Testing (VT) of welded steel structures, defects such as cracks, insufficient penetration, poor fusion, overlap, and incomplete filling of welds are not acceptable. However, there are tolerable ranges for porosity, weld crown, weld erosion, leg length, and throat depth. Traditional visual inspection relies on proficient inspectors, but it is slow, prone to fatigue-induced variations, and subjective in judgment. Moreover, it is challenging to trace the root cause and unsuitable for long-term continuous operations due to the inability to accurately record potential defects, leading to project delays and cost overruns. This technology integrates laser triangulation sensors, portable measurement devices, and encoders, and develops dedicated analysis software for quantitative analysis. It establishes a portable fillet weld profile analyzer and a portable groove weld profile analyzer. By utilizing weld profile data, point cloud reconstruction can be performed, and the geometry parameters of the weld can be calculated using arithmetical algorithms. Data analysis is conducted based on standards such as CNS 13021, AWS D1.1, or relevant specifications to determine compliance with acceptance criteria. Furthermore, this technology enables real-time monitoring during the process, resulting in significant savings in labor and manufacturing costs, making it highly practical.


​Figure 1. Portable weld fillet profile analyzer


​Figure 2. Portable groove butt weld profile analyzer


​Figure 3. Cloud map of welding surface

【Project Planning/Technical Applications】
This innovation allows for the measurement of defects in welding profiles, providing valuable data to investigate their root causes. It facilitates the early detection and prompt correction of such defects during the manufacturing process, thus, lead to higher yields and lower production costs. Additionally, this set involves the customization of dedicated inspection specification to meet practical requirements. The software is tailored to quantify characteristics such as insufficient or uneven leg length, base material angle, effective throat depth, horizontal and vertical effective shear lengths, weld width, convexity, irregular weld crown, and other profile defects, as well as the effective weld area. It also includes parameters related to the base material, such as base material angle deformation and undercut. These aspects are integrated into quality control processes to serve as important records for specifications and process contracts. Furthermore, for the detection of surface micro-defects, a 2D industrial camera is utilized alongside a deep learning framework to extract representative defect features and conduct feature discrimination training. Through iterative learning and validation, an effective method for defect detection is achieved. The results of surface micro-defect detection are illustrated in Figure 5.


​Figure 4. Measurement of welding characteristics


Figure 5. Surface micro-defect inspection

【Future Development】
This research delves into the utilization of 2D and 3D sampling techniques for defect detection, developing methods to identify, classify, and recognize features of defects. These techniques offer valuable insights for real-time algorithmic applications in automated defect detection. Their applications can be divided into two main aspects: process monitoring and operational inspection. During the process monitoring phase, the research enables real-time detection on the production line, thereby reducing the manpower and time required for weld inspection. By identifying changes in weld contour and detecting welding defects, it provides timely alerts for human and equipment errors, preventing the introduction of welding defects and allowing for immediate error correction. The improvement in efficiency leads to more economic productivity and enhanced working efficiency. Furthermore, this technology can also be applied to periodical equipment maintenance. By comparing inspection results over time, it establishes trends in component deterioration, providing necessary references for inspection, maintenance, and replacement decisions for operation. Compared to other non-destructive testing techniques, visual sensing technology offers lower costs and thresholds for automated inspection. This project achieves industrial-grade detection capabilities and efficiency, thereby enhancing the quality and speed of welding works. Also, it contributes to maintaining operational reliability and improving competitiveness in industries such as nuclear energy, petrochemicals, renewable energy, shipbuilding, and machinery.

【Contact Information】
Corresponding Author：Dung-di Yu
Tel：03-4711400 Ext. 6731
E-mail：ddyu@nari.org.tw