Boosting Active Contours for Weld Pool Visual Tracking in Automatic Arc Welding

Abstract

Abstract-Detecting the shape of the non-rigid molten metal during welding, so-called weld pool visual sensing, is one of the central tasks for automating arc welding processes. It is challenging due to the strong interference of the high-intensity arc light and spatters as well as the lack of robust approaches to detect and represent the shape of the nonrigid weld pool. We propose a solution using active contours including an prior for the weld pool boundary composition. Also, we apply Adaboost to select a small set of features that captures the relevant information. The proposed method is applied to weld pool tracking and the presented results verified its feasibility.Note to Practitioners-Welding quality is highly dependent on the human welder's experiences and skills in the gas tungsten arc welding process. In particular, detecting the shape of the nonrigid molten metal during welding plays an important role in improving the welding quality since it provides a prime feedback mechanism for welding control. However, this is a challenging task due to a number of reasons, e.g., strong interference and changing working environment. Applying classical image processing algorithms directly, such as active contour models (ACM), cannot guarantee satisfactory performance due to the above-mentioned reasons. Actually, in practice, ACM more often than not failed to capture the welding pool boundary. In this work, we integrate an applicationdependent prior to the definition of the energy functional of a traditional ACM, thus allowing us to use machine learning techniques to improve the performance of traditional ACM. In density estimation of the prior, AdaBoost is utilized to select features that are most suitable for Gaussian Bayesian classifiers to differentiate wanted boundaries and unwanted ones (e.g., spurious boundaries caused by interfering noise). The proposed approach improves the performance of the traditional ACM and captures the boundary of the

Automated defect classification of Aluminium 5083 TIG welding using HDR camera and neural networks

“…Tables 7 and 8 describe in more details how layers are composed for each architecture. The first layer, "conv: [5,5]× [16]/1-relu", in model 1 in Table 8, contains 16 kernels with each kernel of dimension 5×5, stride 1, and the rectified linear unit, ReLu [41], activation function. It produces 16 feature maps of size 396×483 each.…”

“…Following each convolution layer, there is maximum pooling sampling layer [42]. "max pool: [5,5]/3" translates to a kernel size 5×5 and stride 3, which samples the largest value in the receptive field. The basic blocks in this study are the 5×5 kernels with stride 1 and 2, 3×3 kernels with stride 1 and 2, maximum pooling layers of sizes 5×5 with stride 3 and 3×3 with stride 2.…”

“…To address these limitations, the automation of TIG weld process is crucial. The monitoring and control of the weld pool during the process are critical to the automation in order to ensure quality, consistency and repeatability [3,4,5,6].…”

Coaxial Monitoring of AISI 316L Thin Walls Fabricated by Direct Metal Laser Deposition

“…It generally leads to better results than classical image segmentation methods [ 7 ]. Nevertheless, a challenge for this methodology regards the sensitivity of the geometrical results to the contour initialization [ 8 ]. A methodology evolution, employable to improve the algorithm performances, is the region-based active contours, which models different zones in terms of intensity and sets the region boundaries as the transition between different zones.…”

Automated defect classification of SS304 TIG welding process using visible spectrum camera and machine learning

“…The very same process flexibility gives rise to complex weld pool aspects, which are difficult to assess and control. Several authors tackled the problem of welding processes quality assessment [4,5,6,7,8]. Physically, the welding arc gives rise to a harsh environment and the welding generates a vast amount of information.…”