Bead geometry modeling on uneven base metal surface by fuzzy systems for multi-pass welding

• Published in: Expert systems with applications 2021-12, Vol.186, p.115356, Article 115356
• Main Authors: Horváth, Csongor Márk, Botzheim, János, Thomessen, Trygve, Korondi, Péter
• Format: Article
• Language: English
• Subjects: Algorithms; Bacterial Memetic Algorithm; Base metal; Beads; Coefficients; Contact angle; Coordinates; Error analysis; Fuzzy Systems; Gas tungsten arc welding; Gas welding; Inert gas welding; Machine Learning; Metal surfaces; Modelling; Multi-pass Welding; Polynomials; Position measurement; Process variables; Rare gases; Regression analysis; Regression models; Segments; Shape functions; Weld Bead Profile Modeling; Weld defects
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2021.115356

•A fuzzy system-based weld bead profile model is proposed for multi-pass welding.•Unevenness of the base metal surface is considered in the welding groove.•Bead characteristics and shape are reconstructed from estimated profile coefficients.•Experimental validation demonstrates the adequacy of the proposed model. This paper presents a modeling method of weld bead profiles deposited on uneven base metal surfaces and its application in multi-pass welding. The robotized multi-pass tungsten inert gas welding requires precise positioning of the weld beads to avoid welding defects and achieve the desirable welding join since the weld bead shapes depend on the surface of the previously deposited beads. The proposed model consists of fuzzy systems to estimate the coefficients of the profile function. The characteristic points of the trapezoidal membership functions in the rule bases are tuned by the Bacterial Memetic Algorithm during supervised training. The fuzzy systems are structured as multiple-input-single-output systems, where the inputs are the welding process variables and the coefficients of the shape functions of the segments underlying the modeled bead; the outputs are the coefficients of the bead shape function. Each segment surface is approximated by a second-order polynomial function defined in the weld bead’s local coordinate system. The model is developed from empirical data collected from single and multi-pass welding. The performance of the proposed model is compared with a multiple linear regression model. During the experimental validation, first, the individual beads are evaluated by comparing the estimated coefficients of the profile function and other bead characteristics (bead area, width, contact angles, and position of the toe points) with the measurements, and the estimations of a multiple linear regression model. Second, the sequential placement of the weld beads is evaluated while filling a straight V-groove by comparing the estimated bead characteristics with the measurements and calculating the accumulated error of the filled groove cross-section. The results show that the proposed model provides a good estimation of the bead shapes during deposition on uneven base metal surfaces and outperforms the regression model with low error in both validation cases. Furthermore, it is experimentally validated that the derived bead characteristics provide a suitable measure to identify locations sensitive to welding defects.