Combining soft computing techniques and the finite element method to design and optimize complex welded products

Abstract

One of the main objectives when designing welded products is to reduce strains and deformations. Strains can cause excessive angular distortion. This results in a welded product that does not meet acceptable tolerances. The geometry of the weld bead (height and width) depends on the input parameters (speed, voltage and current), and provides the welded joint with strength and quality. As welded products become increasingly complex, deformations become more difficult to predict as they depend greatly on the welding sequence. This paper shows how a combination of the Finite Element Method, Genetic Algorithms and Regression Trees may be used to design and optimize complex welded products. Initially, Artificial Neural Networks and Regression Trees that are based on heuristic methods and evolutionary algorithms were used in predicting the weld bead geometry according to the input parameters. Then, thermo-mechanical Finite Element models were created to obtain the temperature field and the angular distortion using the weld bead geometry that the best predictive models generated. Finally, optimization techniques that are based on Genetic Algorithms were used to validate these Finite Element models against experimental results, and to subsequently find the optimal welding sequence to use in the manufacture of complex welded products.

Meta‐heuristic multi‐ and many‐objective optimization techniques for solution of machine learning problems

“…Optimization techniques have been used widely for the solution of many engineering and real‐life problems such as the allocation of resources (Kyriklidis & Dounias, ), product design and development (Lostado, Fernandez, Mac Donald, & Villanueva, ; Paz, Pei, Monzón, Ortega, & Suárez, ), process planning and scheduling (X. X. Li, Li, Cai, & He, ), wind farm distribution network optimization (Cerveira, Baptista, & Solteiro Pires, ), vertical transportation optimization in skyscrapers (Koo, Hong, Yoon, & Jeong, ), railway line design and timetable optimization (Castillo, Grande, Moraga, & Sánchez‐Vizcaíno, ), freeway travel cost optimization (Shahabi, Unnikrishnan, & Boyles, ), sustainable road network design (Y. Wang & Szeto, ), bridge design optimization (Bisadi & Padgett, ), construction scheduling (Karim & Adeli, ), mountain railway alignment optimization (W. Li et al, ), road weather information system network optimization (Kwon, Fu, & Melles, ), cost optimization of concrete (Sirca & Adeli, ) and steel building structures (Tashakori & Adeli, ) and composite floors (Adeli & Kim, ), free‐form steel space‐frame roof design optimization (Kociecki & Adeli, ), freeway work zone traffic delay and cost optimization (Jiang & Adeli, ), optimal control of bridges (Adeli & Saleh, ) and buildings (Saleh & Adeli, ), and among others.…”

Residual Stresses with Time-Independent Cyclic Plasticity in Finite Element Analysis of Welded Joints

“…The cooling of the welded joints was accomplished in air at a room temperature of 18 • C. The weld cords of all specimens that were studied were deposited sequentially in the same path sequence and on only one side of the plates to be welded in order to create a butt joint with a single V-groove. It is known that different welding path sequences, if applied to the same specimens, can produce different residual stresses [16]. In this case, the three specimens that were studied were manufactured in the same welding path sequence (same weld cord length and direction of welding) and were attached to the ground at the same point.…”

“…Also, the parameters that defined the cyclical behavior of Chaboche of these new FE models (Yield stress = 235 MPa, b = 17.7, C = 24,506 MPa, γ = 2462.6, Q M = 191.4 MPa, Q 0 = 221.9 MPa, µ =118.3, η = 0.5 and n = 4.1) were adjusted iteratively. In this case, a Mean Absolute Error (MAE) threshold value of about 15 was preset as an acceptable error and used at an industrial level [16]. The iterative adjustment was continued until the MAE of each of the three studied specimens was about 15.…”

“…The minimum mesh size that was considered for the FE models that were generated were 0.75 mm for the plates, 1.0 mm for the weld bead, and 1.5 mm for the refractory surface (Figure 5d). The thirteen parameters that are defined by each of the three FE models were adjusted by evolutionary optimization techniques that are based on Genetic Algorithms (GA) [16] using as reference the temperature field obtained by the thermographic camera. Subsequently, these thirteen Similarly, the thermal conduction phenomena considered six parameters that were based on the different pairs of contacts that made up the welded joints (melting_point, contac_P_init, contac_P_center, contac_P_end, contact_P1_P2, and contact_P_G).…”

“…The thirteen parameters that are defined by each of the three FE models were adjusted by evolutionary optimization techniques that are based on Genetic Algorithms (GA) [16] using as reference the temperature field obtained by the thermographic camera. Subsequently, these thirteen parameters were set as boundary conditions for the three proposed FE models to solve the angular distortion and residual stress fields.…”