Filtern
Dokumenttyp
Schlagworte
- Solidification cracking (24) (entfernen)
Organisationseinheit der BAM
Eingeladener Vortrag
- nein (5)
Laser welding is a widely established manufacturing process in many industry sectors. Solidification cracking as well as the weldability of materials is still since many years a highly contentious issue, particularly regarding the causes of the hot crack formation. Many of studies have been conducted to determine the critical conditions of occurrence of the solidification cracking. In this study a 2D in-situ observation technique in conjunction with laser diodes as the illuminating source has been employed to measure the arising strain field during the laser beam welding process. For the first time the employed technique enabled the in-situ measurement of the transient strain field at the surface of the workpiece directed to the laser beam in the critical range, where the solidification cracking normally occurs. Thus the critical threshold strain values at high temperatures characterizing transition from crack free to crack concomitant welding process could be deduced.
This paper represents the results for proposed optical flow method based on the Lucas-Kanade (LK) algorithm applied to two different problems. The following observations can be made:
- The estimated strain and displacement for conducted tensile test are generally very close to those measured with conventional DIC-technique.
- The LK technique allows measurement of strain or displacement without special selection of a region of interest.
Using a novel optical measurement technique together with the optical flow algorithm, a twodimensional deformation analysis during welding was conducted. This technique is the first to provide a measurement of the full strain field locally in the immediate vicinity of the solidification front. Additionally, the described procedure of the optical measurement allows the real material-dependent values of critical strain characterizing the transition to hot cracking during laser welding processes to be determined.
In this study, a three-dimensional CFD-simulation model was developed to simulate the fluid flow in the weld pool. The CFD-model showed a bulging region in the middle of the depth, which is separated from the top surface and bottom surface by two narrowing regions. It can be concluded that the interaction of the movement of the laser source with the Marangoni vortex leads to a teardrop shape at the upper and bottom surface of the workpiece. Additionally, it shows that the bulging in the weld is a result of the backflows on the upper and lower sides due to the thermo-capillary-driven flows. The weld pool shape was used as a heat source in a two-dimensional thermomechanical model, which allows a highly accurate transformation of the weld pool dimensions obtained from the CFD model. This developed technique allows the consideration of physical aspects, which cannot be considered when using traditional heat sources. The mechanical model has shown that the chronological order of the solidification of the weld has a significant influence on the nature and distribution of the stresses in the weld. High tensile stress has been observed in the bulging region, i.e. in the susceptible region for solidification cracking, when compared to the other narrowing regions, which show compressive stress.
The strain field can reflect the initiation time of solidification cracks during the welding process. The traditional strain measurement is to first obtain the displacement field through digital image correlation (DIC) or optical flow and then calculate the strain field. The main disadvantage is that the calculation takes a long time, limiting its suitability to real-time applications. Recently, convolutional neural networks (CNNs) have made impressive achievements in computer vision.
To build a good prediction model, the network structure and dataset are two key factors. In this paper, we first create the training and test sets containing welding cracks using the controlled tensile weldability (CTW) test and obtain the real strain fields through the Lucas–Kanade algorithm. Then, two new networks using ResNet and DenseNet as encoders are developed for strain prediction, called StrainNetR and StrainNetD. The results show that the average endpoint error (AEE) of the two networks on our test set is about 0.04, close to the real strain value. The computation time could be reduced to the millisecond level, which would greatly improve efficiency.