Study On Numerical Simulation For Laser Welding By Coupled Thermo-hydro-mechanical Method

Laser welding has been widely used in ship manufacturing for its advantages including small welding deformations,energy concentration and narrow heat affected zones.Nowadays,there are two ways of numerical calculation for the welding process: The first is to calculate the molten pool behavior based on computational fluid dynamics theory(CFD);The second it to calculate the stress and deflection based on thermal elastic-plastic finite element method(FEM).But in laser welding,the fluid flow in molten pool is very complex and easy to form the keyhole,which have important influence on the heat transfer and weld bead geometry.Meanwhile,the humps on the weld bead make contribute to the stress concentration and degrade the mechanical property.In order to consider the effects on stress and deflection from molten pool behavior and keyhole,a welding thermos-hydro-mechanical simulation method has been proposed,which improves the simulation accurate for welding mechanical behavior.And a predictive model for weld bead backside width has been developed using simulation model and machine learning algorithm,which improves the predictive performance and enlarge the applied range of simulation.The research works are included as follow:(1)Firstly,the welding thermos-hydro-mechanical numerical simulation method is proposed and the calculation flow is developed.The analysis for welding thermos-hydro-mechanical considering fluid flow of molten pool is realized and the difference of stress and deflection with fluid flow and without fluid flow is analyzed.The simulated results are compared with experimental results and validate the accurate of the developed model.Then,given the feature of dissimilar laser welding,using the thermal and fluid flow multi-field coupling from developed simulation method,a simulation of dissimilar laser welding of Niobium and Ti-6Al-4V is carried out to study the temperature distribution?flow field?keyhole evolution and weld bead formation.The major physical factors,such as recoil pressure induced by vaporisation,surface tension,heat transfer,fluid flow,buoyancy force and their coupling are considered.Results show that the keyhole mainly occurs on the Ti-6Al-4V side due to the differences in physical properties of the materials.The effects of pulse overlapping factor on the weld bead are studied.There is a optimal overlapping factor for the uniform penetration.The mixing of materials mostly occurred in the upper part of the molten pool.The simulated weld bead profile and the phase distribution agree well with the experimental results.(2)In the general CFD model,it is necessary to build the computational domain for both weld material domain and the air upper on it.So when deformation of the weld bead is relative large,it is difficult to dynamically obtain the weld material domain because the geometry of the weld bead is irregular.It makes the development of welding thermos-hydro-mechanical model difficult.In order to overcome this problem,in this paper dynamic mesh method is used in developed welding thermos-hydro-mechanical numerical simulation method to simulate the stress and defletion with relative large deformation of weld bead.An analysis for welding sress and deflection considering molten pool behavior is realized.The effects of humping on stess and defletion are analyzed.The results show that keyhole plays significant role in the hump formation.It is found that the humping defect will increase the localized von mises stress.The distortion is not sensitive to the hump.The simulated weld bead profile agrees well with the experimental results.And under similar welding heat input,the effects of hump on stress and deflection with different welding speed are analyzed.(3)In order to meet the requirement of real-time predicting for weld bead geometry,a predictive model based on machine learning algorithm is developed.Owing to the limits of experimental facilities,the real-time information of molten pool is limited.In this paper,the developed welding thermos-hydro-mechanical numerical simulation method is used in the development of predictive model.The simulation results and experimental data are used as input features to train the predictive model for weld bead backside width model.The results show the adding the simulated data imporves the predictive performance.An analysis for the performance of different machine learning algorithm in small training data is conducted and gird search algorithm is used to optimize the model,the results show the support vector regression is proved better.Then,the generalization ability of optimized model is validated using test data.