Intelligent Evaluation Method For Fatigue Life Of Welded Joints And Its Application

As an effective connection method,welding is widely used in the fields of mechanical engineering,such as petrochemical,rail vehicles,marine ships and so on.In the dynamic loading environment,fatigue failure of the welded joints is easy to occur,and fatigue fracture is the main failure mode.Characteristics of metal material,welding process and welded structure make the design and analysis of welded joints more complex.Artificial intelligence technology can achieve the ability of knowledge acquisition,classification and prediction through complete self-study without any prior knowledge,so as to provide a new way for fatigue life evaluation of welded joints.On basis of the summary and analysis of current methods for fatigue analysis of welded joints,artificial intelligence technology is introduced for fatigue life evaluation of welded joints.Deep researches on evaluation method for fatigue life influence factors of welded joints,fitting of the master S-N curves,establishment of the intelligent fatigue life prediction model and study of the relation of stress and life are carried out.The main contents of this dissertation are as follows:(1)An evaluation method based on rough set theory for fatigue life influence factors of welded joints is proposed.Neighborhood rough set theory can deal with continuous attribute information while discretization is not necessary.Taking advantage of this feature,neighborhood rough set model is established to analyze the fatigue life influence factors of welded joints.Fatigue database of welded joints is established,then the evaluation process by using neighborhood rough set is determined,the forward greedy algorithm is used to complete attribute reduction,and the key influence factor set of the welded joints is obtained.The implicit relationship between the influence factors and fatigue life is studied,and the weights of the influence factors are determined.(2)A master S-N curve fitting method based on fatigue characteristic domain is proposed.Based on the reduction result set of neighborhood rough sets,the automatic classification of fatigue specimens of welded joints is realized,fatigue samples with different value of key influence factors fall to different domains.Hereby,the concept of fatigue characteristic domain is proposed,and the dividing method of fatigue characteristic domain is determined.According to the fatigue test data of the aluminum alloy welded joints,mater S-N curve fitting and master S-N curves fitting based on the characteristic domains are carried out respectively.Goodness of fit statistics results show that when the fatigue characteristic domain is divided,the degree of dispersion level of fatigue sample data and the standard deviation of the S-N curve has been further reduced.(3)Experimental verification of fatigue life evaluation method based on fatigue characteristic domain is carried out.The laboratory fatigue test of aluminum alloy materials 5083 and 5A06 T-welded joints is carried out,the finite element model is established,computation of the structural stress and the equivalent structural stress transformation is completed,the master S-N curve and the master S-N curves cluster based on the fatigue characteristics are fitted.Case study of fatigue life evaluation of a T-welded joint of 5A06+5A06 shows that fatigue life evaluation by using the master S-N curve fitting method based on fatigue characteristic domain is more closer to the actual fatigue life.(4)Two optimized neural network models for fatigue life prediction of welded joints are established.Based on the rough set theory,key factors which influence the fatigue life of the welded joints are obtained.Using the key factors as the input of the neural network,the RS_RBFNN model for fatigue life prediction of welded joints is established.To further improve the prediction accuracy,the RS_PSO_BPNN neural network model is established.The two neural network models are compared by using the fatigue test data of Titanium alloy welded joints.Comparison results show that both of the two models can be used for fatigue life prediction of welded joints,with the same number of training samples,the RS_PSO_BPNN model has higher prediction accuracy than the RS_RBF model.(5)A kind of stress-life model based on PSO_SVM is proposed.Since fatigue tests are time-consuming and costly,it is normally difficult to obtain large sample data that meets the requirements of the traditional statistical methods.Taking advantages of the ability of support vector machines in dealing with small sample data,a support vector machine model based on particle swarm optimization(PSO_SVM)is proposed for representation of the relation of stress and life.Simulation results show that the proposed model has better noise immunity and higher prediction accuracy.When fatigue character domains are divided,the prediction accuracy of the SVM model is further improved.In-depth study of modeling method for fatigue life analysis of welded joints,establishment of evaluation mathematical model for fatigue life influence factors of welded joints,exploration of the optimization of the mastter S-N curve fitting method,establishment of neural network model for fatigue life predtion and the support vector machine model for stress life relation of welded joints are all of great importance for increasing the prediction accuracy,descreasing the prediction complexity and improving the anti-fagiue design method of the welded joints.