Study On Performance Prediction And Acoustic Signal Of Laser Selective Melting Molding Parts

Selective laser melting is an advanced additive manufacturing technology.The principles of this technology is to melt preset-metal powder selectively by high energy laser beam.Doing the melting process time after time in each layer,metal material can be deposited in layers.Compared with the traditional reductive manufacturing,this technology can fabricate any complicated parts very rapidly without process like molding,forging and welding.Though selective laser melting is very advanced,still there are many problems restrict its development.According to its problems,microstructures of formed parts,anisotropy,optimizing parameters and monitoring method were studied in this paper.The specific test methods and test results are as follows:Dendritic growth characteristics and mechanism research were carried out using scanning electron microscope and optical microscope to observe the morphology and microstructure of the molten pool.Based on the DD2 single crystal growth theory,the principle of dendrite growth of selective laser melting parts was explained.The size and average primary dendrite spacing were measured to show characteristics of microstructure.The anisotropic study for forming parts,using multiple forming directions to manufacture tensile test specimen.And aging treatment was used to strengthen the test parts.Property like parts'dimensions,surface forming quality,density and tensile strength were measured using Multiple testing equipment.SPSS software was used for variance analysis.Defect distribution and element segregation were analyzed by optical microscopy and EDS techniques,respectively.The process optimization and mechanical property prediction were studied by controlled variable method and orthogonal test.The cross section of single channel and fracture morphology were studied.The BP-GA neural network was built by Matlab,and the process parameters were used as the features to predict the tensile strength.Acoustic emission signal acquisition and classification were studied.Microphone and camera with filters ware used to record the acoustic emission and plasma topography during single-pass scanning.The forming quality is controlled by the paving layer thickness extracting different forming quality signal,the feature extraction is performed by fast Fourier transform(FFT)and Wavelet analysis.Based on machine learning,SPSS Modeler was used to classify and identify signal features to construct corresponding model.The results of dendrite growth study showed that the molten pool's geometric characteristic is flat and wide with the columnar dendrites in different growth directions.Its growth direction mainly depends on the area at the bottom of the molten pool;The CET transition occurs in the above area of molten pool and the size of the transition zone is related to the depth of remelting area.The results of anisotropic study showed that as the building angle increases,test parts'width increases and thickness decreases.The results of density analysis showed that density of the parts increased by adding support or increasing the building angle that can reduce the overhang surface.While the aging treatment can produce porous oxide layer,which decrease parts density;the mechanical properties anisotropy is caused by the defect distribution and element segregation.The defects and element segregation between the layers are the weak points of the formed parts.The optimization process parameters and prediction mechanical properties study showed that the laser power has the greatest influence on the mechanical properties.The BP-GA neural network can effectively predict the mechanical properties,the coefficient R~2=0.73.The results of acoustic emission signal acquisition and classification study showed that the spectrum are more concentrated when the laser is applied to the substrate,but the spectrum distribution is scattered when the laser is applied to the metal powder.The classification results were outputed by Logistic,Decision Tree and Bayesian network.The recognition rate reaches 91%,which can do qualitative judgment by the acoustic signal.