Study Of In-situ Monitoring Methods In Selective Laser Melting Process

Recently,additive manufacturing(AM),3D printing,or rapid prototyping technology has achieved rapid development.Based on the principle of layered manufacturing,the metal AM technology directly uses the layer-by-layer method to manufacture parts as the digital model.It becomes a means of rapid prototyping and manufacturing combining with the numerical control processing,casting,and forging.The metal AM technology has broad application prospects in many fields such as biomedical,aerospace and lightweight design manufacturing.Selective laser melting(SLM)technology,as one of the most practical metal AM manufacturing technologies,is widely studied at home and abroad.Problems such as the accuracy and mechanical properties of SLM parts have hampered the further development of the SLM technology.In order to overcome the influence of factors and obtain high-quality molded parts,process control is the essential method to improve the accuracy and mechanical performance of the SLM parts.Process monitoring is a necessary prerequisite for the process control.What's more,it is important to make the reasonable and effective process monitoring.In this paper,the status and development of the SLM process monitoring were reviewed and summarized.Considering the situation of the monitoring process,an off-coaxial method was used to explore the deep learning application in the monitoring of the melting process.The melted states were realized by simplified monitoring approaches and effective recognition method.The main contents and innovations of the thesis were as follows:(1)This paper discussed the general problems within.SLM components and existing monitoring methods.It pointed out the requirement of the further monitoring research during the SLM process.The signal generation and their corresponding defects were analyzed.Considered the types of defects to be monitored,an off-coaxial monitoring system was proposed compared with the existing signal acquisition methods.Acquisition of different signals was enabled by this setup during the SLM process.(2)An acoustic signal monitoring system was established for the SLM process.I analyzed generation principles and dynamic characteristics of the acoustic signals.Acoustic signals among different frequency bands were characterized separately.The comparative analysis was conducted with the original signal features,features after dimension reduction with the principal component analysis,and features after dimension reduction with the Fisher linear analysis.The melted states were recognized by the dynamic characteristics of the acoustic signal.The results verified that it was feasible to monitor the SLM process using acoustic signal.(3)A near-infrared(NIR)image monitoring system was set up for the SLM process.The light signal generation,the plume dynamic characteristics,and the spatter dynamic characteristics were analyzed in the NIR images.The relationship between the image characteristics and the SLM process stability was explored.The SLM melted states were classified based on the features from the NIR images as the laser power and laser scanning speed were changed.The results showed that the plume and spatter features extracted by the NIR images were feasible for the melted state monitoring.Effective recognition results could be obtained by the proposed method.(4)The acoustic signal and NIR image were fused to monitor the states of the SLM process.The trend and relationship were quantized with the melted track feature during changing the laser power and the laser scanning speed.Considering the requirements of computational accuracy and time complexity,deep learning was used to extract features from acoustic signals and NIR images.The features were fused to monitor the melting status.SLM states were classified with the features from the deep learning neural network.Results indicated that the state recognition reduced empirical analysis and improved the recognition speed using deep learning method.Fusion monitoring with acoustic signals and NIR image features made the state recognition of SLM process more effective and feasible.In summary,the comparison of the overall test results showed that the SLM states could be monitored by the acoustic signal and NIR image through this proposed SLM platform and monitoring method.This research method provided a new theoretical and reliable monitoring method for the fusion monitoring of the SLM states.