文献类型：期刊文献

英文题名：Machine-learning prediction of selective laser melting additively manufactured part density by feature-dimension-ascended Bayesian network model for process optimisation

作者：Li, Bo[1,2];Zhang, Wei[1,2];Xuan, Fuzhen[1,3]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Addit Mfg & Intelligent Equipment Res Inst, Shanghai 200237, Peoples R China;[3]East China Univ Sci & Technol, Shanghai Collaborat Innovat Ctr High End Equipmen, Shanghai 200237, Peoples R China

年份：2022

卷号：121

期号：5-6

起止页码：4023

外文期刊名：INTERNATIONAL JOURNAL OF ADVANCED MANUFACTURING TECHNOLOGY

收录：;EI(收录号：20222712314944);WOS:【SCI-EXPANDED(收录号:WOS:000819273800003)】；

基金：The work was sponsored by International Collaboration Program from the Science and Technology Commission of Shanghai Municipality in China (No.19110712500), Shanghai Natural Science Foundation (No.20ZR1414000), and National Natural Science Foundation of China (No.52175140).

语种：英文

外文关键词：Machine learning; Feature; Bayesian network; Selective laser melting; Relative density

摘要：Selective laser melting (SLM) is a widely used metal additive manufacturing technique due to its applicability to customising geometrically complex 3D parts. Process optimisation strategies via machine learning (ML) have received great attention due to higher production efficiency. Furthermore, to promote the accuracy of ML models for achieving high as-built density, this work employed a practical approach of feature dimension ascending to obtain twenty-two input variables based on SLM process parameters for a Bayesian network (BN) prediction model. A random forest (RF) algorithm was introduced to the feature screening from the high-dimensional features. The dimension-ascended features for the BN model replaced the source features corresponding to straightforward SLM process parameters such as laser power, laser scan speed, powder layer thickness, and melt hatch spacing. Following feature standardisation, the nonlinear combinations of augmented higher-ordered features contributed to training the BN model for predicting the relative densities of as-built parts using the known process parameters. Three regression evaluation indexes were employed to evaluate the BN models. The verification results declared that the BN model achieved a high prediction accuracy of as-built relative densities based on feature-dimension ascending. The significance of this work lies in that the manual setting of SLM process parameters can obtain a high-credible as-built density. This ML strategy can be employed to quickly find optimised SLM process parameters rather than through many tedious process experiments and destructive sample profiling measurements.