关键词: 熔融沉积成型零件, 质量预测, 10—折交叉验证, 稀疏深度信念网络, 自适应布谷鸟搜索算法, 增材制造

Abstract: Tensile strength,warping degree and surface roughness are important indicators to evaluate the quality of Fused Deposition Modeling (FDM) parts.Accurate and stable prediction of the fused deposition modeling is conducive to the development of FDM technology.For this reason,a quality prediction method for fully filled FDM parts based on optimized deep belief network was proposed.According to the FDM production process,the main variables affecting the quality of parts were selected,and the process parameter combination with the greatest influence on the product quality was determined by using the correlation analysis method,so as to obtain the input variables of the prediction model.The validation error of 10-fold cross-validation was taken as the fitness value,and the optimal hyperparameter combination of Sparse Deep Belief Network (SDBN) was determined based on the grid search.Adaptive Cuckoo Search (ACS) algorithm was used to optimize SDBN,and a quality prediction model for fully filled FDM parts was established.The prediction performance of ACS-SDBN model was compared with that of the ACS-DBN,DBN and BP models,and the results showed that the fully filled FDM parts quality prediction method of sparse constraint Deep Belief Network optimized by Adaptive Cuckoo Search algorithm had better prediction accuracy and stability.

Key words: fused deposition modeling parts, quality prediction, 10-fold cross-validation, sparse deep belief network, adaptive cuckoo search algorithm, additive manufacturing