Machine acceleration time series prediction for dimensional accuracy of 3D printed parts  

作　　者：Jayanta Bhusan Deb Shilpa Chowdhury Soumik Chowdhury Gourab Paul Tonay Pal Jayeeta Deb Sudipta Deb 

机构地区：[1]Department of Engineering&Technology,Western Carolina University,NC,28723,USA [2]Department of Information Systems and Analysis,Lamar University,TX,77705,USA [3]Department of Computer Science,The Memorial University of Newfoundland,NL,A1B3X5,Canada [4]Department of Marketing,University of Dhaka,Dhaka,1000,Bangladesh [5]Department of Management,University of Dhaka,Dhaka,1000,Bangladesh [6]Department of Medicine and Surgery,University of Chittagong,Chittagong,4331,Bangladesh [7]Institute of Health Economics,University of Dhaka,Dhaka,1000,Bangladesh

出　　处：《Data Science and Management》2024年第3期218-227,共10页数据科学与管理（英文）

摘　　要：This study explores the influence of infill patterns on machine acceleration prediction in the realm of three-dimensional(3D)printing,particularly focusing on extrusion technology.Our primary objective was to develop a long short-term memory(LSTM)network capable of assessing this impact.We conducted an extensive analysis involving 12 distinct infill patterns,collecting time-series data to examine their effects on the acceleration of the printer’s bed.The LSTM network was trained using acceleration data from the adaptive cubic infill pattern,while the Archimedean chords infill pattern provided data for evaluating the network’s prediction accuracy.This involved utilizing offline time-series acceleration data as the training and testing datasets for the LSTM model.Specifically,the LSTM model was devised to predict the acceleration of a fused deposition modeling(FDM)printer using data from the adaptive cubic infill pattern.Rigorous testing yielded a root mean square error(RMSE)of 0.007144,reflecting the model’s precision.Further refinement and testing of the LSTM model were conducted using acceleration data from the Archimedean chords infill pattern,resulting in an RMSE of 0.007328.Notably,the developed LSTM model demonstrated superior performance compared to an optimized recurrent neural network(RNN)in predicting machine acceleration data.The empirical findings highlight that the adaptive cubic infill pattern considerably influences the dimensional accuracy of parts printed using FDM technology.

关 键 词：Extrusion-based 3D printing Printer acceleration predictionInfill pattern Root mean square error Dimensional accuracy 

分 类 号：TN9[电子电信—信息与通信工程]