Data-driven simulation for fast prediction of pull-up process in bottom-up stereo-lithography. (June 2018)

• Record Type: Journal Article
• Authors: Wang, Jun; Das, Sonjoy; Rai, Rahul; Zhou, Chi
• Is Part Of: Computer aided design. Volume 99(2018)
• Page Start: 29
• Page End: 42
• Publication Date: 2018-06
• Subjects: 2D shape context descriptor -- Bottom-up SLA 3D printing -- Separation stress finite element model -- Data-driven simulation -- Neural network; Computer-aided design -- Periodicals; Engineering design -- Data processing -- Periodicals; Computer graphics -- Periodicals; Conception technique -- Informatique -- Périodiques; Infographie -- Périodiques; Computer graphics; Engineering design -- Data processing; Periodicals; Electronic journals; 620.00420285
• DOI: 10.1016/j.cad.2018.02.002
• Languages: English
• ISSNs: 0010-4485

Abstract

Abstract: Cohesive finite element simulation is a mechanics-based computational approach that can be used to model the pull-up process in bottom-up stereo-lithography (SLA) system to significantly increase the reliability and through-put of the bottom-up SLA process. This modeling relates the pull-up velocity and separation of the fabricated part during the pull-up process. However, finite element (FE) simulation of the pull-up process for the individual part is computationally very expensive, time-consuming, and not amenable to online monitoring. This paper outlines a computationally efficient data-driven scheme to predict the separation stress distribution in bottom-up SLA process. The proposed scheme relies on 2D shape context descriptor, neural network (NN), and a limited number of offline FE simulations. Towards this end, FE models and results for the cross-section of n -fold symmetric shapes form our databases. The 2D shape context descriptor represents different shapes through log-polar histograms in our database. A backpropagation (BP) neural network is trained using the log-polar histograms of the geometric shapes as inputs and the FE simulated stress distributions as outputs. The trained NN can then be used to predict the separation stress distribution of a new shape. The results demonstrate that the proposed data-driven method can drastically reduce computational costs and apply to any general databases. The comparison between the predicted results by the data-driven approach and the simulated FE results on new shapes verify the validity of the proposed method. Highlights: Data-driven prediction model for stress distribution prediction in SLA system. Efficient framework for in-situ feedback control system in bottom-up SLA. Numerical experiments demonstrate the efficiency of FEA-NN base prediction model.