The automatic support generation methods are indispensable for the Digital Light Processing (DLP), the Stereolithography (SL) and the Fused Deposition Modeling (FDM) 3D printing. However, the automation level of these methods is not high and people need to set the orientation and adjust the supports manually. We model the automatic support generating problem as a two level parametric optimization problem. At the upper level we optimize the orientation. The objectives include the deformation, the model building time, the material quantity for the supports, and the surface quality. At the lower level, we check the conditions for various types of supports given an orientation, including supports for the overhang surfaces (overhang edges, overhang points), for the large flat surfaces, for the stability against the gravitational and external pulling forces when printing, and we generate supports with suitable shapes and density. We need to use the computer graphics and the image processing technologies to process the meshes, and batch evaluations of different orientations are necessary. These cause heavy computational burdens. Based on the Graphics Processing Unit (GPU), we employ the Genetic Algorithm (GA) and the Ordinal Optimization (OO) method to solve the problem, making good use of both the software and hardware technologies. The research can solve the automatic support generation problem systematically, can reduce or even eliminate human interventions, can reduce the failure possibility, and can raise the printing quality.
自动支撑技术对于数字光处理(DLP)、立体印刷(SL)和熔融沉积模型(FDM)等3D打印方法而言是必不可少的,但是该技术当前自动化程度不高,停留在凭经验设定摆放指向并手工辅助调整支撑的阶段。本项目将该问题建模成双层的多目标参数优化问题:顶层为对摆放指向的优化,优化目标有制件变形、构建时间、支撑用料、表面质量等;底层为在给定摆放指向情况下对各种支撑条件的判断,包括悬吊面(悬臂边、悬吊点)、大平缓面、打印过程中的稳定性、外界拉拔力等,并生成形状和密集程度合适的支撑。该问题需要用图形图像方法处理三角网络,并需要批量评估摆放指向等,计算量巨大。本项目计划“软”、“硬”并重,采用遗传算法、序优化方法结合图形处理器(GPU)进行求解。预期研究成果能够系统化地解决自动支撑问题,大幅减少或者消除人工干预,降低3D打印失败率,提升3D打印的质量。
自动支撑技术对于数字光处理(Digital Light Processing)、立体印刷(Stereolithography Apparatus)和熔融沉积模型(Fused Deposition Modeling)等3D打印方法而言是必不可少的,但是该技术当前自动化程度不高,停留在凭经验设定摆放指向并手动辅助调整支撑的阶段。本项目将该问题建模成双层的多目标参数优化问题:顶层为对摆放指向的优化,优化目标有制件变形、构建时间、支撑用料、表面质量等;底层为在给定摆放指向情况下对各种支撑条件的判断,包括悬吊面(悬臂边、悬吊点)、大平缓面等,并生成形状和密集程度合适的支撑。通过采用遗传算法、序优化方法并结合图形处理器并行计算,为3D打印自动支撑这一问题提供系统化解决方案。并分析各种算法的复杂度和执行时间,能够根据制件的目标在满足要求的情况下选择复杂度低的算法,在算法执行之前预估算法的执行时间。针对DLP、SL和FDM等不同打印方式,都能够自动、快速地实现合适支撑的生成。其中,自动支撑的判断条件离不开切片的支持,切片的层与层之间、每一层内部计算切线段、进行填充均是同质化操作,本项目基于切片过程的不同算法,设计了不同的并行策略来合理分配计算负载,提高GPU利用率。为进一步提升3D打印质量,本项目对于3D打印的变形进行了研究,考虑到3D打印涉及到加热、冷却、聚合等反应使蜷曲和收缩现象普遍的问题,本项目构建深度神经网络进行误差预测与补偿以减少3D打印的误差。本项目成果能够系统化地解决自动支撑问题,大幅减少或者消除人工干预,降低3D打印失败率,提升3D打印的效率和质量。并且,本项目组将3D打印与医学相结合,构建了颅骨手术模型,并用于医学教学。项目组共发表论文42篇,其中SCI收录15篇,EI收录22篇。项目执行期间,申请国家发明专利12项,其中8项已经授权,申请并授权PCT/美国专利2项。项目执行期间共授权国家发明专利13项,授权PCT/美国专利4项。
{{i.achievement_title}}
数据更新时间:2023-05-31
Combining Spectral Unmixing and 3D/2D Dense Networks with Early-Exiting Strategy for Hyperspectral Image Classification
一种改进的多目标正余弦优化算法
基于混合优化方法的大口径主镜设计
变可信度近似模型及其在复杂装备优化设计中的应用研究进展
采用深度学习的铣刀磨损状态预测模型
基于切片的3D打印优化研究
基于序优化和遗传算法的大规模交通系统协调控制研究
基于Agent的并行仿真支撑技术研究
基于3D打印及组织工程技术构建精确阴茎形态软骨支撑体