CASIA OpenIR  > 毕业生  > 硕士学位论文
基于GPU的DLP 3D打印切片与支撑生成算法研究
张淅鹏
学位类型工学硕士
导师熊刚 ; 沈震
2018-05
学位授予单位中国科学院研究生院
学位授予地点北京
关键词3d打印 数字光处理 切片过程 摆放指向优化 支撑生成 遗传算法 Gpu并行计算
其他摘要
       近年来,3D打印已经引起了学术界和工业界的广泛关注。相比于传统的制造工艺,3D打印在个性化定制方面具有性能和成本优势,目前已经在越来越多的领域得到应用。在3D打印中,支撑结构用于保证模型在打印过程的稳定性,从而保证3D打印产品的精度和表面质量。目前,对3D打印过程中的支撑设计、3D模型摆放等技术问题的研究尚不充分,并且缺乏并行化相关工作。本学位论文以数字光处理(Digital Light Processing,DLP)3D打印机为研究对象,对切片、3D模型摆放指向、支撑生成等问题进行了研究,并借助并行工具—图形处理器(Graphic Processing Unit,GPU)对问题求解过程进行计算加速。
       本文以提高DLP 3D打印技术中支撑生成过程的计算效率为目标,重点展开以下三个方面的研究:并行化切片方法、模型摆放指向优化及其并行化方法、支撑生成算法及其并行化方法。首先,并行化切片方法的研究有助于加速切片过程,并为后续两个问题的研究提供了模型截面图形与图像数据。接着,本文对模型摆放指向问题进行了研究。最后,在两个前期工作的基础上,本文进行支撑生成问题研究。在研究方法上,本文在算法层面上利用并行遗传算法来优化模型摆放指向,并提出了基于截面图像数据的支撑生成算法,在硬件层面上应用GPU并行计算技术进一步加速问题求解的过程,提高问题求解的效率,具体研究内容如下:
       第一:并行化切片方法。本文在现有的切片算法的基础上,提出基于GPU的DLP打印技术模型切片过程的并行化实现方法。针对切片过程的不同算法,本文提出了不同的并行策略来合理分配计算负载,提高GPU利用率。实验结果表明,相比于CPU实现,本文提出的GPU并行切片方法在不同模型上能取得34倍左右的加速效果。实验结果验证了本文提出了基于GPU的并行切片方法的有效性,表明了GPU在切片应用中的巨大潜力。
       第二:模型摆放指向优化及其并行化方法。本文利用图形学方法建立模型摆放指向与打印时间、表面质量、支撑面积等优化目标的数学关系,给出了模型摆放指向的多目标优化问题数学模型,并提出了基于并行遗传算法的问题求解方法,最后应用GPU实现算法求解的过程。实验结果表明,在复杂模型摆放指向优化中,基于GPU的遗传算法实现相比于CPU实现能取得10倍左右的加速比,在保证解的质量的前提下,显著地减少了摆放指向优化问题的求解时间。
       第三:支撑生成算法及其并行化方法。本文将向下特征体支撑生成算法由图形域推广到图像域,并提出了该算法基于GPU的并行化方法,有效地减少支撑生成
过程中的计算量,提高了算法效率。实验结果表明,本文提出的支撑生成算法能大幅减少计算时间,其GPU并行实现在复杂模型支撑生成上,相比于CPU实现能取得20倍左右的加速比。
       综上所述,本文主要研究以DLP为代表的3D打印技术的切片、模型摆放指向和支撑生成问题,对于其中的相关算法,本文提出的GPU并行实现方法能显著减少算法执行时间,提高问题求解效率,对促进3D打印过程的自动化和智能化有重要意义。
;
    In recent years, three-dimensional printing (3D Printing) has drawn a lot of attention from both the academic and industrial communities. Compared with traditional manufacturing technologies, 3D printing has shown advantages of flexibility and cost for personalized product and it has been used in a wide range of fields. In 3D printing, support structures are utilized to make models stable during the printing process and support generation methods have an important impact on the precision and surface quality of 3D printing product. However, there are only a few studies which have investigated and analysed the digital data processing in 3D printing, such as model orientation optimization, support structure generation and parallel methods. This thesis focuses on the model orientation optimization and support structure generation issues in 3D printing with digital light processing (DLP). A kind of parallel devices, graphic processing unit (GPU), is used to bear massive computing load and accelerate these applications.
 
    This thesis aims at improving the computing efficiency of the automatic support generation process of 3D printing with DLP. There are three points: 1) a GPU-based parallel slicing method for DLP process, 2) model orientation optimization and its GPU-based parallel method, 3) support generation algorithm and its GPU-based parallel method. We first explore the GPU-based parallel method to speed up the slicing process and obtain the layered graphics and image data. Then, we deal with the model orientation optimization problem and provide the support generation solution based on the optimal model orientation. We combine both the software and hardware acceleration in the computational experiments. In software, parallel genetic algorithms (GAs) are used to optimize model orientation and a layered-image-based down character support generation algorithm is proposed to improve computational efficiency. In hardware, GPU parallel computing is used to implement the related algorithms, which can further accelerate the problem solving process and improve the performance of algorithms. The details of the contents are as follows.
 
    1) A GPU-based parallel slicing method for DLP process. We first introduce mesh slicing algorithm and polygon rasterization algorithm and then propose a GPU-based parallel slicing method for DLP process. Different parallel strategies are proposed to achieve good load balance and efficient GPU utilization. Test results on several models show that the proposed GPU-based slicing method can obtain a speedup ratio of about 34 when compared with CPU implementation. Experimental results validate the effectiveness of the proposed parallel method and show the power of GPU in the slicing issues of DLP process.
 
    2) Model orientation optimization and its GPU-based parallel method. By using geometric computing, we formulate the model orientation as a multi-objective problem, aiming at minimizing the building time, surface quality, and support area. The genetic algorithm (GA) is used to solve the optimization problem and the process of the genetic algorithm is parallelized and implemented on the GPU. Experimental results show that when dealing with complex models,  compared with CPU implementation, the GPU-based genetic algorithm can speed up the process by a factor of about 10, which helps to significantly reduce the optimization time and ensure the quality of solutions.
 
    3) Support generation algorithm and its GPU-based parallel method. We use image information in the down character support generation algorithm instead of graphic information and propose its GPU-based parallel method, which simplifies the calculation and improves the efficiency. Experimental results show that the proposed algorithm can reduce the support generation time and bring performance improvements. When dealing with complex models, compared with CPU implementation, the GPU-based implementation can achieve a speedup factor of about 20.
 
    In summary, we focus on the slicing, model orientation optimization and automatic support generation problems 3D printing with DLP and propose GPU-based parallel methods for the algorithms involved in the processes, which help to reduce the execution time and improve the efficiency. The research achievements can help to make the 3D printing processes more automatic and intelligent.
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/20994
专题毕业生_硕士学位论文
作者单位中国科学院自动化研究所
推荐引用方式
GB/T 7714
张淅鹏. 基于GPU的DLP 3D打印切片与支撑生成算法研究[D]. 北京. 中国科学院研究生院,2018.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
基于GPU的DLP 3D打印切片与支撑生(4308KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[张淅鹏]的文章
百度学术
百度学术中相似的文章
[张淅鹏]的文章
必应学术
必应学术中相似的文章
[张淅鹏]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。