3D列印之超彈性材料蒲松比精密量測與4D列印之平面網格變形──面具製作搭配反向設計

Precise Measurement of Poisson’s Ratio of 3D-printed Hyperelastic Materials and Transformation of 2D grids into 3D gridshells by 4D Printing: A Case Study of Face Mask Design

徐亦賢(Yi-Xian Xu)

指導教授 ： 莊嘉揚

國立臺灣大學/工學院/機械工程學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101683

摘要

本研究分為兩個方向，兩者皆以3D列印作為出發點，第一部分針對3D列印材料的性質精密量測作為目標；第二部分則針對3D列印材料於4D列印的應用進行新穎加工方法的探討，看似不存在關聯，但第一部分的內容為量測方法的探討，其結果可應用於第二部分的材料作為未來相關研究的使用。
隨著3D列印技術的普及，快速成型(Rapid prototyping)超彈性材料也成為3D列印不可或缺的特色之一，這也成為軟性機器人製造與設計時常用的方法。然而超彈性材料因類似橡膠的性質，使其存在週次軟化的現象，此現象對於機器人性能有顯著影響。由於超彈性材料在大變形下難以進行精確量測，使3D列印材料性質的研究多以硬材料為研究目標，且在週次軟化的研究中，並無針對其非線性蒲松比進行量測，因此本研究旨在精確量測3D列印超彈性材料的機械性質，其中透過量測熱塑性聚氨酯(TPU 85A, NinjaTek, Ninjaflex)於週次拉伸試驗中的機械性值了解週次軟化現象的非線性蒲松比變化。TPU 85A為現今熔融堆疊式3D列印中剛性最低的材料，其優異的彈性成為許多軟性機器人材料的首選。為了精確量測其性質，本研究使用平面影像關聯法搭配Reference sample compensation (RSC)修正方法來達到此目的，該方法能有效修正進出紙面位移帶來的量測誤差。在此修正方法下，成功觀測到蒲松比隨著週次軟化與遲滯變化的現象，在第一週次下蒲松比由較低的值0.45 ± 0.005提升至較高的值0.48 ± 0.005，而在後續週次下蒲松比維持在較高值且隨著拉伸應變的增長有些為變化，此蒲松比變化導致試片在最大拉伸應變為17.5%下體積有些微的增加(≈ 1%) 。此發現可協助有限元素法使用者進行軟性機器人之輔助設計，也有助於了解週次軟化現象的物理機制並提供其他研究來驗證週次軟化的理論模型。
4D列印為3D列印技術的延伸，係使用智能材料使3D列印的物件可透過外界刺激而變形，此機制可以應用於機器人或形狀變形等領域。形狀變形的優勢使其能透過簡單的結構轉換為複雜結構，克服3D列印對於複雜曲面印製的困難。過去研究未使用形狀記憶聚合物組成的平面網格透過4D列印來進行立體網格的製作，且反向設計其平面網格的過程困難且繁瑣。因此本研究希望以形狀記憶聚合物作為平面網格材料經4D列印的過程來做為立體網格的加工方法，並以人臉的理想模型系統(Model system)來測試此方法可行性，而立體網格的反向設計除了以人為設計外將搭配深度學習來加速該過程。而其中本研究透過熔融堆疊式3D列印機在列印智能材料(SMP55)時會在材料中殘留預應力的特性，找出一個能穩定列印且又能使SMP55能產生高達60%收縮率的列印參數，並透過與PLA組合的雙層結構使其能向上與向下彎曲，在這兩種材料四種組合配置形成的平面網格來達成4D列印。人為設計的部分藉由尋找平面網格不同設計下變形的規律，搭配有限元素法成功完成了三個日本能面的製作，也驗證了該方法於立體網格加工方法的可行性，且能藉由本研究設計的過程可將此技術應用於其他立體網格的製作。深度學習的部分，本研究將人臉面具的平面網格設計參數化來生成隨機的人臉面具，以此大量的隨機面具來進形人臉面具反向設計模型訓練的依據，其中條件式深層卷積生成對抗網路作為反向設計的模型架構，神經網絡根據目標的深度照片來生成平面網格設計，而其生成的人臉面具與目標的深度照片在結構相似性的計算下有77%的相似度，其結果仍存在一定的進步空間。在本研究網絡模型訓練的結果、參數化隨機人臉生成與4D列印平面網格列印的方法之上，未來能使反向設計的神經網絡模型更加完善。

關鍵字

週次軟化 ； 熱塑性聚氨酯(TPU) ； 平面影像關聯法 ； 蒲松比 ； 4D列印 ； 形狀變形 ； 反向設計 ； 生成對抗網路 ； 形狀記憶聚合物

並列摘要

This research is divided into two directions, both of which use 3D printing as the starting point. The first part is aimed at precise measurement of the properties of 3D printing materials; the second part is a novel application of 3D printing materials by 4D printing. The discussion of two directions do not seem to be related, but the content of the first part is the discussion of measurement methods, and the results can be applied to the materials of the second part for future related research.
With the popularization of 3D printing technology, rapid prototyping of hyperelastic materials has become one of the indispensable features of 3D printing, which has also become a popular method in the manufacture and design of soft robots. However, due to the rubber-like properties of hyperelastic materials, cyclic softening is a phenomenon which has a significant impact on the performance of the robot. As hyperelastic materials are difficult to have accurate measurement under large deformations, the researches of the properties of 3D printing materials mostly focus on hard materials. Besides, in the study of cyclic softening, the non-linear Poisson's ratio has not been measured. Therefore, this study aims to understand the effect of cyclic softening on the mechanical properties of hyperelastic materials. We experimentally study the cyclic softening phenomenon by measuring the changes of mechanical properties of thermoplastic polyurethane (TPU85A, NinjaTek, Ninjaflex) in the cyclic tensile test. To measure the properties precisely, we use two-dimensional digital image correlation (2D-DIC) combined with the reference sample compensation (RSC) method. This accuracy-enhanced method can effectively eliminate the measurement errors induced by the unavoidable out-of-plane displacements and lens distortion. We find that the Poisson’s ratio of TPUs also exhibits large hysteresis in the first cycle and then approaches a steady state in subsequent cycles. Specifically, it starts from a relatively low value of 0.45 ± 0.005 in the first loading, then increases to 0.48 ± 0.005 in the first unloading, and remains largely constant afterward. Such a change in the Poisson’s ratio results in a slight volume increase (≈ 1%) at a maximum strain of 17.5%. Our findings are useful for those who use finite element method to analyze the mechanical behavior of TPU, and shed new light on understanding the physical origin of cyclic softening.
4D printing is an extension of 3D printing technology. By printing smart materials, the 3D-printed objects have the ability to deform again due to certain external stimuli such as heat. This mechanism can be used in fields such as robots or shape morphing. The advantage of shape morphing makes it possible to convert a simple structure into a complex structure to overcome the difficulty of 3D printing of complex surfaces. In the past, the study did not use the 2D grid composed of shape memory polymer to produce the 3D gridshell through 4D printing, due to the difficulty and cumbersomeness of the process in reversely designing of 2D grid. Therefore, this research hopes to use shape memory polymer as the material of 2D grid to process the 3D gridshell through the 4D printing process, and to test the feasibility of this method with the ideal model system of the human face. In addition to the human design, the reverse design of the 3D gridshell will be combined with deep learning to accelerate the process. We use the fused deposition modeling 3D printer to print smart materials (SMP55), which will residue pre-stress in the material to achieve 4D printing. In the experiment, we found a printing parameter that enables SMP55 to be stably printed and produce up to 60% shrinkage. By combining with PLA as a double-layer structure, it can bend upwards and downwards after 4D printing. The 2D grid formed by different configuration of combinations of SMP and PLA can achieve a variety of 3D gridshells. In the part of human design, by looking for the law of deformation under different designs of 2D grids, we combined the finite element method to make three Japanese Noh masks, Nomen. The success of human design has verified the feasibility of this method in the processing method of 3D gridshell, and the process of this research design can be used to apply this technology to the production of other 3D gridshell. In the part of deep learning, we use different functions to parameterize the law of deformation to generate random face masks. This large number of random masks can let us use conditional deep convolution generative adversarial networks (cDCGAN) to conduct reverse design based on three-dimensional face information. The neural network generates the design of 2D grid based on the depth photo of the target. By the calculation of structural similarity, the similarity of the output and target is 77%, which is still room for improvement. Based on the results of network model training, the generation of parameterized random face and the 4D printing methods of 2D grids in this study, the reverse-designed neural network model can be more perfect in the future.

並列關鍵字

Cyclic softening ； Thermoplastic polyurethane (TPU) ； 2D-DIC ； Poisson’s ratio ； 4D printing ； Shape morphing ； Inverse design ； Generative adversarial networks (GAN) ； Shape memory polymer

參考文獻

[1] E. Song, J. Li, S. M. Won, W. Bai, and J. A. Rogers, "Materials for flexible bioelectronic systems as chronic neural interfaces," Nature Materials, vol. 19, no. 6, pp. 590-603, 2020.

[2] R. Ghaffari et al., "Soft wearable systems for colorimetric and electrochemical analysis of biofluids," Advanced Functional Materials, vol. 30, no. 37, 2019.

[3] X. Ning et al., "Assembly of advanced materials into 3D functional structures by methods inspired by origami and kirigami: A review," Advanced Materials Interfaces, vol. 5, no. 13, 2018.

[4] A. M. Abdullah, X. Li, P. V. Braun, J. A. Rogers, and K. J. Hsia, "Kirigami‐inspired self‐assembly of 3D structures," Advanced Functional Materials, vol. 30, no. 14, 2020.

[5] C. Yu et al., "Electronically programmable, reversible shape change in two‐and three‐dimensional hydrogel structures," Advanced Materials, vol. 25, no. 11, pp. 1541-1546, 2013.