=Paper=
{{Paper
|id=Vol-2692/paper5
|storemode=property
|title=Generative Design for Social Manufacturing
|pdfUrl=https://ceur-ws.org/Vol-2692/paper5.pdf
|volume=Vol-2692
|authors=Carlos González-Val, Santiago Muiños-Landin
|dblpUrl=https://dblp.org/rec/conf/ecai/Gonzalez-ValM20
}}
==Generative Design for Social Manufacturing==
https://ceur-ws.org/Vol-2692/paper5.pdf
Generative design for Social Manufacturing
Carlos González-Val1 and Santiago Muiños-Landin1
1
Technological Centre AIMEN, C/ Relva, 27 A. Torneiros - 36410 Porriño - Pontevedra, Spain
Abstract— Social Manufacturing is a novel approach where of services (see 1), technology or design each of them with
different members of a community can interact within a different degree of explicitness regarding agent (community
cyber-physical social space in order to achieve specific and member) knowledge. One key element within the SM
personalized solution for manufacturing processes. In such
digital scenario, interactions, driven by information flows, can landscape is the so called prosumer, a consumer that
be divided in different branches depending on the actors participates actively in social manufacturing assuming also
involved in the whole process. One particularly critic branch the role of a producer. The more involved the prosumers
for such collective production, is the one that captures the (agents) get into the product manufacturing, the more
information related with product design, due to its direct reach the final product results through self-organizing and
link with creativity. Here we show how active tools based
on Artificial Intelligence triggers artificial creativity that social-enabled mechanism. However, contrary to what would
can be used for the user capabilities augmentation. We be an ideal SM context, where all the members of a
show how the use of deep generative models, based on community are pure prosumers, the current approaches to
Variational Autoenconders, offers solutions for a particular SM shows the presence of customers and also services
social manufacturing platform for furniture design combined providers in an independent way. In such scenario, the
with additive manufacturing to drive the transition from a
digital framework to a real context. symmetry conditions of the assumed roles distribution, or
the fact that all the actors might be not equally active, results
I. INTRODUCTION in a potential weakness of the whole SM workflow for the
The co-creation of fully customized products can be emergence of a collective production.
achieved though the exchange of data and processes between In order to achieve such customization of products
members of a community [1], [2], [3], [4]. This is the essence based on collective behavior, one particularly critic channel
of social manufacturing (SM), a novel approach where that captures the interactions within the CPSS is the one
such data is shared within a cyber-physical social space composed by the information related with product design.
(CPSS) [5], [6], triggering massive decentralized co-creation Such criticality comes not just from the above mentioned
processes. As a collective phenomena, interactions define risks regarding the possible absence of agents providing
behavior and emergent dynamics of the individuals and the such knowledge to the CPSS. But also from the intrinsic
group [7], [8]. Such interactions can take place with the difficulty to define variables that hold the essence of a
environment but also among individuals of the community, design, to represent it as information to be stored, shared and
meaning an individual or cognitive component and a social customized. To overcome such issue Artificial Intelligence
one of the information hold by the whole system. While provides different approaches. Specifically Deep Generative
in a biological framework such information driven by Design tools such as Generative Adversarial Networks [12]
physical interactions is enconded in complex biochemical (GANs) and Variational Autoencoders [13] (VAE) have been
networks [9], [10], [11], the way artificial agents interact shown as powerful frameworks to provide solutions in a
is usually sensed, digitalized and processed by hardware wide range of complexity dimensional reduction problems
and software, and through the use artificial intelligence and generation [14], [15]. And more interestingly for the
algorithms these interactions derives in artificial collective scope of this work, they have also been reported as active
behavior. Considering the digital social space defined within tools for 3D shapes generation [16] of specific products.
a SM framework, the information that quantifies such In this context. these kind of methods by training through
interactions comes from very different sources related with existing large 3D datasets such as Shapenet [17], manage
the different actors involved in a manufacturing process. to encode, in terms of distributions, the values of the most
Even in some cases, this information is embedded in the representative variables of certain family of shapes into what
actions taken by the members of the community, coming is called a latent space. And introducing variational methods
from experiences out form the CPSS. Which results in a high they trigger the generation of new models based on the
difficulty to extract information hidden in a space different information stored in the latent space. We propose the use of
from that where cyber-physical interactions take place, this information captured in such latent space to overcome
and that could be used by intelligent algorithms to drive the sparsity within the CPSS channel that holds information
production strategies. With this in mind, one can visualize in terms of product design.
the cyber-physical social space as a multidimensional one, In this work we present an active tool based on generative
with multiple channels holding information flows in terms modelling for augmented design to be developed within the
Copyright ©2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0)
Social Manufacturing community
Makers Producers Service Providers Media
Customers Designers Pure Prosumers
Customized Product
Requirements
Cyber-Physical Social Space
Constrains
Preferences
Fig. 1: Cyber-physical social space for Social Manufacturing Different roles within Social Manufacturing Frameworks are projected
into an interation space (CPSS). In real frameworks most of the actors are not pure prosumers so that many different roles appears and
interact. From the interactions different information channels can be separated and summarized in three main branches (requirements,
constrains and preferences) to drive a model for a final customized product.
european project INEDIT [18]. The main goal of this project These two layers are used to conform a set of n Gaussian
is to build a Do it Together ecosystem to demonstrate the distributions with mean and standard deviation given by the
potential of real SM approaches within Circular Economy output of the layers. These distributions are then randomly
[19]. For that the INEDIT platform brings together a sampled to conform the latent vector of the given input. In the
very wide spectrum of stakeholders around the furniture decoder, the reverse process is applied, using 3D Transposed
manufacturing providing four cross use cases: sustainable Convolutions with kernel size (4 x 4 x 4) and stride of two to
wood panels manufacturing and 3D-printing of wood as a increase the spatial dimensions along the decoder, resulting
disruptive approach [20], 3D printing of recycled plastic in a tensor of the same dimensions than the input tensor.
and ‘smartification’. Keeping the as a central principle the This architecture was trained in pytorch [26] using 3d
idea of designing globally and producing locally. The part models of chairs, taken from the dataset Shapenet [17]
of our contribution to this project that we present in this under the category ”chair”, and converted to a voxelized
paper covers the augmentation of the design capabilities of a representation of shape [32 x 32 x 32] via the library Kaolin
potential user of the INEDIT platform. For that we developed [27]. A total of 4 samples were used.
a VAE following the β-VAE approach [21], [22] developed The loss function to minimize is a compound function
for a progressive processing of the information during the between Binary Cross-Entropy and Kullback-Leibler
training process of the network. We will show how our VAE divergence [28], adjusted with a gain proportional to the
works and provide results for the creation of digital models epoch number, as shown in 1, with i being the epoch number,
of furniture. Furthermore we provide real examples of such yv the input value of each voxel and yˆv the predicted value
outcomes obtained through additive manufacturing[23]. of each voxel. With this loss, we look that, at the beginning
of the training, the focal point of the optimization is the
II. R ESULTS AND D ISCUSSION reconstruction, meaning, the weights in the convolutional
For the development of the generative model, we used layers. However, as the training advances, the loss gives
a Variational Auto Encoder structure, adapted for 3D more weight to the disentanglement of the latent variables,
convolutions, as shown in Figure 2. In the encoder section, therefore optimizing the latent representation.
each stage comprises a 3D convolutional layer, a batch X
normalization [24] and a Leaky ReLU activation [25]. The Loss = − (yv log yˆv + (1 − yi v) log(1 − ŷv ))
convolutional layer uses a kernel of (4 x 4 x 4) with a stride v∈V
of two, therefore, decreasing the spatial dimensions while n
X (1)
increasing the number of filters. The final vector of features + iβ σj2 + µ2j − log(σj ) − 1
is then reshaped into a tensor of size [n] and connected j=0
with fully connected layers to the ”mean” and ”std” layers. This training procedure was performed for a different
Fig. 2: Network architectureThe network architecture. In blue: the input kernels and in green: the output kernels The figure is so wide
in order to hold also the printed models to compare with the outcome of the digital platform.
number of dimensions for the latent space and for different provide a powerful tool for designers but also a help tool for
number of epochs. We find that the optimal point that users without any previous experience or expertise in design.
minimizes both KL-divergence and reconstruction is n = 50 The results have shown how Variational Autoenconders
and Epochs = 100. provide a robust method for creativity augmentation, which
After the training, the samples were obtained by random might be a fundamental active tool to increase the knowledge
sampling the latent space with samples from a N (0, 1) encoded in the CPSS. This works shows how Artificial
distribution. The voxelized representations were transformed Intelligence might enhance the capabilities hidden in Social
to a 3D mesh using the marching cubes algorithm [29], and Manufacturing frameworks.
the artifacts of this transformation were removed using a
Laplacian smoothing operation [30] and a re-mesh filter. R EFERENCES
Finally, the samples were 3D printed at 1:10 scale using [1] K. Ding, P. Jiang, and X. Zhang, “A framework for implementing
social manufacturing system based on customized community space
a consumer 3D printer with a wood-filled filament 1 . The configuration and organization,” Advanced Materials Research, vol.
results, while not being comparable to the fabrication of a 2450, no. 712, pp. 3191-3194, 2013.
real-scale chair, demonstrate the possibility of manufacturing [2] R. Duray, “Mass customization origins: mass or custom
manufacturing?,” International Journal of Operations Production
complex shapes with a wood-like appearance, texture, touch Management, Vol., pp. 314-328, 2002.
and smell. [3] B. D. . F. F. S. Da Silveira, G., “Mass customization: Literature
review and research directions,” International Journal of Production
III. C ONCLUSIONS Economics, 72(1), 1-13, 2001.
[4] B. Mohajeri, “Paradigm shift from current manufacturing to social
We have developed a augmented design tool based on manufacturing,” Aalto University, 2015.
Variational Autoencoders for its integration in a Social [5] F. Wang, “From social computing to social manufacturing: The coming
Manufacturing platform. Specifically we have presented industrial revolution and new frontier in cyber-physical-social space,”
Bulletin of Chinese Academy of Sciences, vol. 6, no. 1,pp.658-669,
a contextualization of our tool within the INEDIT 2012.
project framework as a co-creation landscape for furniture [6] P. Jiang, K. Ding, and J. Leng, “Towards a
manufacturing based on the Do it Together concept. The cyber-physical-socialconnected and service-oriented manufacturing
paradigm: Social manufacturing,” Manufacturing Letters, vol. 7 pp.
presented approach, adapted from β-VAE model, will 15-21, 2016.
[7] J. Parrish and W. Hamner, “Animal groups in three dimensions: How
1 FormFutura Filaments: https://www.formfutura.com/ species aggregate,” Cambridge University Press, 1997.
Fig. 3: Results Main results of the VAE at different stages. A.) First outcome of the network expressed in voxels with a 32 x 32 x 32
resolution. The figure shows the values of two different outputs from the same class (chairs). B.) Shows the results shown in A after
smoothing. C.) Shows the outcome of our VAE in real scenario. The results obtained from the algorithm are now obtained through additive
manufacturing.
[8] T. S. P. F. M. . A. G. Camerlink, I., “The influence of experience on S. Mohamed, and A. Lerchner, “beta-vae: Learning basic visual
contest assessment strategies,” Sci Rep 7, 14492, 2017. concepts with a constrained variational framework,” in ICLR, 2017.
[9] J. C. Pomerol, “Artificial intelligence and human decision making,” [23] B. Berman, “3-d printing: The new industrial revolution,” Business
Eur J Oper Res 99, 3–25, 1997. Horizons, 55 (2), pp. 155-162, 2012.
[10] K. R. W. D. M. . S. M. N. Resulaj, A., “Changes of mind in [24] S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep
decision-making,” Nature 461, 263–266, 2009. network training by reducing internal covariate shift,” arXiv preprint
[11] Y. Dayan, P. Niv, “Choice values,” Nat Neurosci 9, 987–988, 2006. arXiv:1502.03167, 2015.
[12] I. Goodfellow, J. Pouget-Abadie, B. Xu, M. Mirza, D. Warde-Farley, [25] K. He, X. Zhang, S. Ren, and J. Sun, “Delving deep into rectifiers:
S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” Surpassing human-level performance on imagenet classification,”
NIPS,2, 5, 6, 8, 2015. 2015.
[26] A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. DeVito,
[13] D. Kingma and W. M., “An introduction to variational autoencoders,”
Z. Lin, A. Desmaison, L. Antiga, and A. Lerer, “Automatic
Foundations and Trends R in Machine Learning: Vol. xx, No. xx, pp
differentiation in pytorch,” in NIPS-W, 2017.
1–18. DOI: 10.1561/XXXXXXXXX, 2019.
[27] K. J., E. Smith, J.-F. Lafleche, C. Fuji Tsang, A. Rozantsev, W. Chen,
[14] W. Wang, Z. Gan, H. Xu, R. Zhang, G. Wang, D. Shen, C. Chen, and T. Xiang, R. Lebaredian, and S. Fidler, “Kaolin: A pytorch library for
L. Carin, “Topic-guided variational auto-encoder for text generation,” accelerating 3d deep learning research,” arXiv:1911.05063, 2019.
pp. 166–177, 01 2019. [28] V. Prokhorov, E. Shareghi, Y. Li, M. T. Pilehvar, and N. Collier, “On
[15] X. Zhang, Y. Yang, S. Yuan, D. Shen, and L. Carin, “Syntax-infused the importance of the kullback-leibler divergence term in variational
variational autoencoder for text generation,” in Proceedings of the autoencoders for text generation,” 2019.
57th Annual Meeting of the Association for Computational Linguistics, [29] W. E. Lorensen and H. E. Cline, “Marching cubes: A high resolution
(Florence, Italy), pp. 2069–2078, Association for Computational 3d surface construction algorithm,” SIGGRAPH Comput. Graph.,
Linguistics, July 2019. vol. 21, p. 163–169, Aug. 1987.
[16] C. Nash and C. Williams, “The shape variational autoencoder: A deep [30] H. Badri, M. El Hassouni, and D. Aboutajdine, “Kernel-based
generative model of part-segmented 3d objects,” Computer Graphics laplacian smoothing method for 3d mesh denoising,” vol. 7340,
Forum, vol. 36, pp. 1–12, 08 2017. pp. 77–84, 06 2012.
[17] A. A. Chang, T. Funkhouser, L. Guibas, P. Hanrahan, Q. Huang,
Z. Li, S. Savarese, A. Savva, S. Song, H. Su, J. Xiao, L. Yi, IV. ACKNOWLEDGEMENTS
and Y. Yu, “Shapenet: An information-rich 3d model repository,”
arxiv:1512.03012, 2015. This research has received funding from
[18] “Eu publications office. “cordis site for inedit project”. last viewed: the European Union’s Horizon 2020
13th dec. 2019. url: https://cordis.europa.eu/project/rcn/224848,”
research and innovation programme under
[19] Y. Kalmykova, M. Sadagopan, and L. Rosado, “Circular economy
e from review of theories and practices to development of the project INEDIT with Grant Agreement
implementation tools,” Resou. Conserv. Recyc. 135, 190201, 2018. 869952.
[20] J. Kietzmann, L. Pitt, and P. Berthon, “Disruptions, decisions, and The authors want to thank the comments
destinations: enter the age of 3-d printing and additive manufacturing,”
Business Horizons 58, 209–215, 2015. and fruitful discussions with all the members of the Artificial
[21] C. Burgess, I. Higgins, A. Pal, L. Matthey, N. Watters, G. Desjardins, Intelligence Group at the Robotics and Control Department
and A. Lerchner, “Understanding disentangling in β-vae,” 04 2018. of AIMEN.
[22] I. Higgins, L. Matthey, A. Pal, C. Burgess, X. Glorot, M. M. Botvinick,