3Dprint.AI: Towards Open-Source, Large-Scale 3Dprinting Beyond Intelligence

Part I/II: Theoretical Part

This article is a part of the ongoing research of Georgios Drakontaeidis at the Bartlett School of Architecture, Research Cluster I (RCI), 2019.

Figure 1. AI matrix.


”There is more digital after the digital, either we like it or not”


(Carpo, 2018).

Abstract

At the threshold of the digital revolution in fabrication, emerging technologies such as additive manufacturing are dramatically transforming both design and making (Gershenfeld, 2015). However, the current limitations of additive manufacturing regarding cost, time and quality are limiting its future development, especially its implementation in large scale. From the abovementioned issues arises the hypothesis of this thesis, namely, the technological advances in machine learning algorithms and the open innovation in open source communities have the potential to revolutionise large-scale additive manufacturing, enabling it to shape the future of making. To test this hypothesis, this thesis introduces 3Dprint.AI, an open-source machine-learning framework for large-scale additive manufacturing, which aims to engage users and technology in the application of additive manufacturing at an architectural scale in a sustainable and affordable way.

3Dprint.AI operates in two modes, the manual and the automatic one. It involves the following steps i) video recording and object recognition while printing, ii) adaptive feedback to the robot, iii) documentation of all printing settings and results, and iv) distribution of those settings either via google sheet files-manual mode- or through an online distributed network to enable their use by machine learning algorithms to make predictions-automatic mode-.3Dprint.AI is implemented in the design and development of a high-quality extruder, which entails the integration of the above steps in manual mode . Printing tests are conducted to create data sets for use to predict the extruding outcomes and train machine learning algorithms of the community. The thesis concludes by proposing software development as an additional operation to the above open-source framework, which can save computational power and earn an income for the community.

keywords: robotic 3d printing, additive manufacturing, artificial intelligence, open-source, innovation, blockchain, large-scale manufacturing

Contents

01| Introduction

  • 1.2 Hypothesis
  • 1.1 Definitions
  • 1.3 Method
  • 1.4 Steps
  • 1.5 Intended contributions

02| Background

  • 2.1 Problem statement
  • 2.2 Historical and social background
  • 2.3 Debates on the main idea
  • 2.4 Trajectories in large scale AM, in machine learning in design- fabrication, in open source communities

03| 3Dprint.AI, an open-source machine learning framework for large-scale additive manufacturing

  • 3.1 General description
  • 3.2 Technical specification: Steps
  • 3.3 Organization of community and data
  • 3.4 Problems and challenges

04| Implementation: Manual mode

  • 4.1 The extruder
  • 4.2 Implementation steps
  • 4.3 Evaluation and further steps

05| Conclusions

  • 5.1 Contribution
  • 5.2 Concluding remarks

Bibliography

Image credits

………………………………………………………………………………………………………………………………………………………………………

1| Introduction

Figure 2. Robots, assembling line.

1.1 Definitions

‘’Machine Learning (ML) is the practice of using algorithms to parse data, learn from it, and make a determination or prediction. The machine is “trained” using large amounts of data and algorithms .’’ (Copeland, 2016).

‘’ Additive manufacturing (AM) is the process of joining materials to make objects from 3D model data, usually layer upon layer, as opposed to subtractive manufacturing methodologies.‘’ (Wohlers Associates, 2010).

Robotic AM is an additive manufacturing process using industrial robots due to their high speed, accuracy and freedom of movement in multiple axes and large scales. In this aricle, the term large-scale AM is entirely referring to robotic AM.

1.2 Hypothesis

Gershenfeld (2012) declared the coming of a new digital revolution in fabrication, while Bitonti (2016) rejected the claim stating that the full potential of technologies such as AM had barely begun to be discovered. Despite its apparent prospects, the limitations of AM are confining to a prototyping scale. Bitonti (2016) noted that even though AM has the potential for large scale application, the industry uses it only as a prototyping technology due to its limitations.

Anderson (2102) demonstrated that open innovation in both hardware and software can empower current technologies like AM in a global Maker’s framework. However, even though the computational power and algorithmic availability make the application of machine learning to AM promising, it remains a conceptual and technical challenge. To overcome these obstacles, this thesis introduces 3Dprint.AI, an open-source machine-learning framework for large-scale additive manufacturing, which can accelerate its development and minimize its limitations. The 3Dprint.AI framework has been created at a time when AM is facing numerous limitations in making the substantial advance to challenge, intensively and decisively, all other subtractive manufacturing techniques.

This thesis is motivated by the idea of bringing AM to an architectural scale by creating an online platform where users and labs may exchange data and knowledge about hardware and software for implementing AM on a large scale. However, this thesis aims not at creating the platform but at studying the challenges posed by the framework in which the platform would operate.

1.3 Method

The proposal for a machine-learning framework for large scale AM draws upon research in several related fields. Bringing together the Makers culture of DIY-ism in open hardware and software, the ma-
chine learning trends in digital design, and the use of robotics in digital fabrication, this thesis proposes a framework with the potential to raise AM to architectural scale by challenging the current state of the industry.

The analytical axis of the thesis revolves around the questions – Can open source innovation and machine learning algorithms advance large scale AM?

The critical angle of the thesis is centred on the questions of whether an open community for large scale AM is possible. Are there any ethical issues, apart from technical ones that prevent users from participating? Will users trust a blockchain network for distribution of data or the high skills required for participating in the community and yet maintain a
degree of inaccessibility to it? How will the financial needs of the community be met efficiently?

As an alternative to bibliographical research for finding answers to these questions, two polls, one at the Autodesk University conference 2019 on ‘the future of making’ was taken after the idea of this thesis was presented and discussed in detail and the other from Bartlett students. Even though the results of the poll cannot be totally objective since the limited number of participants, the responses to the poll illustrate the thoughts, of the possible members of the community.

Finally, most of the 3Dprint.AI operations were implemented manually, using an extruder developed for all the students of Research Cluster I. The creation of the community was criticised using the extruder outcomes of other clusters at the workshop. The above design of the extruder was given to other students who quickly gained the skills of spatial printing in the conducive intellectual environment and exchanged designs and data in the workshop.

Figure 3.MX3D robot printing metal.

1.4 Steps

The thesis is developed in the following steps:

Chapter 2 will frame the content of the thesis historically while debating the main points of the framework. Trajectories in large scale AM, in machine learning and in open source communities will be analysed and criticised.

Chapter 3 will introduce and the 3Dprint.AI framework in manual and automatic mode. It will analyse the organisation of community and data. It will present both technical and ethical challenges and analyse plans for its economic efficiency.

Chapter 4 will introduce the extruder, which was developed for RC1 by implementing most of the steps of the framework in the manual mode. It will evaluate the experience from the B made workshop of Bartlett where all the designs were given to the students from other clusters.

Chapter 5 will evaluate and criticise the research described in this thesis and, conclude by laying out the path for the future development of the framework.

1.5 Intended Contributions

This research intends to engage three different groups of readers. For architects – design researchers, it aims to problematize on the existing limitations of digital fabrication and illustrate the potential of creating an open-source community for applying machine learning in AM for pushing both designs and fabrication out of their current limits. For re-
searchers in artificial intelligence and programmers, this thesis aspire to inform them about the problems of AM and motivate them for the implementation of machine-learning algorithms in AM in order to
revolutionize it. For makers and thinkers, it aims to demonstrate a framework for making things in large scale using robotic AM while clearly highlighting the challenges, dangers and prospects of it.

02| Background

Figure 4. Daedalus Pavillion detail, Ai build, 2016.

2.1 Problem statement

The problem statement of the thesis is based on the problems that have persisted historically in AM implementation on a large scale; though it is continuously gaining momentum at the prototyping scale. At the same time, according to Carpo (2017), machine learning algorithms have been developed and successfully applied to many different fields from digital design to fabrication. Furthermore, Anderson (2012) has contended that the open-source communities have revolutionised codes and systems and online interaction among the users can lead to great innovations. Consequently, this thesis tackles the question of whether open innovation and AI algorithms can be applied to large scale AM and accelerate its development.

2.2 Debates on the main idea of the framework

On the one hand, it can be assumed that the computational power and the algorithmic innovation in the post-digital era can lead to great innovation using AM. However, on the other hand, Hebron (2016) has pointed out that machine learning algorithms need to be explored more intensively because, according to the researchers, they are still at a primitive stage.

Moreover, although AM tends to be a source of innovation, the cost and the complexity of the process prevent people from experimenting with it (Rayna, Striukova and Darlington, 2015). Indeed, learning, the skills required for using 3d printers or robots tend to be time-consuming and effort-intensive. Consequently, users tend to quickly lose interest in such machines, preferring not to use them. However, digital platforms can help overcome these difficulties since, most of the time, they enable multiple interactions between users to exchange knowledge and ideas on specific issues and quickly gain skills (Srnicek, 2017). Many digital companies such as Google, service start-ups such as Uber, and manufacturing industries, such as Siemens have started using platforms for informing customers about their services and products, to instruct them and make their services and products more accessible to people (Srnicek, 2017).


Furthermore, an open-source machine-learning framework for large scale AM tends to be a technologically complex and computationally intensive procedure. Therefore, the technical problems that can arise at every step in the process of its creation can make the framework non-functional and just a waste of time and energy. However, Pearson (2016) has noted that the governance inside open source communities can distribute the responsibility of resolving the technical issues among the community’s users for converting a concept into a functional model since the flexibility of open source environments enables easy and fast modifications of the system.

However, it may be assumed that the emerging ethical questions over the loss of intellectual property within the open-source community, the unreliability and corruptibility of the systems it produces, and the lack of ethical-economical rewards for the users are becoming serious impediments in attracting wider participation in the community. In addition, the possibility of the use of technology in unexpected and often illegal ways – such as unauthorised development of guns – can endanger public safety.

However, Pearson (2016) argues that in open source environments there are specific rules of hierarchy, protection of data and evaluation of users to ensure the ethical and legal aspects of the communities.

03|3Dprint.AI, an open-source, machine-learning framework for large-scale AM.

Figure 5. 3Dprint.AI.

3.1 General description

3Dprint.AI is introduced as a conceptual framework using robotic arms for printing, which technically has the potential for practical implementation. The idea of 3Dprint.AI was presented in the Autodesk University Conference on June 2019 and the poll of the lecture audience presented below was the feedback for the thesis arguments.


As an open-source framework, 3Dprint.AI aims to connect labs, institutions, companies, start-ups and even single, verified users. The worldwide existence of labs engaged in experimenting with large scale AM, in universities and industries, felt the lack of a framework focused on this technology as a hindrance in converting 3Dprint.AI into a promising
and technically feasible technology.

3Dprint.AI works in two modes – the fully automated mode using machine learning algorithms and the manual one (Figure 6). These modes are incorporated into the following steps: i) recording and object recognition process while printing, ii) adaptive feedback process, iii) documentation of printing settings and iv) saving-distributing data either by google sheets-manual mode or blockchain network-automatic mode. After saving, data can be processed for improving the settings and making predictions either by machine learning algorithms or manually through a critical study of them. Manual mode can be used at the beginning of the community when there is a small amount of data and the appropriate algorithms have not been written and tested. Like most of the open-source communities, 3Dprint.AI is flexible enough to enable addition or removal of steps as necessary for 3Dprint’s future implementation.

As described above, its real innovation is in the identification of printing errors in real-time and prediction of the future results while all the information is available to the community. Members can contribute data to the community while they can also test the predictions made by algorithms. In this manner, after many iterations, it would be possible for the community to develop a software for spatial extrusion with robotic arms, fill the existing gap in this field, and make spatial extrusion more accessible by users.

The tools needed for the framework is a robotic arm, an extruder, and a computer. In the beginning, the robotic arm can be a 6-axis robot. The extruder can be made by the community for its use to be shared by all the members. to use the same hardware, for avoiding collisions. Every step of the process is important and demands high precision settings and tooling, as every possible error can lead to different conclusions and decisions. Technical, organisational and ethical issues of this framework are discussed below.

Figure 6. The framework.

3.2 Technical specification: Steps

I) Printing record && object recognition

A camera mounted near the extruder of the robotic arm is used for recording all the processes at close quarters. In the automated mode, the camera sends all the recordings to a local computer, which uses an object recognition program to compare the printed model with the digital one. The algorithm scans every layer – lattice – of the printing model, comparing it with the photos for specific printing errors. Printing errors such as in wrapping material, over-extruding etch are commonly known to the community and are very easy to find their photos.

The camera can also be used for live streaming of the process to the robot users so that the process can be implemented manually for an experienced user to identify many extruding errors without the use of the software.

II) Adaptive feedback process

When the local computer recognises a printing error, it searches the database of possible printing errors to find a match for this error and suggest the steps to follow to rectify it. The computer sends feedback to the robot to change its g-code, which can control most of the printing processes from the robot movements to temperature controls.

Most of the times printing errors occur due to high speed or temperature errors. These parameters can be changed easily to improve the printing results significantly. Even though this step seems too complicated, it is given effect gradually during printing. The same process can be implemented manually by changing the g code of the robot either through the official operation software of the robot such as RobotStudio or by stopping the process to change the parameters of g code through the manual control of the robot.

III) Documentation of printing settings, problems and solutions

Once the printing is over, the users have to save all the settings of the process such as motor speeds, temperature etch, and the errors that occurred during printing. If they managed to rectify the errors, they must document the actions they performed to make the rectification. They also must take photos of their model, especially where errors occurred. If a user cannot fix an error, it would be documented as an unsolved issue and would be advertised to the community. As users contribute solutions, large amounts of data accumulate. For every possible error, every user may contribute a different solution, of which the optimum one may be used. The documentation may be done in an excel-google sheet document, which can be available to all the members of the community. The datasheet for every printing iteration can either be sourced automatically from the printing software or the user may create it manually.

IV) Saving and distributing data to the community

After documentation, users can use the blockchain network of the community to distribute their data (Figure 36). Through machine learning, the existing data can be analysed to determine precisely what further tests are needed to make the predictive model more confident (Zelinski, 2018). However, training a machine learning system tends to be a highly computationally intensive process, which is often impractical or even impossible to perform on a single consumer-grade machine (Hebron, 2016). For this reason, some users of the community would be designated to implement all the computational calculations to provide ready solutions to the users.

Join our weekly newsletter to receive:

  1. Latest articles & interviews
  2. AI events: updates, free passes and discount codes
  3. Opportunities to join AI Time Journal initiatives

As at the beginning of the community, the available data can be shared even via google sheet documents and some experienced users may decide which parameters were causing errors and propose further tests to other members of the community (manual mode).

Figure 7. Private blockchain network of 3Dprint.AI community.

3.3 Organization of the community and data

3.3.1 Governance & hierarchy

Pearson (2016) showed that both governance and hierarchy are essential in open source communities, whose organisation, though democratic in many ways, does not vary much from traditional software Development Company. In the 3Dprint.AI framework, the above characteristics tend to be crucial because evaluation and confirmation of the data submitted by the users before adding it to the database is vital for the reliability of the system. Furthermore, governance is necessary for distributing specific roles and responsibilities among the users when organising new projects, solving issues and answering questions asked by the community, and even defeating virus attacks and malware.

Srnicek (2017) has demonstrated that such open source communities, even though they have governance and rules; enable the users to act in unexpected ways. Indeed, Pearson (2016) has criticised that no matter when and how users interact and participate in online communities, it is the hierarchy which makes the communities flexible and facilitated its divisions into many sub-systems as the number of users increases.

According to Pearson (2016), open-source communities have a tight vertical hierarchy structure. At the base of the pyramid are the contributors who develop a small part of work, followed by the maintainers who organise the larger parts of the project and evaluate the work by the contributors, and at the top are the project leaders who collect and assemble all the different parts of the projects (Pearson, 2016).


At least one maintainer will be required to initiate 3Dprint.AI. The maintainer will begin by inviting more users, and as the community expands, the maintainer would become a project leader while other users become maintainers and contributors. Every new user would be a contributor and deliver work to the maintainer. The user must prove technical skills, willingness and responsibility to the community to become a maintainer (Figure 35).

Project leaders would be responsible for 1) the evaluation of their contributors, 2) strategic decision-making about the orientation of the future research of their group. For example, they will decide which printing materials they will test, which extrusion method etch. The main roles of maintainers would be: i) to evaluate the slicing settings that the con-
tributors send and submit them to the network, ii) to advertise to the community new printing problems needing solutions according to their leader’s plan, iii) to divide work of conducting specific tests among the contributors. 3Dprint.AI contributors will i) run printing tests and record slicing settings, ii) discuss with other contributors or maintainers the issues that they face while printing, iii) feed the community with feasible creative ideas about various upgrades to hardware and software. In 3Dprint.AI community, reputation would be gained through contribution and leadership through both contribution and experience (Pearson, 2016).

Figure 8. Hierarchy of 3Dprint.AI community.

3.3.2 Data distribution via a blockchain network

According to Rosic (2019), centralised datasets used to be easy targets for hackers who wanted to either steal valuable information or just corrupt the system. Laurence (2017) noted that blockchain networks tend to be impenetrable to attacks and sensitive user information could be protected with cryptographic algorithms.


In the 3Dprint.AI framework, a private blockchain network is used. Indeed, every data sheet sent by the contributors and evaluated by maintainers would be stored and available to everyone inside the community to see the data. However, no one would be able to change the data once it is stored in the blockchain. As a result, machine-learning algorithms developed by the community may be used by the community to run multiple iterations and make predictions. One or more contributors can then test the predictions and if proved right, the maintainers may add them to the blockchain network. Consequently, the data added to the blockchain may be relied on as valid, safe and used by the community.


The data in the blockchain, for better organisation and easier access, would be classified into three distinct groups (Figure 9), the first, the documentation of the printing settings and the errors occurring while extruding. The second, predictions made by machine learning algorithms. The third, the predictions of experienced users implemented manually. All groups would be advertised to the community and every group will be categorised into subgroups according to the printing error they refer to.

3.3.3 Categorization of data

The categorisation of the data within the above three groups would be based on some important printing parameters such as speed, the complexity of geometry, printing temperature, material, calibration, and multiple errors. Machine learning algorithms would be used only with the information in the first two groups as the third group contains only advice from shared by the users. The above categorisation would help the better organisation of data to make it easily comprehensible to the learning system (Hebron, 2016). While 3Dprint.AI grows more categories may be added and some others deleted or excluded from use due to obsolescence.


Figure 9. Categorization of data of the community.

Another important use of categories of data is in analytics, which can reveal trends, relationships between errors or even make predictions to define the future direction of research for the community. Making groups and subgroups will assist in making better distribution of human resource for achieving specific goals and tasks of the community. Data analytics can illustrate printing errors and their interrelation to facilitate manual and algorithmic predictions.

3.4. Problems and challenges

Figure 10. Robot thinking.

3.4.1 Technical

I. Proper dataset

The main problem with 3Dprint.AI is the use of the training data. Hebron (2016) averred that the quality of a given dataset relates to the characteristics of completeness, accuracy, consistency and timeliness.
Indeed, datasets, which are not complete, accurate, consistent and contemporary or recent within a certain time framework may lead to inappropriate decisions because of large amounts of outdated information proceeding. In AM, often there is more than one reason causing a problem and it is very challenging even for experienced users to identify all of them. As a result, instead of solving one issue, the datasets could create more, causing more failures in the printing process. Hence, Hebron (2016) noted that an unreliable dataset is worse than no dataset, as computational power is wasted. However, this problem can be minimised by several iterations of the error until the best solution is found and documented in the online database after careful evaluation of the data provided by the contributors.

II. Change of g-code

The changing of the g-code of an industrial robot while it is in operation can be very challenging because any mistake can greatly damage its mechanism. More specifically, every single movement of a 6-axis robotic arm consists of many small sub-movements along its every axis. To avoid such damage and further failures in its environment, every change in these movements should be within the safe axis limits of the robot. However, the existing software for robots such as RobotStudio or even the manual controllers of the current robots is equipped for such functions. At the same time, the parameters of the extruder such as temperature and stepper motor speed are independent of the robot’s parameters such as robot speed, which makes the making of the changes easy and safe. Thus, at the beginning of the community, the goal of the framework may be set as maintaining the robot at a steady speed while all the other independent parameters of the extruder may be changed to achieve optimum results with minimum risk.

III.Proper documentation

The proper documentation of all settings is a vital concern because it can affect the predictions of machine learning algorithms. The manual docu-
mentation through excel files should be validated in order to be useful to the community. However, at the later stages of development of the community, the documentation, at least the part regarding the printing parameters can be automatically exported to a maintainer of the community from the printing software. To avoid confusion and errors it is necessary that the instructions from the maintainers to the contributors are focused. Indeed, contributors can change printing parameters inside a specific range and not test random changes. Also, the same tests may be implemented by different contributors for checking the parameters submitted to the maintainers. The comparison of printing outcomes with the same parameters produced by different robots can also be used for testing the reliability of the contributor’s hardware.

IV. Same robot & extruder

It is possible that the makers in the community may use different robots, build different extruders and use different slicing software, which would make the provision a universal solution would be very challenging or impossible. In this step, many errors may occur since the difference in the hardware can mislead the makers. To avoid these problems, 3Dprint.AI must initially specify one model of a robot, one extruder, and one specific extruding material so that a safe solution can be proposed to the users. Later, when the community is ready with a larger amount of data for the use in its algorithms, a variety of hardware may be added to provide the users with a wider choice.

V. Environmental conditions while printing

According to Leonidou (2019), one of the most important factors regarding printing quality and consistency is the environmental conditions in which the printer operates, that is to say, users with the same machine, extruder and slicing settings can face different failures –ex. bad adhesion; if they op-
operate the equipment under different environmental conditions such as different temperatures. This issue can be solved through extruding under almost the same environmental conditions controlled inside an enclosure or room and the specifics of the environment to the documentation.

VI. Complexity of the process

The complexity of the computationally intensive process remains a crucial issue in the framework. In desktop 3d printing, where the extruding plate is most of the times up to 30cm, almost all the printing problems such material collision, layer separation etch can be solved through proper slicing of the 3d model and excellent machining. However, Leonidou
(2019) demonstrated that in large scale extrusions various other parameters such environmental conditions tend to affect much the printing outcome.

At the same time, on large scales, the cost, time and energy consumed are much more important. However, although camera recording and object recognition while printing is computationally intensive, they provide the advantage of continuous supervision over the process. However, in geometries which are simple or easy to extrude, users can just not use camera recording and adaptive feedback process but merely document the setting- errors that occur in order to produce training data for machine learning algorithms developed by the community.

3.4.2 Ethical

Figure 11. AI Ethics.

Despite the technical challenges, there are many ethical issues that emerge from the operation of 3Dprint.AI community. Most important questions are critically discussed below and two polls are presented (Figure 12). More specifically, after an issue with the Autodesk app during the poll in the Conference, only 10 answers out of 80 were recorded, making the result not reliable. For this reason, a second poll with Bartlett students was also held and received 70 answers. Even though both polls cannot be totally objective due to a limited number of participants, they can illustrate the thoughts and arguments of people who can join 3Dprint.AI community in the future.


As an overall trend, both polls illustrate the positive feedback towards 3Dprint.AI community with yes or maybe yes to be the main answers. However, a per cent of almost 10 % of feedback is maybe no or no. These arguments are taken into consideration in the discussions below.

Figure 12.Polls answers. Generally, answers are supportive of the idea of 3Dprint.AI community.

I. Authorship and ownership issues

With the development of the digital culture, such buzzwords as ‘collaboration’, ‘open source’, and ‘open innovation’ have gained currency (Picon2016). These buzzwords imply such fundamental issues as authorship. According to Rui (2016), the fundamental requirement for claiming authorship is originality. For an authored work to receive rights
of property through copyright it should be original, otherwise, it would merely be a copy of some other original work (Rui,2016). However, in a framework such as open 3Dprint.AI, it is unclear who has the authorship of data and who has the authority to use that data. In other words, ownership is uncertain and therefore is becoming a more and more central question (Fok and Picon, 2016).

Rui (2016) noted that the basis of all copyright law is the idea that authorship produces a proprietary right over the authored work. Rui has also criticised the remarkable idea of the copyright that only the creator of a work owns the exclusive right for its duplication. Furthermore, he has demonstrated that even literary critics such as Roland Barthes and Harold Bloom have powerfully argued that it is impossible to assert a claim of full originality of any creative work. Nevertheless, in the 21st century, many types of ownership over collaborative-creative work has become possible using digital technologies and humans are just beginning to foresee the consequences of this trend (Picon, 2016).

Consequently, in open 3Dprint.AI framework, when many users contribute to the creation of training dataset, in order to avoid legal disputes and arguments, all users have to accept the pre-defined terms and conditions. Indeed, they have to agree about the authority and use of the dataset; the full authorship of which can be reposed, since the beginning, with a university or a research institute such as Bartlett. However, according to Pearson (2016), when the community has gained popularity, it can be converted into an organisation with a specific legal structure which devolves all the rights and obligations upon its members.

II.Access restrictions

According to Gershenfeld (2012), in digital fabrication, there is always the threat of theft of intellectual property when makers have the right to replicate designs. Anderson (2012) has argued that the problem of copies in open source environments could be creative because when it is happening, it is a sign of success, as what starts from copying may become a real innovation after improvements. However, Gershenfeld (2012) has clarified that only some designs should be shared with online communities while others, including patents, should be protected.

Another important threat arising from free access in AM tools is the production of guns (Gershenfeld, 2012). Indeed, the democratisation of design and fabrication raises vital questions regarding free access to all to such tools. More specifically, humans tend to think and decide about technology in their individual ways. Therefore, when the technology is artificial intelligence, applied to large scale AM, it is axiomatic that access to it should be restricted only to verified users for the protection of public safety.

Consequently, 3Dprint.AI can start operating and allow access to its dataset to universities, research institutes, industries, corporations, start-ups, fab labs and other selected and verified users. It cannot be an open-source framework data that is open to everyone because all the possible negative consequences of its use for individuals and society have to be avoided.

III. Trust to the system

According to Schneier (2019), blockchain has failed to achieve adoption because businesses who are always protective of their sensitive data do not trust it. On the other hand, Anderson (2012) has argued that individuals and companies like Google are participating in open source platforms, sharing their data and trusting these communities because they aspire to meet more people, and accelerate the innovation the process to a speed far greater than can be achieved by conventional development.

Furthermore, Schneier (2019) has criticised that, in blockchain networks, trust is shifting from humans to technology and if it is hacked, there is no recourse for the users who lose all their data. However, Laurence (2017) has shown that the cost of high computational power needed for an attack on a decentralised blockchain network defeats the purpose of hacking it. In fact, at the time of writing this – 2019 –a whole blockchain network has never been hacked successfully.

Moreover, Schneier (2019) noted that no matter how reliable a technological system – blockchain –is, there are always dangers and threats, especially when it is scaled up. For this reason, many companies are using only the blockchain that they have created and control. On the contrary, Laurence (2017) noted that blockchain developers have invented tools and subsystems which afford extra safety to data when the network is scaling up. Laurence argued that these safety systems have made existing blockchain networks even more difficult to hack or interrupt, particularly when these networks are private.

Therefore, relying on the above experts’ opinions and poll results, the 3Dprint.AI framework would use a private blockchain network to secure its data against hackers while allowing access to community members. In this way, even companies and individuals whose data are sensitive could participate in the community with assurance that only members have access to the data and hackers cannot steal it.

IV. Participation of the community

Anderson (2012) gave numerous reasons for individuals and organisations to participate in open source communities, such as gaining new skills and reputation through them. On the other hand, Carpo (2017) argued that even though there many open-source environments, mass collaboration has not happened in design as yet because design professionals have rejected this potential and favoured protecting their privileges. However, Anderson (2012) described the maker movement as a trend, which will evolve in the coming years and engage more people in open source environments.


In the 3Dprint.AI framework, the participation of the community is a crucial issue and the different groups of people who would be expected to engage should be examined separately for their specific characteristics. In Anderson’s opinion (2012), it may be assumed that the interest of both makers and companies to advance large scale AM, the significant number of labs worldwide and the current open-source trend together can create the momentum by which at least some makers or labs would be motivated to participate in 3Dprint.AI community.

However, Inaccessibility of 3D.print.AI for the following reasons is becoming a major argument. Firstly, it is both difficult and time consuming for people to engage with robots for large-scale AM. Secondly, someone has to teach prospective members the basic skills and enable to further develop those skills since most of the users may be unfamiliar with industrial robots.


For these reasons, awards and payments, either ethical or financial, could motivate makers who are highly skilled, to share their knowledge with other members of the community and build mutual trust with them. The hierarchy of users can contribute towards this goal as maintainers who could distribute tasks, teach the contributors to do the tasks and supervise their work. Also, the high-quality skills that members may gain in the community can attract users to join.


Companies and start-ups can also join as they can quickly gain knowledge, they can meet new people even develop some of their projects in the community. The private blockchain network can guarantee the security of their data. Companies would continue to participate to benefit from the freely available and growing set of hardware and software. They can
use it to solve their printing issues, for which, otherwise, they have to pay.

3.4.3. Economic efficiency of the community

According to Pearson (2016), when open source communities amass large memberships, wish to protect their property and to survive economically convert into organisations like Linux. By this way, they can receive money from donors while additive advisory committees ensure that the community stays vibrant. Pearson (2016) has found that advisory committees guide communities on the technical, social and political issues and examine these issues to prevent them from becoming problems. Kumar
( 2015) has shown that open source communities earn funds from companies that are interested in their research or by providing support to companies outside the community or by selling their products or from advertisers on their website or apps. Pearson (2016) observed that such organizations in the USA enjoy low tax rates to ensure their existence.

As mentioned before, 3Dprint.AI can start its operation in an institution like the Bartlett and connect labs, companies, start-ups; and users. It can receive money by all the above means. The economic efficiency of this framework is crucial as it develops both hardware-extruder and software about large scale AM. It would need some initial investment to begin its operation, building the extruders and pay some users who spend time in the community for organising projects or teaching the contributors AM skills. This fund may be raised by crowdfunding or by research grants from institutions or, even, by self-funding by its first users. Finance is a crucial issue which can affect the number of people who will engage.

After some members are enrolled and a reliable data set is created, members of the community can develop software for slicing 3d models and strategizing toolpaths like the Cura or AIsync, which then can be sold to companies, industries, and labs who do not participate to earn an income for the community. Kumar (2016) demonstrated that even big companies like Microsoft and Google find it less expensive to pay open source communities like Linux for creating codes than creating the codes themselves. This can happen with the 3Dprint.AI by communicating with companies and industries in the field of their future research agenda and recent developments in AM and AI.

Figure 13. General diagramme which illustrates the feedback the idea of the framework and the framework itself.


Next part (II/II): Implementation of the theory in robots, research contributions and future plans.


to be continued..

AI+3Dprinting+opensource=3Dprint.AI

Bibliography

  • Ai-build (2019). Ai Build – Technology. [online] Available at: https://ai-build.com/technology.html [Ac- cessed 2 May 2019].
  • Anderson, C. (2012). Makers: The new industrial revolution. 1st ed. New York: Crown Business,p.83,101,102.
  • Autodeskresearch.com. (2019). Project Dreamcatcher. [online] Available at: https://autodeskresearch.com/projects/Dreamcatcher [Accessed 19 May 2019].
  • Bitonti, F. (2016). When Matter Becomes Media. In: A.Picon and W. Fok, ed., Digital Property: Open-Source Architecture, AD Volume 86, Issue 5. Hoboken: John Wiley & Sons, Inc, pp.101-104.
  • Brasseur, V. (2018). Forge Your Future with Open Source. 1st ed. Raleigh, NC: Pragmatic Bookshelf, pp.23-43.Carpo, M. (2017). The Second Digital Turn: Design Beyond Intelligence. Cambridge&London: MIT Press,pp 6, 132.
  • Carpo, M. (2017). The Second Digital Turn: Design Beyond Intelligence. Cambridge&London: MIT Press,pp 6, 132.
  • Carpo, M. (2018). The Post-Digital Will Be Even More Digital, Says Mario Carpo. Available from: https://www.metropolismag.com/ideas/post-digital-will-be-more-digital/ [Accessed: 7th July 2018]
  • Copeland, M. (2016). The Difference Between AI, Machine Learning, and Deep Learning? | NVIDIA Blog. [online] The Official NVidia blog, Available at https://blogs.nvidia.com/blog/2016/07/29/whats-dif-ference-artificial-intelligence-machine-learn-
  • ing-deep-learning-ai/ [Accessed 2 May 2019].
  • Fablabhouse.com. (2019). FabLab House. [online]Available at: http://www.fablabhouse.com/en [Accessed 4 Jul. 2019].
  • Gershenfeld, N. (2012). ’How to Make Almost Anything: The Digital Fabrication Revolution’, ForeignAffairs, pp. 43-45,55-57.
  • Gershenfeld, N. A. (2015) Fab: The Coming Revolution on Your Desktop—from Personal Computers to Personal Fabrication. Basic Books, pp 3-17.
  • Hebron, P. (2016). Machine Learning for Designers. 1st ed. Sebastopol: O’Reilly Media, Inc.,p.7,21,47,48,51,55,59.
  • Hermann, M., Pentek, T. and Otto, B. (2016). Design Principles for Industrie 4.0 Scenarios. In: 49th International Conference on System Sciences (HICSS).Washington DC,: IEE computer society, pp.3928-3935.
  • Johnson, J. (2019). Rocket Plan: How 3-D Printing Is Unlocking A New Space Race. [online] Forbes.Available at: https://www.forbes.com/sites/jenniferjohnson/2018/05/16/rocket-plan-how-3-d-printing-is-unlocking-a-new-space-race/ [Accessed 15 Jun.2019].
  • Keating, S. (2014). ‘Beyond 3D Printing: The New Dimensions of Additive Fabrication.’ In Follett, Jonathan (Ed.), Designing for Emerging Technologies: UX for Genomics, Robotics, and the Internet of Things(379-405). O’Reilly Media, pp 380,383-388,393,401.
  • Kumar, A. (2015). How do Open Source Companies, Do programmers make money? [online] Microsoft: TheWindows Club. Available at: https://www.thewindowsclub.com/open-source-companies-programmers make-money [Accessed 8 Jun. 2019].
  • Lansard, M. (2019). The best 3D printing forums, Facebook groups and communities in 2019. [online] Aniwaa. Available at: https://www.aniwaa.com/best-3d-printing-3d-scanning-forums-facebook-groups-and-communities/ [Accessed 1 Jul. 2019].
  • Laurence, T. (2017). Blockchain For Dummies. 1st ed.Hoboken: John Wiley & Sons, Inc., pp.19,21.
  • Leonidou, L. (2019). Engineering the AiCell. [Blog] AiBuild TechBlog. Available at: https://medium.com/aibuild-techblog/engineering-the-aicell-e65a4deb-0f3c [Accessed 19 May 2019].
  • Matejka, J., Glueck, M., Bradner, E., Hashemi, A., Gross-man, T. and Fitzmaurice, G. (2018). Dream Lens: Exploration and Visualization of Large-Scale GenerativeDesign Datasets. In: 2018 CHI Conference.Toronto: Autodesk Research, p.3.
  • MX3D. (2019). MX3D Bridge. [Online] Available at: https://mx3d.com/projects/bridge-2/ [Accessed 15 Jun. 2019]
  • Pearson, B. (2016). Common Characteristics of an Open Source Community – Open Source Today. [online] Open Source Today organization. Available at:http://opensourcetoday.org/common-characteristics-open-source-community/ [Accessed 26 May2019].
  • Picon, A. & Fok, W. (eds.) (2016). Digital Property: Open-Source Architecture. AD Volume 86, Issue 5.
  • Hoboken: John Wiley & Sons, Inc, pp 7,8.
  • Picon, A. (2016). From Authorship To Ownership. In: A. Picon and W. Fok, ed., Digital Property: Open Source Architecture, AD Volume 86, Issue 5. Hoboken: John Wiley & Sons, Inc, pp.37, 38.
  • Rayna, T., Striukova, L. and Darlington, J. (2015). Co-creation and user innovation: The role of online 3D printing platforms. Journal of Engineering and Technology Management, 37(0923-4748), pp.90-102.
  • Relativity Space. (2019). Relativity Space | Stargate —Relativity Space. [online] Available at: https://www.relativityspace.com/stargate [Accessed 1 Jul. 2019].
  • Rosic, A. (2019). What is Blockchain Technology? A Step-by-Step Guide For Beginners. [online] Block-geeks. Available at: https://blockgeeks.com/guides/what-is-blockchain-technology/ [Accessed 26 May 2019].
  • Rui, D. (2016). Serving, Owning, Authoring. In: A. Picon and W. Fok, ed., Digital Property: Open-Source Architecture, AD Volume 86, Issue 5. Hoboken: John Wiley & Sons, Inc, pp.18-21.
  • Schneier, B. (2019). There’s No Good Reason to Trust Blockchain Technology. [online] WIRED. Available at: https://www.wired.com/story/theres-no-good-reason-to-trust-blockchain-technology/ [Accessed 8,Jun. 2019].
  • Srnicek, N. (2017). Platform capitalism. 1st ed. Cambridge, UK: Polity Press, p.46,47.
  • Tamke, M., Nicholas, P. and Zwierzycki, M. (2018) ‘Machine learning for architectural design: Practices and infrastructure’, International Journal of Architectural Computing, 16(2), pp. 123–143.doi: 10.1177/1478077118778580.
  • The Bartlett School of Architecture. (2019). Architectural Design MArch. [online] Available at: https:// www.ucl.ac.uk/bartlett/architecture/programmes/postgraduate/march-architectural-design [Accessed 8 Jul. 2019].
  • Wohlers Associates (2019). What is Additive Manufacturing? | Wohlers Associates. [online] Available at: https://wohlersassociates.com/additive-manufacturing.html [Accessed 2 May 2019].
  • Zelinski, P. (2018). Where AM Meets AI. Additive manufacturing, Volume 7, Issue 1, p.25.
  • Chaillou, S. (2019). The Advent of Architectural AI, A Historical Perspective. Boston: Harvard Graduate School of Design, pp.2-15.
  • Negroponte, N. (1972). The architecture machine.Cambridge Mass.: The MIT Press, pp.1-20.

Image credits

  • Figure 1. Author: graphic design, image source:[online] http://hdqwalls.com/matrix-5k-wallpaper [Accessed 2 May. 2019].
  • Figure 2. Image source: [online] https://www.daytranslations.com/blog/2019/01/10-trends-that-will-change-the- manufacturing-industry-13371/[Accessed 2 May. 2019].
  • Figure 3. Image source: [online] https://mx3d.com/projects/bridge-2/ [Accessed 2 May. 2019].
  • Figure 4. Image source: [online] https://ai-build.com/
  • daedalus.html [Accessed 2 May. 2019].
  • Figure 5. Author: graphic design, image source: [online]http://heranet.info/projects/hera-2009-humanities-as-a-source-of-creativity-and-innovation/electronic-litera-ture-as-a-model-of-creativity-and-innovation-in-practice/ [Accessed 5 July. 2019].
  • Figure 6. Author.
  • Figure 7. Author.
  • Figure 8. Author.
  • Figure 9. Author.
  • Figure 10. Image source: [online] http://www.gizmodo.cz/?tag=ethics [Accessed 5 July. 2019].
  • Figure 11. Image source: [online] https://emerj.com/ethics-and-regulatory/establishing-ai-ethics-public-and-private-sector/
    [Accessed 5 July. 2019].
  • Figure 12. Author : graphic design, polls:[online] https:// live.eventbase.com/polls?event=autodeskuniversi ty2019&session=1180201908 [Accessed 11 July. 2019]. && [online] https://docs.google.com/spreadsheets/d/1oFLQycmy1ptX1Mh8v76-9jrw_Dgfb2mk3uqrBNJjUnI/ edit?usp=sharing
  • Figure 13. Author.

Opinions expressed by AI Time Journal contributors are their own.

About Georgios Drakontaeidis

Contributor Georgios is a postgraduate student at the Bartlett School of Architecture in March Architectural design programme (2018-2019) and scholar of Lilian Voudouri Foundation. He is a member of Research Cluster I of Bartlett, which explores automated design and fabrication systems, with a focus on robotic 3D printing technologies and their application on an architectural scale. Georgios also holds a diploma in Architectural Engineering from N.T.U of Athens where he graduated achieving a first honours award. In July 2018 he submitted his diploma project which was the first 3D-printing project at his university. Throughout his undergraduate studies, he received a number of significant awards and scholarships. He participates in international conferences and architectural workshops.

View all posts by Georgios Drakontaeidis →