Want to write a piece for 3DHEALS Expert Corner? Email us: email@example.com
There are several main reasons that frequently motivate the innovators:
- Do cool things that could not be done before (e.g. flying, electricity, etc.).
- Make life better by a magnitude of a million times, etc. and not just minor increments (e.g. discovery of antibiotics).
- Save time, labor, and money that would recreate the industrial revolution and new economies.
While the general public is fascinated with both artificial intelligence and 3D printing as powerful new technological tools, and their potential future impact in healthcare, there has not been any known “killer applications” that utilize AI to improve existing 3D printing applications, in or out of healthcare/life sciences. The easy answer could be that both technologies are still relatively new, or that people who focus on AI applications are not necessarily interested in 3D printing, and vice versa. Or, maybe it’s because we simply do not have enough solutions to problems at hand.
Some of the proposed ways AI can improve 3D printing include the following [1-7]:
- Improve prefabrication design process
- Defect/Failure Detection
- Real-Time 3D printing Control/Failure compensation
- Predictive Maintenance/Inventory
- Workflow (Cost) optimization
- Chemical reaction/photopolymerization using ML-based algorithm to maximize control (chemicals and energy input)
There is an interesting analogy that I came across from professor Hiroya Tanaka,  with the following image(Figure 1). This shows that the subject “3D printing” has the visible physical components (tip of the iceberg) and the much larger invisible components in the realm of software, including data science, advanced 3D modeling, 3D object storage and retrieval, and AI/ML/Deep learning. While this is in accordance with the belief that “software eats the world” by the Silicon Valley, I would argue that all of these components will be equally important to the achieve the theoretical promises 3D printing as a successful manufacturing alternative.
That said, it is still helpful to do a brief review of where we are in terms of the intersection of these two technologies. Hopefully, this article can inspire interesting discussions, and even better, some new startups that Pitch3D can host very soon.
Artificial Intelligence/Machine Learning/Deep Learning
Artificial intelligence is an “intelligence” that is demonstrated by machines, which can perceive its environment and take actions to maximize its chance of success through the “learning” and “problem-solving” process. Machine learning is the scientific study of algorithms and statistical models that computers use to perform a specific task without human instructions, relying on patterns and inference instead. There are unsupervised ML (no human input) and supervised ML (human input). Finally, deep learning, also known as hierarchical learning, is based on artificial neural networks. There are also supervised and unsupervised DL.
The relationships among the concepts of artificial intelligence, machine learning, and deep learning (using artificial neural networks) are best demonstrated in the following diagram. (There are more sub-categories within each of these concepts that interested readers can easily find on the internet.)
It is my theory that inventors can be lucky, but the inventions are never accidental. Inventions that changed human history (e.g. robots, computers, 3D printers, microbiology) are results of the continuous search for answers over long periods of time, from different perspectives and angles, and sometimes only after thousands of years.
The current status of healthcare applications using 3D printing is not so favorable because of several reasons:
- 3D printing is still expensive, not just from the hardware and material cost, but also labor cost, and waste due to print defects and failures.
- Lack of efficient and affordable design software. This is, in particular, a problem for the healthcare sector.
- 3D printing is unable to achieve affordable (customized) mass production due to workflow challenges.
- Lack of good quality control processes and tools, especially for the heavily regulated healthcare sectors.
The list can go on.
However, challenges also present opportunities, and AI/ML seem to be potential solutions to these worthy problems because AI/ML do somethings better than humans in many ways:
- Computers are able to process large amounts of data, learn, and implement actions in a more consistent fashion.
- Computers require little resources to function (i.e. electricity, minimal to no need for human operation).
- Computers can function well even in a toxic or harsh environment. (e.g. high temperature, toxic fumes)
- “Skillset” (algorithms) can be more rapidly “learned” and disseminated in a consistent way than human learning.
- Computers can store and retrieve large amounts of information almost instantaneously.
That said, creating the right AI/ML algorithm to 3D printing is no easy task because of the following:
- Successful AI/MI for the 3D printing process requires extensive knowledge of the specific 3D printing technologies, including but not limited to the design process, control of machine components, material science, post-processing. For example, the strategies behind optimizing the SLA based 3D printing process  will be very different from laser sintering metal 3D printing. 
- Finding high-value problems based on the end goal of production. Either it is focused on reducing wasted time or precious materials, or ensuring end product mechanical properties that could result in serious clinical outcomes. 
- Data collection. For example, for 3D printed anatomical models, a good AI/ML product focusing on optimizing the segmentation process will significantly decrease the bottleneck effect of entering the field for many hospitals and clinics. However, the lack of such a product is because of a lack of enough training datasets. 
- Intrinsic limitations of existing monitoring systems. Researchers are currently using either photos or videos to train their AI/ML algorithms. Smoothly incorporating the monitoring systems without interrupting the printing process will be challenging. [1, 4, 5] However, such integration will be required to achieve “real-time” 3D printing monitoring and subsequent “fixing” or “failure compensation” of the prints. 
- Forming a successful team that can tackle problems along the entire 3D printing process from design to final product requires a group of people from different disciplines.  For example, to accomplish real-time SLA 3D printing support modification[Figure 3], Dr. Iuganson proposed in his thesis a team structure that would include the following:
- 3D printing engineer
- Sensors technician
- Automation CT engineer
- Laser and optics engineer
- Machine learning specialist develops a set of steps for correction of the printing and generating supports if the problem is predicted.
- Data scientist creates a code for the machine to change the design structure and generated supports
- AI research scientist analyses and implements the information in the AI system to add a new feature of real-time control over the design and supports.
Now, imagine that everyone on this team has to understand what is going on and can also communicate effectively with one another!
Solutions seem to be coming, but just not here yet.
GE Additive, Sculpteo, Autodesk, and many more all appear to actively develop AI/ML-based solutions to optimize various value points of the 3D printing process.  Align Technology just announced a new AI/ML-based visualization/predictive tool SmileView based on 60 million patient datasets. (Align is also actively hiring AI/ML engineers.)  It is my hope that perhaps more entrepreneurs can venture into this exciting intersection of two powerful emerging technologies.
Perhaps this IS where we will find the “killer app” in 3D printing.
- Artificial Intelligence in 3D Printing (Thesis by Dr. Reino Iuganson)
- Deep Learning for Advanced 3D Printing
- The potential for machine learning algorithms to improve and reduce the cost of 3-dimensional printing for surgical planning (Trevor J. Huff, Parker E. Ludwig & Jorge M. Zuniga) ISSN: 1743-4440 (Print) 1745-2422 (Online) Journal homepage: https://www.tandfonline.com/loi/ierd20
- Machine Learning “Fixes” 3D-Printed Metal Parts—Before They’re Built
- Automated Process Monitoring in 3D Printing Using Supervised Machine Learning
- Artificial Intelligence Applications in Additive Manufacturing (3D Printing)
- Multi-Objective Optimization Design through Machine Learning for Drop-on-Demand Bioprinting
- New Dental Product: SmileView from Align Technology