By: Ralph Zoontjens
As companies are getting the hard work done in times of digital transformation, artificial intelligence is one of the key enabling technologies. And while 3D printing is starting to digitize the manufacturing industry, it can be augmented with sophisticated computational resources as well. The field of AI has numerous implications, some more surprising than others.
AI is the culmination of today’s abilities for computers to solve complex problems where humans do not suffice. Some forms of AI emulate the human “wetware” brain, while others result in a form of thinking more alien to humans, making the entire concept difficult to grasp.
Fundamentally, until we reach the age of quantum computing, microprocessors can only see a world of data and are completely blind to the actual world. So the challenge for companies is to make it relevant and meaningful in conjunction with their own way of thinking. The best-known branches of AI are:
- Machine Learning. With machine learning (ML), systems are programmed to train themselves based on input datasets with the purpose of detecting relationships that humans cannot. It is a form of unsupervised learning, meaning that the expected ways and outcomes are not predetermined; the computer is on its own.
- Neural Networks. This is a form of learning where software emulates the organic brain. The neural network (NN) system experiences stimuli going through a network of nodes that assign numerical weights in order to learn the best behavior. Compare it to a human brain that learns to shift weights more in favor of beneficial foods rather than sweet foods as time progresses.
- Deep Learning. Computers use different layers of processing data in order to dive deeper and deeper into the complexity of a problem. Instead of finding the right approach to a certain task, they learn to understand the whole organization of the reality they are faced with.
- Reinforcement Learning. Reinforcement learning (RL) is a form of supervised learning where computers are set to learn a certain desired result, for example, identifying faces or winning games of Monopoly. Program functions called agents run certain functions that they adapt based on the rewarding feedback they get toward achieving the result. This ultimately creates a kind of artificial person that can perform a task toward absolute artificial perfection. In a related form called genetic algorithms (GA), the underlying mechanism mimics the workings of genes that mutate, cross over, and adapt to be most beneficial in a certain environment.
- Edge AI. Edge computing is a system architecture where data-collecting devices are installed on the “edge” of a network—as close to the source of data as possible. For example, with the rise of the internet of things (IoT), sensors installed onto machinery fuel network analysis in order to build collective knowledge that in turn can be used to optimize processes toward peak efficiency levels.
- Conversational AI. Robots and machines of the future can be interacted with using plain speech. The algorithms are smart enough to learn commands not installed in the robot’s original programming.
Printing on the edge
3D printing has been all the hype for its unlimited possibilities for decades, literally freeing manufacturing from the mold. While injection molding is great for quick line production of high volumes of identical parts, the promise of 3D printing includes customization and being free from mold release directions, which allows for the wildest and most optimal of form geometries, as well as on-demand manufacturing.
The necessary platform innovations for getting machines to work with precision and a wide range of materials have largely been completed. The next level of 3D printing is where it is seriously embedded into industrial processes, and it is rapidly being adopted by giants like Volkswagen, NASA, and Adidas. And with the scale-up comes a bevy of new developments. Here are 10 ways in which artificial intelligence will bring 3D printing to its true heyday.
- The automatic designer
State-of-the-art product design modeling software such as Fusion 360, Solidworks Xdesign, Autodesk Dreamcatcher, and CATIA can generate shapes that optimally fit a certain set of requirements.
The designer can choose to set parts of the model in stone so the AI will leave it untouched. But where it is free to let loose its computationally creative powers, the shape will automatically be evolved to an optimal configuration to meet certain mechanical loading conditions, and work best with a type of material and printing process or cost constraints.
In a similar vein, AI can optimize the porosity and lattice structure of the infill material throughout the different areas of the volume to optimize strength and make the design lightweight. It can also generate so-called metamaterials that display complex dynamic behavior such as bending and stretching because of their internal structure.
One day, when computers learn to build up enormous bodies of design knowledge, the star designer can become obsolete, but for now human creativity is still paramount in developing the computer’s creation into a desirable end product.
- Precise print settings
Before a model can be printed, it goes through a preparation process called slicing. While two-dimensional printing can already be a challenging task, with 3D or even 4D printing the parameters deviate even more from press-of-a-button solutions.
The latest generations of slicer software offer expert users more than 200 settings, such as layer height, shell thickness, speeds for different sections of the print, and support and infill material options. You can imagine the improvement that occurs when a neural network, like a master pianist, learns to control all these parameters.
And since most 3D printers construct objects in layer-by-layer fashion and use support material to avoid printing in mid-air, object orientation is one of the many key issues that determine printability, part quality, strength, cycle time, and visual impact.
Another perk of AI-enabled slicing is that it can fit parts better toward intended dimensional tolerances. Finding the optimal settings for a certain product type and its material requirements is an expertise on its own.
- Manufacturing planning
As the industry matures, fablabs become print farms and 3D printing turns into additive manufacturing. When having to manage dozens of rapid prototyping projects on a daily basis, production planning starts to be an organizational challenge. And human-level intuition is often insufficient because it fails to take in enough information, resulting in unrealistic planning.
When one aspect is overlooked, the intended planning falls short and the entire process will fail. Mistakes occur in assigning the right jobs to the right machines in the right time window, plus internal communication takes up a major chunk of time.
Companies like London-based AMFG are forerunners in developing the technology to streamline this process and minimize machine downtime, just like manufacturing process planning tools do for the established industries. This way, the vision of the automated AM factory comes closer to fruition.
- In-print monitoring
3D printing generates a high scrap rate. To the inexperienced operator, as much as 50% of the total raw material can go to waste. If the printer could spot an error before job completion, it prevents material loss and possible equipment damage.
Filament runout sensors are only a small part of the puzzle. The key enabler here is computer vision. Using advanced camera systems or OCT scanners, 3D printers of the future will detect failures such as stringing, warping, clogged nozzles, material jams, air bubbles, and surface defects such as burn marks, cracks, and residues.
Inkbit is one of the commercial pioneers, with further development currently on the agenda of several European research firms.
In the automotive and aerospace areas a demand exists for high-performance materials such as nickel-chromium alloys. From a metallurgical point of view, blending two metals is very complex and requires a precise recipe for success. Similar challenges present themselves with printing glass, cement, silicones, epoxies, and composites.
To keep the welding process stable during prints, it has to be continuously monitored and the parameters adjusted. AI here learns to find the sweet spot that creates usable parts, as exemplified by Fraunhofer’s FutureAM project. The self-learning neural networks employed can also learn to optimize toolpaths for accelerating the print and adding part performance. Another related application is having ML teach the machine to improve full-color prints so the coloring comes out supremely sharp and high-fidelity.
Imagine asking Alexa to come up with the perfect material for your special 3D printing project, and she instantly gives the spot-on answer. The core technology is recommender systems.
These filter large amounts of data and find the links that would take humans hours if not days to do. With the expanding range of available materials this is a welcome solution.
An expert system could also work for making a selection out of the thousands of available standard components to be inserted into 3D printed assemblies. It is precisely what German precision mechanical engineering firm Heidelberg intends to do. Instead of deploying their high-profile experts with limited availability, the alternative will be to have clients consult a software agent called Performance Advisor Technology (PAT).
Modern logistics companies use AI to automatically find the optimal way to load their trucks based on the different parcel sizes. A similar application transfers to 3D printing.
In processes like selective laser sintering (SLS) and direct metal laser sintering (DMLS), objects are produced by fusing particles in a bed of powder. This means that during their creation, printed parts are supported by their own material, and the entire volume of the bed can be filled up with parts on top of one another, even nested inside each other.
Because of the easily removable structure of supports, this is also partially feasible with SLA processes.
AI-powered software enters the scene by finding the optimal layout to maximize the number of parts produced in a single print run. In SLS and other sintering processes, the lasers cause a heat distribution that impacts part quality, which the AI can prevent as well. Startups such as Printsyst are now establishing patents for AI-driven solutions to automatically prevent suboptimal prints.
- Mesh preparation
Before 3D files, like Grasshopper files, are printable, they have to be converted into a mesh format (STL, OBJ, PLY, AMF, Collada, etc.) that stores geometry by means of triangles. Since different industries work with different native formats, conversions do not always run smoothly and can result in files that take hours to manually repair.
Existing auto-repair functions are marginally effective and will be greatly assisted by artificial intelligence. The most notorious issue here is the phenomenon of rat’s nests: inexplicable accumulations of intersecting triangles that are a nightmare to remove.
With the help of AI, five hours turns into five seconds. It can also detect unneeded level-of-detail as a result of automatic file conversions, and adjust resolution to its ideal value for the application while improving file size and processability in the meantime.
- Cloud connectivity
When 3D printers are adorned with smart sensor networks, all of the captured raw data can instantly be transferred to databases on the cloud. Using the right KPI performance metrics, statistical algorithms can detect the most critical anomalies in system parameters that are causing bottlenecks. And they can do so for all machines worldwide. This next level of process control results in a hyper efficient production pipeline.
- Moral machines
When left on its own, AI can turn into an uncontrollable monster, so we need to keep check of when and where exactly to deploy it. With the hype train of digital transformation and the democratization of manufacturing power, anybody who jumps on the bandwagon of 3D printing can create intelligent products and use 3D printers for the hardware side of things.
What started with semi-functional guns can indeed turn into the perils of maleficent or out-of-control robots, so it is paramount that creators keep connecting and sharing projects to keep on the right track of innovation. An innovation that can be of use here is artificial moral agents (AMAs), which are 3D printers that are imbued with a sense of morality and can advise whether or not to carry out certain projects.
About The Author
Ralph Zoontjens is a product designer born and based in Tilburg, the Netherlands. After graduating with a dual degree in Industrial Design from Eindhoven University of Technology he set up his company IDZone to specialize in 3D printing, which he aims to popularize by revealing its potential for unbounded geometry, customization, and generative design, as well as blog writing about the various topics that envelop his practice. In his spare time he likes distance running, playing chess, or spending time with his son.