Updated: Jan 22
It’s always fascinating to see the clashing of two industry 4.0 technologies. In the additive manufacturing industry, the use of artificial intelligence has been rife, with software developers and research institutions alike seeking to leverage the predictive power of machine learning across two primary applications.
The first is process control. While industrial 3D printing has advanced to technological levels never seen before, it’s still not perfect, and part defects like porosities are still a major issue. In light of this, machine learning algorithms are used for either defect detection or parameter modulation purposes (or both), all with the intent of reducing print failures and saving on time and material costs.
Although not as common, the other major application of AI is in the formulation of new 3D printing materials.
Defect Detection and Closed-Loop Control
The development of closed-loop control systems has been a key goal for additive manufacturing engineers for years now, but hard-coding such complicated algorithms can be a real pain. This is where AI comes in.
Late last year, a team of researchers from Argonne National Laboratory and Texas A&M University used real-time temperature data, together with machine learning algorithms, to refine a defect prediction algorithm for laser powder bed fusion 3D printing. Specifically, the team made correlative links between thermal history and the formation of subsurface defects that occur during 3D printing. This ML model allowed them to predict exactly when and where a defect might occur simply by tracking the temperature of the build chamber with in-situ infrared cameras.
As such, this ability to identify where porosities are likely to form just from infrared imaging is an extremely powerful tool. It eliminates the need for costly individual part inspections, which are not always feasible when dealing with high production volumes.
Taking a more visual approach, Dr. Joshua Pearce of Michigan Technological University has previously developed an open source, computer vision-based software algorithm capable of print failure detection in the FDM process. Interestingly, the ML-based algorithm utilizes the power of a single webcam pointed at the build plate. It tracks, on a layer by layer basis, any printing errors, and defects that appear on the exterior or interior of the printed part, modifying parameters such as nozzle temperature and print speed in real time in an attempt to salvage the part.
Although not as ‘industrial’, this AI approach serves the same purpose - it saves the user material costs and time by reducing the occurrence of print failures.
AI in 3D Generative Design Solution
3D printing is a complicated process, and AI can significantly help improve this technology to make it more efficient. One such area is the generative design. Artificial intelligence and machine learning are being increasingly used to make 3D Models. Netfabb solutions from Autodesk provides solutions that use machine learning to develop multiple design solutions. The solutions take benchmarks provides by a designer or engineer, e.g. size of product, weight, strength, style, materials, cost, and any number of other criteria, and then use cloud computing and machine learning to create multiple design solutions.
Using AI to develop new 3D printing materials
Formulating new materials is certainly not easy, and requires extensive and costly testing procedures to ensure high-performance material properties. Much of the work in materials development is data crunching, whereby certain heat treatments or additives are mapped onto corresponding properties. So not use the data processing capabilities of AI for the job?
A key player in this sector is Intellegens, a University of Cambridge spin-off. The company’s business model revolves around its Alchemite platform, which uses machine learning to design new materials specifically for 3D printing. Intellegens has previously used its platform to design a new nickel-based alloy for Direct Laser Deposition as part of a research project. In this case, the deep learning capabilities of Alchemite were used to pinpoint property-to-property relationships very quickly, enabling the team to use a large database of thermal resistance measurements to extrapolate just ten data entries of alloy processability - impressive.
Similarly, researchers from New York University’s Tandon School of Engineering have previously used ML tools to reverse engineer glass and carbon fiber 3D printed components. Specifically, the team applied AI algorithms to CT scans of 3D printed parts, enabling them to predict the printing toolpaths behind certain properties such as structural strength, flexibility, and durability.
Then, the team was able to reconstruct the components with an accuracy of 0.33%, along with all of their high-performance properties. Interestingly, this project actually raised concerns around the security of IP relating to specially 3D printed composite parts, as their development processes were shown to be bypassed quite easily simply by running them through ML algorithms.
It’s not hard to see, then, how AI can be both an extremely powerful technology and a malicious tool for those with ill intents. Regardless, it’s here to stay, and its application in the development of closed-loop control systems and materials alike is undeniably handy. As far as predictions go, it probably won’t be long before we see component-based search engines that predict which STL files we need next for our basement DIY projects.