Most of us savor spaghetti. The word that describes the long, thin staple of Italian cuisine, however, is best avoided in the world of additive manufacturing. Typically used to describe the tangled mess of stringy plastic that often results from a failed print, the term gives 3D printers sleepless nights.
For 3D hobbyists, an open-source software like Spaghetti Detective can monitor prints and notify users if it detects a possible print failure. But that won’t cut ice with big manufacturing companies where the failure of a single critical part during an industrial 3D printing process can pose a very serious problem. Not only does it lead to a colossal waste of material and time but it also lowers productivity and adds to the cost of manufacturing a part.
Artificial intelligence (AI)-powered 3D printing promises to effectively bridge this gap.
Armed Forces Prove Early Movers
Almost two years back, global security and aerospace company Lockheed Martin and the Office of Naval Research said they were jointly exploring how to apply AI to train robots to independently oversee and optimize the 3D printing of complex parts. The researchers said they would apply machine learning techniques to monitor variables and control the robot during fabrication.
These are critical considerations since aerospace-grade metals, including several recipes of titanium alloy, are supplied by foundries. They come with guaranteed strength, porosity and thermal tolerance characteristics. 3D-printed metals, on the other hand, may look identical to traditionally manufactured pieces but may not fit the bill on close inspection.
Lockheed Martin’s research was aimed at helping machines make decisions about how to optimize structures based on previously verified analysis. That verified analysis and integration into a 3D printing robotic system was core to the $5.8 million contract signed in October 2018.
Last month, army researchers said they discovered how to detect and monitor the wear and tear of 3D-printed maraging steel through sensor measurement. Maraging steels—a portmanteau of martensitic and aging, referring to an extended heat-treatment process—are steels that possess superior strength and toughness while remaining ductile.
“3D printed parts display certain attributes, due to the manufacturing process itself, which, unchecked, may cause these parts to degrade in manners not observed in traditionally machined parts,” explained Jaret C. Riddick, director of the Vehicle Technology Directorate at the U.S. Army’s Combat Capabilities Development Command’s Army Research Laboratory.
The study was recently published in the International Journal of Advanced Manufacturing Technology. Todd C. Henry, a mechanical engineer at the laboratory who coauthored the study, says the sensor technology he is developing offers a way to track individual parts, predict failure points, and replace them a few cycles before they break.
An experimental validation set was done to assess the real-time fatigue behavior of metallic 3D-printed maraging steel structures. These findings are now being applied to the 3D printing of stainless steel parts and using machine learning (ML) techniques, instead of sensors, to characterize the life of parts, according to Henry.
How AI Systems Help
Machine and deep learning, subsets of AI, can sift through mountains of data and make very good predictions. Of course, all this is subject to the data being good. ML is broadly about teaching a computer how to spot patterns and use mountains of data to make connections to accomplish specific tasks. A recommendation engine is a good example.
Deep learning, an advanced ML technique, uses layered (hence “deep”) neural networks that are loosely modeled on the human brain. Neural nets enable image recognition, speech recognition, self-driving cars and smart home automation devices, among other things.
Driving Home the Point
In August 2018, researchers from Kansas State University’s Department of Industrial and Manufacturing Systems Engineering (IMSE) integrated supervised machine learning, a camera, and image processing software to create a production-quality monitoring system for assessing 3D-printed parts in realtime. They used a LulzBot Mini 3D printer for this purpose.
The researchers used the Support Vector Machine (SVM), which is a supervised machine learning model. SVM applications are typically found in bioinformatics, image and text recognition, among other applications.
The images calculated at checkpoints were loaded as inputs to the vectors of the training models. Two categories of training models were loaded into the system: good (identical to the ideal parts) and bad (defective parts). The SVM training algorithm built a model that classified any new model loaded into the system as either a ‘good’ or a ‘bad’ print. If a model was classified as ‘bad’ during the initial stages of printing, a corrective measure was employed to stop the printing process to prevent further wastage, and the part was reprinted.
The researchers noted that the main drawback of the proposed method is that the printing process needs to be paused while the images of a semifinished part are taken. Another drawback is that since only top view images are taken, the proposed method might not be able to detect the defects on the vertical plane. Researchers believe that they can overcome these drawbacks by incorporating cameras on the sides of the printer as well to detect defects on both the horizontal and vertical planes.
Geometric Accuracy Control
Researchers at Purdue University and the University of Southern California are using ML to ensure among other things that the pieces of an aircraft fit together more precisely and can be assembled with less testing and time. The technology allows a user to run the software component locally within their current network, exposing an application programming interface (API). The software uses ML to analyze the product data and create plans to manufacture the needed pieces with greater accuracy.
The researchers have developed a new model-building algorithm and computer application for geometric accuracy control in additive manufacturing systems. They claim the improved accuracy ensures that the produced parts are within the needed tolerances and that every part produced is consistent and will perform the same way, whether it was created on a different machine or 12 months later.
Layer-By-Layer Scanning
In June 2019, Inkbit, a startup out of MIT, announced that it had paired its multi-material inkjet 3D printer with machine vision and machine learning systems.
Inkbit uses a proprietary 3D scanning system to generate a topographical map of each layer after deposition. Any discrepancy is corrected in subsequent layers. The data are also used to train a machine learning algorithm that enables the printer to learn the properties of each material and anticipate its behavior. This ensures that parts are built quickly and accurately every time. The layer-by-layer scanning also allows Inkbit to generate a full 3D reconstruction of each part as it was printed, providing a complete digital record of every print.
Machine vision (MV) comprises methods used to provide imaging-based automatic inspection and analysis for applications such as automatic inspection, process control, and robot guidance, usually in industry. A machine vision system uses a camera to view an image. Computer vision algorithms then process and interpret the image before instructing other components in the system to act upon that data.
“The company was born out of the idea of endowing a 3-D printer with eyes and brains,” Inkbit cofounder and CEO Davide Marini said back in June 2019.
Autonomous Decisions on the Fly
Similarly, London-based Ai Build develops AI and robotic technologies for large-scale additive manufacturing. It has developed an automated AI-based 3D printing technology with a smart extruder. Its AiMaker attaches itself to industrial robotic arms and is able to 3D print large objects at high speed with great accuracy.
Combining advanced AI algorithms with real-time manufacturing data from its sensors and cameras, AiMaker detects problems and makes autonomous decisions on the fly in a bid to achieve the best possible print quality.
IoT Control
ML is currently also being used to solve the accurate quality of the 3D printing problem by using generative design and testing in the prototyping or prefabrication stage. Bengaluru-headquartered HyCube Works, for instance, has an AI-based smart extruder that can detect problems such as clogging while 3D printing. The Indian startup also uses the Internet of Things (IoT) technology to control its 3D printers remotely.
“3D printing involves numerous and complex parameters to be controlled and monitored in the process to achieve an acceptable level of accuracy. Trial and error methods for finding the correct lattice positions or design of appropriate support structures, are not a sustainable or speedy solution,”explained cofounder Reethan Doijode.
HyCube Works, according to Doijode, plans to “integrate supervised algorithms for defect detection in real-time build control of the 3D-printed part in the machine, and any disorientation of printing layers or failures in the part would be monitored. The vision system comprehensively scans each layer of the object as it is being printed to correct errors in realtime.”
Peek into the Future
Big companies like GE, with their massive R&D budgets, are taking further steps. In January 2019, GE Research and GE Additive integrated edge computing with 3D printers to give the latter “‘digital eyes’ to track each layer of every build” in a bid to help manufacturers know in real time whether a part build is good or has to be scrapped.
Edge computers sit directly on machines—in this case, 3D printers. Edge systems equipped with machine-learning algorithms can run instant analysis and supply insights to operators, and eventually to the printers themselves. GE believes that while the system can transform the 3Dprinting process by monitoring the building of parts in real time, the Holy Grail is controlling the process at the speed and precision required to prevent or fix minute defects on the fly.
Meanwhile, researchers believe that additive manufacturing can get a further boost if AI-powered 3D printers collaborate as teams.
A multidisciplinary robotics team at the New York University (NYU) Tandon School of Engineering, hosted by NYU’s Center for Urban Science and Progress (CUSP) and supported by a $1.2 million grant from the National Science Foundation (NSF), is working on one such concept. It is designing autonomous systems for 3D printers on robotic arms attached to mobile, roving platforms.
Functioning in teams—a concept called collective additive manufacturing—these printers are equipped with ML and other AI capabilities and could repair bridges, tunnels and other civic structures; work in ocean depths and disaster zones; or even head to space to work on the Moon, Mars, and beyond.
The team plans to demonstrate the effectiveness of the algorithms by 3D printing new concretes using mobile printers at NYU Tandon. 3D printers already churn out plane parts and even houses. This AI boost will only take the technology to another level of manufacturing.