If errors occur during the 3D printing process of a product, these errors are usually not discovered until afterwards. In the worst case, this makes the product unusable, wasting a lot of time and therefore money. Researchers at GE labs are working on a technology that could provide a solution. The researchers want to use high-resolution cameras, machine learning and artificial intelligence to notice deviations in the printing process in real-time, so that they can be corrected by the printer.
In additive manufacturing, an object is created by fusing together several layers of material. There are various techniques for this, one example being selective laser sintering (SLS). In SLS, a layer of powder is placed on the print bed, after which a laser is used to harden the powder in specific areas. A new layer of powder is then placed on the print bed and this powder is cured again using the laser. The new layer of material thereby fuses with the previously printed layer, creating an object layer by layer.
Deviations in powder particles
Unfortunately, in additive manufacturing, errors can occur in the printed object due to factors beyond the printer's control. For instance, variations in the powder particles can lead to deviations in the printed object. It is therefore necessary to carefully check objects produced using additive manufacturing systems. Currently, this checking is often done manually after the production process is completed.
Researchers at GE labs want to use a combination of artificial intelligence and machine learning to detect defects in printed layers during the printing process. "Using artificial intelligence and machine learning, we are essentially turning 3D printers into their own inspectors," explains Joseph Vinciquerra, head of GE labs' Additive Research Lab. "By eliminating the need to inspect parts after they are fully built, we can take days or even weeks off the entire manufacturing process and achieve a breakthrough in productivity."
Detecting problems in printed layers
The team is experimenting with this technology by printing simple geometric shapes, such as a cylinder. The printer used for this purpose is equipped with high-resolution cameras that capture each printed layer to record anomalies barely visible to the human eye. These include minimal pits or streaks in layers due to variations in the powder used. They then use a CT scanner to examine the printed object for problems caused by these anomalies.
All data collected in the process is analysed by a machine learning algorithm to discover correlations between problems in the final product and anomalies in the printed layers captured by cameras. "The more often you do this, the smarter the system becomes," Vinciquerra explains. "The computer vision will eventually have had enough training to tell us that a problem is imminent."
Correcting deviations
Vinciquerra and his team are working on a method to feed this information back to the 3D printer and adjust the control of the 3D printer accordingly. This enables the production system to take action during the manufacturing process to correct detected deviations in printed layers and prevent these deviations from causing problems in the final product. "The idea is that the machine has a compensation strategy based on what computer vision sees," Vinciquerra explains.
To make this possible, the researchers want to create a virtual representation of the manufacturing process. This representation represents what the printed object should look like under ideal conditions at a certain point in the production process. By continuously comparing the actual printed object with this virtual representation, it is possible to detect deviations in real time. If, according to the machine learning algorithm, these deviations will lead to problems in the final product, the printer can make changes to the rest of the printing process to compensate for the deviations.
Flexibility in production
"If we can spot a defect very early in the printing process, we have the option to stop and start again," Vinciquerra said. "If an object is already halfway through the printing process, we can evaluate whether the defect can be fixed by adjusting the rest of the layers still to be printed. We have never had this level of flexibility in production."
By the way, GE labs is not the only party working on developments to increase the quality of parts printed with SLS. Carnegie Mellon University, for instance, is working on a system that allows the quality of the powder used in SLS printers to be inspected prior to the printing process. This will look for abnormalities in the size of powder particles, the way particles are grouped and the shape of particles, among other things. This makes it possible to reject poor-quality powder and ensure that only suitable powder is used.
Author: Wouter Hoeffnagel
Source: GE labs
Source: Carnegie Mellon University
