Welding How Machine Learning Simplifies the Laser Process

Related Vendors

EMPA Eidgenössische Material- Prüfungs-und Forschungsanstalt

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

Laser-based metal processing enables the automated and precise production of complex components, whether for the automotive industry or for medicine. However, standard procedures require costly preliminary trials. Empa researchers have developed how machine learning makes laser processes more precise, cost-effective, and efficient.

Laser-based processes allow for the precise welding of metal components or the creation of more complex parts using 3D printing—quickly, accurately, and automatable. Therefore, laser-based processes are used in numerous industries, such as the automotive and aerospace industries, where the highest precision is required, or in medical technology, for example, to manufacture custom-made titanium implants. The problem: Laser processes are technically demanding. The interactions between laser and material make the process sensitive to the slightest deviations— whether in material properties or laser parameter settings. Even minor fluctuations can lead to production errors.

"In order for laser-based processes to be used flexibly and achieve consistent results, we are working on better understanding, monitoring, and controlling laser-based processes," says Elia Iseli, research group leader in the Empa department of Advanced Materials Processing in Thun. In line with these principles, Giulio Masinelli and Chang Rajani, two researchers from the group, aim to make laser-based production methods cheaper, more efficient, and more accessible—by means of machine learning.

Two Different Modes: Evaporate or Melt?

First, the two researchers focused on additive manufacturing, namely the 3D printing of metals using lasers. This process, technically called Powder Bed Fusion (PBF), works somewhat differently than conventional 3D printing. Thin layers of metal powder are melted with the laser precisely at the right spots, gradually "welding" the finished component out of it.

With PBF, complex geometries are possible that can hardly be realized with other methods. Before production can begin, a series of preliminary tests is necessary. This is because, in laser processing of metal, including PBF, there are fundamentally two modes: In the so-called "Conduction Mode," the metal is merely melted. In the "Keyhole Mode," or deep welding, it is partially vaporized. The slower "Conduction Mode" is suitable for thin and very precise parts. Deep welding is somewhat less accurate but much faster and also suitable for thicker workpieces.

Where exactly the boundary between these two modes lies depends on a multitude of parameters. For the best quality of the finished product, the correct settings are needed—and they vary greatly depending on the material being processed. "Even a new batch of the same starting powder can require completely different settings," says Masinelli.

Reduce Preliminary Trials by About Two-Thirds

Normally, a series of tests must be conducted before each batch to determine the optimal settings for parameters such as scan speed and laser power for the respective component. This consumes a lot of material and requires supervision by a professional. "Many companies cannot afford PBF for this reason," says Masinelli.

Masinelli and Rajani have optimized this series of tests using machine learning. They utilize data from optical sensors that are already present in the laser machines. The researchers have taught their algorithm to "see," based on these optical data, which welding mode the laser is currently in during a trial. Based on this, the algorithm determines the settings for the next trial. This reduces the number of required preliminary tests by about two-thirds while maintaining the same quality of the final product.

"We hope that our algorithm will allow even non-experts to use PBF equipment," Masinelli summarizes. For the algorithm to be used in industry, it would just need to be integrated into the firmware of laser welding machines by the equipment manufacturers.

Optimize Welding Process in Real-Time

In another project, Rajani and Masinelli focused on laser welding and went even further. They optimized not only the preliminary trials but also the welding process itself. Even with optimal settings, laser welding can be unpredictable, especially if tiny defects on the metal surface come under the laser beam.

"Influencing the welding process in real-time is currently not possible," says Chang Rajani. "That exceeds the capabilities of human experts." The speed at which data needs to be evaluated and decisions made is a challenge even for computers. Therefore, Rajani and Masinelli used a special type of computer chip for this task, a so-called Field-Programmable Gate Array (FPGA). "With FPGAs, we know exactly when they will execute a command and how long the execution will take—which is not the case with a conventional PC," explains Masinelli.

Nonetheless, the FPGA in their system is also connected to a PC, which serves as a kind of "backup brain." While the special chip is busy observing and controlling the laser parameters, the algorithm on the PC learns from this data. "If we are satisfied with the algorithm's performance in the PC's virtual environment, we can 'transfer' it to the FPGA and make the chip smarter all at once," Masinelli explains.

The two Empa researchers are convinced that machine learning and artificial intelligence can contribute a lot in the field of laser processing of metals. Therefore, they continue to develop their algorithms and models and expand their application area—in collaboration with partners from research and industry.

Machine Authentication

Machine Operator Authentication - Why knowing "who" matters