In the world of 3D printing or additive manufacturing, laser-based metal processing is a popular technique that allows for automated, precise, and rapid production of intricate components.
Laser-based processing of metals involves using a laser as the energy source to manipulate the metal. A laser is an amplified beam of light or electromagnetic radiation that can propagate in a straight line with little divergence.
This makes lasers highly useful in material processing, where it is used for machining, joining, and surface engineering. In additive manufacturing, lasers are used to melt materials and fabricate components layer-wise.
Additive manufacturing is simply creating a product layer-by-layer. It started with using plastics as a material, thanks to the ease of processability. But it has now grown to include all types of materials, including metallic materials.
Metallic materials are known for their attractive properties, such as excellent electrical conductivity and high strength, ductility, and melting point, making them highly useful in biomedical, energy, architecture, and military applications.
Laser processing of metals meanwhile offers unique benefits of high energy density, a narrow heat-affected zone, and little contamination. That’s why laser processes are used in numerous sectors, especially where maximum precision and high customization are required. But it has its own complications and technical challenges.
“To ensure that laser-based processes can be used flexibly and achieve consistent results, we are working on better understanding, monitoring, and control of these processes.”
– Elia Iseli, Research group leader in Empa’s Advanced Materials Processing laboratory
With this goal, researchers Giulio Masinelli and Chang Rajani from Empa in Thun are making laser-based manufacturing techniques more accessible, affordable, and efficient, using machine learning.
Understanding the Advantages and Challenges of Laser Powder Bed Fusion (PBF-LB)
In the wider field of laser-based metal processing, Powder Bed Fusion is a popular one, which involves using a laser to melt thin layers of metal powder in the exact spots and welding them all together to produce the final component.
Powder Bed Fusion with Laser Beam (PBF-LB), meanwhile, is a specialized technique that has been gaining a lot of attention in recent years. In this prominent additive manufacturing technology, lasers that emit very high powers are used to specifically melt metallic powders layer-wise before mixing them into customized and highly precise components.
This technique allows for the production of complex geometries while offering customization capabilities and ensuring material efficiency.
These characteristics make PBF-LB specifically beneficial for industries like automotive, medical, aerospace, and consumer products, where we need lightweight and complex parts, personalized designs, precision, weight reduction, and quick prototyping, respectively.
While versatile and efficient, the technique faces several obstacles in reaching broader adoption and achieving optimization.
This includes having difficulty in identifying the ideal processing framework for the metallic powder being used.
“Even a new batch of the same starting powder can require completely different settings.”
– Masinelli
The high-energy input needed for metal melting in this technique actually creates complex physical mechanisms that negatively affect the quality of parts. These mechanisms include inconsistencies in material properties, the influence of atmospheric gases, and the interaction of the laser with the vapor plume. All of these phenomena bring problems in identifying parameters.
This is primarily due to the two modes. One is the conduction mode in which the metal is just melted, and it is ideal for thin and precise components. The other option is keyhole mode, in which the metal can be vaporized in some cases. It is faster but also less precise, making it suitable for thicker components.
The boundary between these modes, however, depends on different parameters, and achieving the best quality in the final products requires the right settings, which vary depending on the material being processed.
The complex interactions between the material and laser also make the process sensitive to really small variations, which can then lead to issues in production, and this makes the technique time and resource-intensive. As a result, PBF-LB needs the laborious fine-tuning of parameters to achieve consistent results.
It doesn’t just end here either. The samples produced at this stage are then analyzed using different techniques such as microstructural analysis, density measurements, and X-ray computed tomography (CT).
These methods provide detailed information on internal structures and find defects, which are critical to evaluating the quality and performance of PBF-LB parts, but again, they need specialized equipment and expert knowledge on top of being expensive and time-consuming.
“That is why many companies cannot afford PBF in the first place.”
– Masinelli
To address all these problems, researchers from Empa utilized machine learning to make laser processes more efficient, cost-effective, and precise.
Click here to learn how additive manufacturing is reshaping industries.
Leveraging AI for Real-Time Control in Metal 3D Printing
For sample analysis, researchers have turned to real-time monitoring methods using sensors such as acoustic emission (AE), high-speed thermal imaging, and optical sensors.
Real-time monitoring has been chosen due to its ability to detect undesirable events during the manufacturing process. This allows for immediate adjustments, in turn, saving resources by removing and re-melting defects.
These real-time monitoring techniques are usually based on machine learning (ML) algorithms.
A field of study in artificial intelligence, ML is concerned with the development of statistical algorithms that learn from data. These algorithms extract meaningful patterns from high-dimensional data and then make predictions, in the case of metal processing, that’s on part quality, without needing to explicitly program complex physical models.
These AI approaches aren’t without limitations, though. The challenges include the model learning to detect changes in process parameters instead of the process regime and defect formation.
The natural drift in machine parameters over time also presents a barrier to the generalization of these models, limiting the practical applicability of AI models in real-world manufacturing environments. Then there are issues with automation, which require specialized equipment and get complicated by multiple parameters, exploring which is also challenging and resource-intensive.
There is a clear need for algorithms that can autonomously navigate the PBF parameter space, considering multiple process variables, to identify optimal conditions and understand the underlying melting regimes.
This need is now being addressed by researchers from Empa, who have proposed a new method that employs unsupervised collection of optical data with a focus on melting regime identification without requiring labeled data or extensive post-processing analysis.
Implementing Unsupervised Learning to Optimize PBF-LB Parameters
The novel unsupervised technique developed by Empa researchers focuses on two main parameters: laser power and scanning speed, which are identified as having the most significant impact on the melting regime.
While the focus of the study1 was on these two parameters, the technique can also be used for additional process parameters. In the future, the researchers will incorporate gas flow rate, hatch spacing, and layer thickness into their algorithm to enable a more comprehensive exploration of the PBF-LB parameter space.
For now, the proposed method accurately points out the transition between conduction mode and keyhole mode.
The unsupervised approach also provides a basis for extracting processing maps without depending on labeled data, which offers a considerable advantage in PBF-LB, where getting labeled data is both costly and challenging.
The study actually builds upon this foundation and introduces an original method that combines parts of active learning (selecting the most informative data points) and Bayesian optimization (iterative sampling strategy utilizing a probabilistic model) to derive processing maps efficiently.
What makes the approach different is that it starts with no data and then progressively builds the dataset by deciding where to perform each new experiment, thus allowing the experimental process to optimize.
Notably, despite employing an iterative approach for refinement, the model remains unsupervised throughout the process, as it doesn’t require labeled data. To identify the melting regimes, the algorithm relies on features extracted from the optical data, and the results are then used to train a Gaussian Process Classifier (GPC) to provide a probabilistic estimation of the map.
As for the iterative facet, the algorithm selects new trial settings based on those fields that have high uncertainty in the predictions, which improves the processing map’s estimation.
Basically, the algorithm is taught to detect which welding mode the laser is in during a test run, using data from optical sensors already incorporated in the laser machines. Based on this, the algorithm sets the parameters for the next test.
“We hope that our algorithm will enable non-experts to use PBF devices,” said Masinelli. It only needs to be integrated into the firmware of the laser welding machines by manufacturers.
Evaluating the Effectiveness of the AI Model in PBF-LB Applications
The new algorithm introduced by researchers to eliminate the need for extensive parameter tuning, which limits PBF-LB’s broader adoption, independently identifies the melting regimes using data from photodiodes.
And when tested in the laboratory, the team found the method to be highly accurate, achieving an F1-score of 89.2% across two materials. To evaluate the performance, researchers printed multiple parts in two materials.
The first one was Ti-6Al-4V, which is one of the most commonly used (alpha-beta) titanium alloys, having excellent corrosion resistance and a high specific strength. The other one was 316L stainless steel, a low-carbon version of 316 stainless steel, which is commonly used in food processing, pharmaceutical equipment, medical devices, jewellery, luxury watches, wastewater treatment, and in the chemical industry.
In particular, the team conducted melt pool inspections to verify the algorithm’s predictions.
The evaluation showed that the approach reduced the need for experimental trials by 67% in both metals while maintaining robust performance. This can significantly lower the cost of parameter exploration. Meanwhile, there was only a maximum of 8.88% decrease in the F1-score in comparison to a traditional full factorial experiment design.
The study stated:
“These results underline the efficiency of our method in the context of autonomous processing map derivation for advanced manufacturing processes.”
The method introduced here, researchers believe, can “greatly improve” both the efficiency and reliability of PBF-LB, which could lead to its broader adoption by enhancing its overall effectiveness across various sectors. According to the study:
“Our results demonstrate the potential of this method to streamline PBF-LB optimization, making it more feasible for industrial applications and paving the way for its broader adoption.”
Enhancing Laser Welding Processes Through AI and FPGA Integration
Besides optimizing preliminary experiments, the researchers also improved the welding process in another project.
When it comes to laser welding, even with ideal settings, the process can still yield unpredictable deviations, and even a minor one can lead to serious defects in the product.
“It is currently not possible to influence the welding process in real time,” said researcher Rajani. “This is beyond the capabilities of human experts.”
– Researcher Rajani
In fact, even computers struggle with the speed at which data needs to be examined and decisions must be made. The researchers used a specialized type of computer chip here.
This chip is called a field-programmable gate array (FPGA), which is designed for the purpose of high-performance computing (HPC) and prototyping. The chip can be programmed after being released from the manufacturer and adapted for different use cases without needing to physically alter the hardware. Their versatility, combined with high performance, makes them highly valuable in aerospace, automotive, and telecommunications industries.
Masinelli noted:
“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.”
The researchers linked the FPGA to a PC to serve as a “backup brain”. As the chip observes and controls the laser parameters, this data is also utilized by the algorithm on the PC for learning.
“If we are satisfied with the performance of the algorithm in the virtual environment on the PC, we can ‘transfer’ it to the FPGA and make the chip more intelligent all at once.”
– Masinelli
The researchers believe that ML and AI have the potential to contribute significantly to laser-based metal processing. As such, they will continue to develop their algorithms and models, as well as expand their area of application, in collaboration with other research groups and industry partners.
Exploring Investment Opportunities in 3D Printing Technologies
Now, a key player in metal additive design and manufacturing is Colibrium Additive. It is part of General Electric Company (GE +0.16%), which is now doing business as GE Aerospace.
Previously known as GE Additive, it was relaunched as Colibrium Additive last summer, and as part of the rebranding, Concept Laser and Arcam EBM were retired.
“While we are changing our name, we maintain our unwavering focus on our customers, quality, and reliability. We will continue to lead the Additive Manufacturing industry from the front and positively disrupt it.”
– CEO Alexander Schmitz
General Electric (GE +0.16%)
When it comes to the 3D printers offered by Colibrium Additive, they include Electron Beam Powder Bed Fusion (EB-PBF) printers, Laser Powder Bed Fusion (L-PBF) printers, and Binder Jet.
As for the company’s market performance, it has been really thriving over the last few years.
With a market capitalization of over $260 billion, GE shares are currently trading at around $244, up a substantial 46% this year. The company’s stock is fast nearing its peak at around $290, which was hit in 2000. Its EPS (TTM) is 6.35, and the P/E (TTM) is 38.46, while the dividend yield available to shareholders is 0.59%.
General Electric Company (GE +0.16%)
Meanwhile, the company financials show a strong Q1 2025 in which GE recorded a total revenue of $9.9 billion, an increase of 11% while total orders jumped by 12% to $12.3 billion.
This strong start to 2025 was driven by commercial services, stated CEO H. Lawrence Culp, Jr., while noting the macroeconomic dynamics that require the company to take strategic actions, such as controlling costs and leveraging available trade programs.
Operating profit surged 38% in 1Q25 to $2.1 billion while adjusted EPS had a 60% increase to $1.49. During this period, GE also reported $1.5 billion in cash from operating activities (GAAP) while free cash flow increased 14% to $1.4 billion. The company also reported a commercial services backlog of over $140 billion.
Amidst all this, Propulsion & Additive Technologies only grew 1%, with the company noting that prices and volume offset lower shipments resulting from a planned soft start in equipment sales.
In its annual report this year, GE stated “declines in the Additive Manufacturing industry due to slower adoption of technology,” but at the same time, it noted Colibrium Additive to be “a critical business for current and future technology at GE Aerospace as we continue to focus on where it can create the most value.”
Latest General Electric (GE) Stock News and Developments
Conclusion
As AI continues to advance and transform industries, it’s also helping redefine what’s possible in modern manufacturing by accelerating process optimization and enabling real-time adaptability.
By significantly reducing the time and cost associated with parameter tuning and defect detection in PBF and achieving real-time control in laser welding, laser-based additive manufacturing is poised for broader adoption, paving the way for a new era of efficient, accessible, and customized production.
Click here for a list of the top 3d printing stocks.
Studies Referenced:
1. Masinelli, G., Schlenger, L., Wasmer, K., Ivas, T., Jhabvala, J., Rajani, C., Jamili, A., Logé, R., Hoffmann, P., & Atienza, D. (2025). Autonomous exploration of the PBF-LB parameter space: An uncertainty-driven algorithm for automated processing map generation. Additive Manufacturing, 87, 104677. https://doi.org/10.1016/j.addma.2025.104677