Milling force-induced elastic deformation modeling and physical constraint transfer prediction of thin-walled grid conical shell

Imagine a thin, curved shell made of metal, shaped like a cone and reinforced with a grid of ribs, much like the skeleton of a bird's wing or the framework of a bridge. These structures are vital in aerospace and deep-sea exploration because they are incredibly strong yet light enough to carry heavy loads without adding excessive weight. However, making them is a delicate balancing act. When a machine tool cuts away material to shape these shells, the force of the cutting tool pushes against the thin metal. Because the metal is so thin and the support from the ribs is uneven, the shell bends slightly under this pressure. This bending is invisible to the naked eye during the cut, but it changes the depth of the cut. When the tool moves away and the pressure is released, the metal springs back to its original shape, leaving the final surface slightly different from what was intended. This results in a part that is uneven in thickness, which can compromise its strength and safety.

The challenge for engineers has been predicting exactly how much this bending will happen. Traditional methods either rely on complex mathematical theories that struggle with the irregular patterns of the ribs, or they use computer simulations that are incredibly accurate but so slow and expensive that they cannot be used for every single cut on a real factory floor. There is also a lack of real-world data; running enough physical tests to learn every possible variation would take years and destroy countless expensive parts. Researchers needed a way to combine the speed of data-driven predictions with the reliability of physical laws, especially when they only had a small number of real experiments to work with.

A team of researchers at Dalian University of Technology has developed a new approach to solve this problem for these specific grid-patterned conical shells. Instead of trying to guess the final shape directly from the machine's settings, they broke the problem down into two connected steps. First, they focused on predicting the force the cutting tool exerts on the metal. To do this, they used a technique called transfer learning. They started by training a computer model on a vast amount of simulated data, where the computer could run thousands of virtual cutting scenarios quickly. This gave the model a strong understanding of how the tool generally behaves. Then, they took this knowledge and refined it using a much smaller set of data from actual milling experiments. This allowed the model to learn the specific quirks of the real machine and the real metal without needing thousands of physical tests.

Once the model predicted the cutting force, it passed that information to a second part of the system. This second part used a set of physical rules derived from the theory of how thin shells bend. It took the predicted force and the specific stiffness of the metal at that exact spot—considering how close the tool was to a rib or a gap—and calculated exactly how much the shell would bend. By linking the data-driven prediction of the force with the physics-based calculation of the bend, the researchers created a system that is both fast and physically consistent. They verified this method by milling a real aluminum alloy cone shell, which started with a thickness of 10 millimeters and was machined down to a final rib height of 3.5 millimeters. The machine used a 10-millimeter diameter cutter spinning at 5,000 revolutions per minute.

The results showed that this hybrid method was significantly more accurate than using data alone or using traditional simulation methods. When tested against real measurements taken during the process, the new model correctly predicted the elastic deformation in more than 93 percent of the cases, with some test modes reaching over 95 percent accuracy. In contrast, models that relied only on data or only on simulations without this specific combination of techniques showed lower accuracy and struggled more with sudden changes in the metal's stiffness. The study demonstrates that by treating the cutting force as a crucial middle step and using a small amount of real-world data to correct a large amount of simulated data, engineers can reliably predict how these complex, thin-walled parts will behave. This capability offers a practical path toward manufacturing these critical aerospace and deep-sea components with the high precision required for their safe operation.