An Intelligent CNC Machine Tool Based on Random Forests Parameters Prediction Model

In the world of modern manufacturing, the journey from a digital design to a physical metal or plastic part relies heavily on a specific type of automated machine known as a computer numerical control, or CNC, tool. These machines are the workhorses of industry, capable of carving complex shapes with a precision that human hands cannot match. To make them work, engineers must write a set of instructions called G-code, which tells the machine exactly how to move its cutting tools, how fast to spin them, and how deeply to bite into the material. For decades, setting these speeds and feeds has been a task reserved for highly skilled experts. These specialists rely on years of experience and general rules of thumb to decide how a machine should behave. While this method works, it is slow, inconsistent, and often leads to different settings being used for the same job on different machines, simply because different operators have different habits. As products become more complex and production cycles shorten, the industry is searching for a way to make these setups faster and more reliable without losing the quality that comes from human expertise.

A team of researchers at the University of Pula in Croatia has proposed a new way to bridge the gap between human experience and automated efficiency. Instead of trying to replace the expert or the machine, they developed a system that learns from the factory floor itself. The researchers gathered data from three different milling machines in a workshop, analyzing 251 previously successful machining operations. They built a computer program that reads the G-code files from these past jobs, identifying exactly what kind of cut was being made, what material was being used, and what size tool was involved. By studying these records, they created a digital model that could look at a new, untested program and suggest the best settings for speed and feed based on what had worked well in the past. This approach does not require any new hardware or expensive sensors; it simply uses the information already present in the programs that operators have already written and tested.

The core of their method involves a type of computer learning called a random forest. Imagine a group of experts, each with a slightly different perspective, who vote on the best answer to a problem. In this case, the "experts" are mathematical decision trees that analyze the data from the past jobs. The researchers fed their system four key pieces of information: the specific machine being used, the type of material being cut, the kind of operation being performed, and the diameter of the cutting tool. The system then learned to predict four critical settings: how fast the tool should spin, how fast the machine should move, how wide the cut should be, and how deep it should go. The study focused on four common types of milling operations: drilling holes, flattening surfaces, cutting outlines, and hollowing out pockets. They tested this system on parts made from engineering plastics, specifically polypropylene and two types of polyethylene.

The results showed that the system could learn the hidden patterns of the workshop with surprising accuracy. The model found that the identity of the machine itself was the most important factor in deciding how fast the tool should spin. This makes sense, as different machines have different power levels and mechanical limits; a robust, heavy machine can handle higher speeds than a lighter one. The size of the cutting tool was the dominant factor for determining how fast the machine should move and how deep the cut should be. The type of operation also played a significant role in every prediction. Interestingly, the specific type of plastic being cut mattered very little for most settings, suggesting that the machine's capabilities and the tool's size are far more influential than the material in this context. The system was particularly good at predicting the width of the cut, almost perfectly matching the values used in the original successful programs. It was less certain about the depth of the cut, which is understandable because the depth often depends on how much material needs to be removed in a single pass, a detail that is not always clear from the code alone.

One of the most revealing findings was how much human habit influences the process. The researchers noticed that even when different operators used the same machine to cut the same material, they often chose very different speeds and feeds. One operator might be aggressive, pushing the machine to its limits, while another might be conservative, prioritizing safety over speed. These personal preferences created a wide range of data, which traditional rules often fail to capture. The new system, however, was able to synthesize this variety of human experience into a single, consistent recommendation. It does not force a single "correct" answer but rather suggests the most probable optimal settings based on the collective history of the shop. The researchers integrated this model into a two-step workflow. First, the system analyzes a new G-code program and suggests improved parameters. Second, it presents these suggestions to a human operator, who retains full authority to accept, reject, or modify the advice. This ensures that the final decision remains in human hands, respecting the operator's judgment while providing a data-driven starting point.

The study confirms that a machine learning model can effectively extract the accumulated knowledge of a manufacturing floor and apply it to new situations. By analyzing 251 operations across three distinct machines, the researchers demonstrated that they could predict machining parameters with a high degree of reliability, particularly for spindle speed and cut width. The system successfully identified that machine identity and tool diameter are the primary drivers of these decisions. While the model is not a perfect replacement for human intuition, especially in cases where the depth of cut is ambiguous, it offers a powerful tool for reducing the time spent on setup and standardizing the quality of production. The approach requires no hardware modifications and works with the existing software that factories already use. As the manufacturing industry moves toward more flexible and automated systems, this method offers a practical way to leverage past successes to improve future performance, ensuring that the wisdom of experienced operators is preserved and shared across the entire production line.