Who Make First CNC Machine? The Real Origin of NC and CNC

• 1
  Who Make First CNC Machine: The Team Behind the 1952 Mill
• 2
  How a Punched Card Actually Moved a Cutter
• 3
  From NC to CNC: What the Computer Changed
• 4
  What 70 Years of Evolution Bought the Machine Shop
• 5
  NC, Early CNC and Modern CNC Compared

People ask who make first cnc machine as if one name would settle it. The record points to a partnership. John Parsons ran Parsons Corporation in Traverse City, Michigan. In the late 1940s his shop was milling helicopter rotor blade templates and aircraft turbine parts for the U.S. Air Force. Hand-fed cranks could not hold the contour tolerances the Air Force wanted, and every part drifted a little from the last one.

Parsons did not set out to invent a new machine category. He set out to stop trusting human hands. His idea was to encode coordinate positions on punched cards and let motors drive the table to those positions. The Air Force funded the work, then pushed the harder part of the problem toward a university lab.

That lab was the Servomechanisms Laboratory at MIT. In 1952 the team there showed a modified vertical milling machine driven by servo motors on three axes, X, Y and Z. Punched cards carried the instructions. The machine is usually called the first NC machine tool, and it is the correct answer to the question of who make first cnc machine, with the caveat that it was a project, not a person.

The word order trips people up. The machine was not computer numerical control. It was numerical control, NC. The controller could not store a program, could not do math, and could not compensate for tool wear. It read a card and moved. That is all.

A punched card is not a program in the modern sense. It is a fixed sequence of position commands. Each hole pattern tells the control which axis to move and how far. The control converts that into a voltage, the servo amplifier drives the motor, and the leadscrew turns.

The loop closes at the motor, not at the cutter. There is no feedback from the workpiece, no probing, and no adaptive adjustment. If the cutter deflects under load, the machine never knows. If the room warms up and the leadscrew grows, the machine never knows. The only correction available is one the operator makes by hand between setups.

This is the practical boundary of NC. Position accuracy depends on the mechanical build of the machine and the quality of the card. Two identical cards on two different machines can produce two different parts. On a warm afternoon a long part can drift outside tolerance while the control reports nothing wrong.

It also explains why aerospace drove the work. Rotor blade templates and turbine components had to match a master, not just each other. Repeatability across a batch mattered more than raw speed. NC delivered repeatability by removing the operator's wrist from the feed crank, which was the whole point.

NC and CNC differ in one place: where the program lives. NC reads a fixed medium, usually a punched tape or card deck. Change one coordinate and you punch a new tape. CNC keeps the program in memory. The control can store several programs, edit a feed rate at the panel, and run a tool path that no tape could describe.

Memory changes what is possible, not just what is convenient. A stored program can hold cutter compensation tables, tool length offsets, and subroutines. That means one program can cut a family of parts, and a tool change no longer invalidates the geometry. It also means the control can pause, probe, and resume.

The shift landed through the 1960s and 1970s as minicomputers became small enough and cheap enough to bolt onto a machine tool. The mechanics did not change much. The intelligence did. A 1975 CNC mill and a 1955 NC mill use the same ballscrew, the same spindle, and roughly the same servo idea.

That distinction still matters when you read a quote. A shop with modern controls can compensate for tool wear between parts. A shop running an old control on a tape reader cannot. The tolerance on the drawing does not tell you which one you are buying. The machine controller does.

The gains since 1952 fall into three buckets: axes, feedback, and the cutting envelope. Axes came first. Three axes can reach a face. Five simultaneous axes can reach a face at an angle without a second setup, so an undercut or a blended fillet stays on one coordinate system.

Feedback came next. Modern controls read glass scales and encoder counts thousands of times per second. Thermal growth gets modeled, not ignored. At GreatLight, 16 simultaneous 5-axis machining centers run alongside 27 three-axis machines and 16 mill-turn centers, and the controller compensates for tool wear and spindle growth on the fly.

The envelope grew too. Maximum processing size reaches 4,000 mm, with travels of 4,000 × 400 × 150 mm on the large frame and 750 × 1,150 × 550 mm on the medium frame. A Ø400 mm rotary table handles parts that must be indexed between faces.

Tolerance is where the story lands. GreatLight holds ±0.005 mm (±0.0002 in) on qualified features, with surface finish from Ra 0.2–0.8 μm on fine finishes to Ra 1.6–3.2 μm as machined. None of that was reachable with a card reader. All of it is routine on a machine that stores a program.