Who Is the Father of CNC Machine? Three Engineers, One Chain of Work

• 1
  Who Is the Father of CNC Machine? A Name and a Caveat
• 2
  John Parsons and the Punched-Card Origin of Numerical Control
• 3
  The MIT Servo Lab and the First Numerically Controlled Mill
• 4
  From Punched Tape to G-Code: Where the Computer Entered
• 5
  What the Fathers of CNC Actually Invented, in Machining Terms
• 6
  Who Did What: Three Contributors Compared

Ask ten machinists who the father of CNC machine is and most will say John T. Parsons. That answer is defensible. Parsons and his team at Parsons Corporation, working in Traverse City, Michigan, built the punched-card system that computed and cut the curved templates used to inspect helicopter rotor blades in the 1940s. His 1948 proposal to the U.S. Air Force for a digitally controlled contouring machine is what funded the next step. Without Parsons pushing the idea, the servo-driven mill would not have been commissioned when it was.

The caveat is that Parsons did not build a CNC machine. He built a numeric control system, a card reader, and a set of calculations. The machine that followed was designed by the Servomechanisms Laboratory at MIT, funded by the Air Force, and it cut metal in 1952. So the honest answer is a chain: Parsons defined the problem, MIT solved the servo problem, and the control builders turned it into a product.

Why engineers still argue about the title: the word 'father' implies one person, but numeric control needed three separate inventions to work. Position feedback, a stored program of coordinates, and a machine tool that could move on command. Each came from a different shop or lab. Parsons owned the first invention, MIT the second, and industry the third.

The problem Parsons faced in the mid-1940s was not machining. It was inspection. Helicopter rotor blades have a twisted, curved profile, and the check fixtures used to verify that profile had to be hand-filed from templates. Those templates were laid out by hand from coordinate tables, which took weeks and drifted from part to part. His fix was to compute the coordinate points on punched cards, then feed the cards to a machine that positioned a cutting head.

That machine was not a milling machine in the usual sense. It was a card-fed positioning system that drilled or cut a series of discrete points along a curve. The path was not smooth, but the coordinates were right. This is the first practical example of a stored program driving a machine tool. The program was a deck of cards, not a text file, but the logic is the same one a modern controller reads.

Parsons took the idea to the Air Force in 1948. The Air Force was buying swept-wing aircraft and needed contoured parts faster than hand layout could deliver. The contract that followed went to MIT, not to Parsons, because the servo problem was considered the harder engineering challenge. Parsons supplied the concept, the requirement, and the early hardware experience.

What Parsons did not have was closed-loop control. His system indexed to a position and stopped. It could not follow a continuous contour while the cutter was in the material at feed rate. That gap is the reason the MIT machine is usually treated as the first true NC mill and Parsons as the man who asked for it.

In 1949 the Servomechanisms Laboratory at MIT began work on a milling machine controlled by a stored program of coordinates. The team included graduate students and staff who had worked on radar tracking and gun directors during the war. Their experience was in servo loops, not in chip making, and that shaped the design. The project was funded by the U.S. Air Force, which wanted contoured airframe parts cut without a skilled tracer hand.

The machine was a modified Cincinnati Hydro-Teletype vertical mill. Three axes were driven by hydraulic servo motors. Position feedback came from resolvers, and the command data arrived on punched paper tape. The tape held a sequence of coordinate instructions, read block by block. The controller compared commanded position to actual position and drove the servo until the error was near zero. That is a closed loop, and it is the defining feature of NC.

First cuts were made in 1952. The machine could follow a curved path at a controlled feed rate, which no tracer or template system could do as accurately on a complex contour. The tape was fragile, the electronics filled several racks, and the whole setup was far too expensive for a job shop. None of that mattered. The concept was proven.

Why this matters to a working machinist: everything in a modern controller descends from that loop. A Fanuc or Siemens control still reads a block, compares it to an encoder reading, and corrects the servo. The numbers have changed. The loop has not.

For roughly two decades after 1952, NC machines were hard-wired. The controller could not be reprogrammed without changing the logic, and the tape had to be prepared on a separate machine. Changing a part meant making a new tape and often rewiring part of the control. Setup was slow and the machines were concentrated in large aerospace plants that could afford them.

The shift to CNC came when a general-purpose computer replaced the fixed logic. The first commercial CNC controller is usually credited to Fanuc, which introduced one in 1956, and the concept spread through the 1960s and 1970s. Now the control logic lived in software. A new part needed a new program, not a new circuit board. That single change is what pushed NC out of the aerospace plant and into general machining.

The programming language standardized as G-code, formalized in the 1960s and later published under ISO 6983. It survived because it describes motion in a way both humans and controllers can read. A line such as G01 X50.0 Y20.0 F250 tells the control to move in a straight line to a point at a feed rate. Every CAM system, from a small shop package to a full aerospace suite, posts to some dialect of that language.

The reason an engineer should care about the timeline is that machine behavior still reflects it. Old controls had limited look-ahead, so programmers used many short blocks and slowed down on corners. Modern controls look ahead hundreds of blocks and adjust feed automatically. Same G-code, very different motion.

Strip away the history and the inventions reduce to three ideas that a machinist uses every day. The first is the stored program: the part geometry lives outside the operator's hands, as data. Once geometry is data, it can be edited, copied, sent over a network, and verified before the first cut. That is the reason a shop can machine a part in Dongguan from a file written in Ohio.

The second is the closed servo loop: the machine measures where it is and corrects toward where it should be. This is the difference between a machine that repeats a position and a machine that holds a tolerance. A control with position feedback can compensate for thermal growth, tool wear, and cutting force. It is why a modern machine can hold ±0.005 mm (±0.0002 in) on a production run.

The third is interpolation: the control computes the intermediate points between two programmed coordinates so the tool follows a smooth path. Straight-line and circular interpolation are the common forms. On a five-axis machine the control also interpolates the two rotary axes, which is what allows a ball nose cutter to stay normal to a curved surface.

None of these ideas is about a single machine. They are about moving control of geometry from the bench to the controller. Once that move happened, accuracy stopped depending on how steady a hand was and started depending on how good the loop and the program were.