Zhixin bought 10 CNC and 10,608 robots in Gree

• 1
  Why Zhixin bought 10 CNC and 10,608 robots together
• 2
  How robot loading changes the CNC process window
• 3
  Grippers, nests, and the tolerance chain
• 4
  What 10,608 robots actually buy a plant
• 5
  Where CNC robot cells go wrong
• 6
  Keeping tolerance and finish in an unattended cell
• 7
  When to automate the CNC load, and when not to

The headline count is lopsided. Ten machining centers against ten thousand robots sounds like a rounding error on one side. Look at the numbers differently. In a Gree supply chain, one upstream machining cell feeds hundreds of downstream stations: press tending, welding, packing, leak testing. The 10 CNC machines are the bottleneck; the robots are the distribution network around them.

Ten CNC machines is a capacity statement about stiffness and spindle hours. Ten thousand robots is a statement about handling. When a manufacturer announces both at once, it means the machining step was removed from the manual queue and placed at the head of an automated line. Everything downstream now waits on spindle cycle time.

That changes what the CNC cell has to deliver. Surface finish and tolerance still matter, but so do chip evacuation timing, in-machine probing cadence, and part presentation at the door. A robot cannot compensate for a burr or a thermal drift the way an operator can. The process has to be self-correcting before the robot ever reaches in.

So the real question for an engineer reading this news is not how many robots. It is which parts justify a robot-loaded cell, and which parts should stay on a manned machine. That is what the rest of this page covers.

• 1
  CNC sets taktRobots scale around spindle cycle time, not the other way around.
• 2
  Handling is cheapSimple pick-and-place units cost far less than an extra machining center.
• 3
  The cell must self-checkAutomated loading removes the human who used to catch drift.

A manual operator loads a part, closes the door, watches the first cut, and adjusts. A robot does none of that. The load position repeats to whatever the gripper and nest can hold, typically a few hundredths of a millimeter, not micrometers. Tolerances below ±0.005 mm still depend on the machine, not the robot.

That is why robot-loaded cells lean on in-machine probing. A touch probe re-datums the workpiece after clamping, and the control shifts the work offset. Without that step, nest wear shows up as a slow drift in one axis, and it takes a few hundred parts before anyone notices a trend.

Chip control becomes a hard constraint. Robots do not clear nests. Long stringy chips in aluminum will wrap a gripper jaw and cause a misload that the door interlock may not catch. Through-spindle coolant, chip fans, and air blast are not optional in a lights-out cell; they are the difference between a stable week and a scrap bin.

Thermal behavior also shifts. Without an operator opening the door every few minutes, the machine reaches a steadier thermal state. That can improve size control on long runs, but it also means the first parts after a weekend stop are colder than mid-shift parts. Warm-up cycles are part of the process, not a formality.

• 1
  Probe every partRe-datum after clamping to absorb nest wear and casting variation.
• 2
  Break the chipPeck cycles and high-pressure coolant prevent gripper wraps.
• 3
  Warm upRun a dummy cycle before the first production part each shift.

The tolerance chain in a robot cell runs from the machined feature, through the nest, through the gripper jaw, to the robot flange. Each link adds error. A cast nest with 0.1 mm of wear moves the whole chain. A hardened, replaceable insert keeps it closer to 0.02 mm over a production run.

Robot repeatability is usually quoted at the flange under ideal load. Add a 5 kg part on a 300 mm arm extension and the effective repeatability drops. For heavy steel parts, a short-reach robot with a stiff wrist beats a long-reach unit with a higher payload rating. Reach costs accuracy.

Gripper choice follows part geometry, not preference. Three-jaw self-centering grippers suit round turned parts. Parallel grippers with soft jaws suit prismatic milled parts with a consistent datum face. Vacuum is fast but fails on rough or oily surfaces, and it gives no positive confirmation of grip.

Confirm the grip. A simple air-pressure or proximity sensor on the jaw tells the controller whether a part is present. Without that feedback, a dropped part becomes a collision, and a collision in a lights-out cell can run for hours before anyone walks by.

• 1
  Replaceable nest insertsHardened inserts hold position and are cheap to swap.
• 2
  Short reach winsStiffness matters more than payload headroom for heavy parts.
• 3
  Sense the jawGrip confirmation prevents silent misloads.

Robot count is a poor efficiency metric on its own. Two identical arms in a cell that shares one CNC machine do not double output; they split the same spindle hours. The useful ratio is robot hours per spindle hour, and it should sit near the handling time fraction of the cycle.

If a 6-minute cycle includes 40 seconds of load and unload, a single robot can serve several machines only if the travel between them is short. Put the machines 8 m apart and the robot spends more time walking than gripping. Cell layout decides whether the robot count is real capacity or decoration.

The bigger gain in a Gree-scale deployment comes from consistency, not speed. A robot loads the same way at 3 a.m. as at 3 p.m. That flattens the distribution of part sizes, which lets downstream gauging use tighter limits and catch problems earlier.

Cost per part falls mainly through unattended hours. A cell that runs through the night at reduced feed still produces parts that would otherwise wait for a morning shift. The savings show up in spindle utilization, not in labor rate alone.

• 1
  Watch robot-to-spindle ratioExtra arms do not add spindle minutes.
• 2
  Keep travel shortLong robot walks waste the cycle time you saved.
• 3
  Bank the night shiftUnattended hours are where the payback lives.

Most automation failures are not robot failures. They are fixture failures, chip failures, and sensing failures. A nest that wears 0.05 mm a week looks fine on Monday and produces out-of-tolerance parts by Friday. Scheduled nest checks catch it; a once-a-quarter review does not.

Tool wear is the second silent killer. An operator hears a change in cut sound and swaps an insert early. A robot keeps loading until the probe says the size drifted. If the probe only checks one feature, a worn tool on another feature can run for hours. Probe the tightest tolerance, not the easiest one.

Part mix is the third. Automation rewards repetition. A cell that runs five part numbers with different datums spends more time on changeover and re-teaching than it saves. Where the mix is genuinely variable, a pallet system with manual loading often beats a robot arm.

Plan the recovery path before the first part. What happens when a gripper drops a part inside the enclosure? Can an operator reach it without a ladder and a lockout? Cells that ignore this end up running only when someone is standing nearby, which defeats the purpose.

• 1
  Check nests on a scheduleWear is gradual and invisible until parts fail.
• 2
  Probe the tightest featureEasy-to-reach features hide tool wear.
• 3
  Design the recoveryA cell nobody can clear safely is not unattended.

Holding ±0.005 mm over a night shift requires the machine to be stable and the metrology loop to be closed. In-machine probing after roughing and again after finishing gives the control two chances to correct. A single probe pass at the end only reports the error; it does not fix it.

Finish targets shift with automation too. Ra 0.8–1.6 μm is a realistic in-cell target for most aluminum and steel parts with the right insert and coolant. Getting to Ra 0.2–0.8 μm usually means a separate finishing pass or hand work, and hand work breaks the unattended flow.

Coolant concentration and filtration matter more when nobody is watching. A clogged filter raises temperature and changes size. A drop in concentration promotes rust on steel parts that sit in a nest overnight. Both are cheap to monitor and expensive to ignore.

Document the offsets. When a cell is re-datumed, the work offset changes. If those changes are not logged, a size trend across a week is impossible to reconstruct. A simple log tied to the machine control is enough.

• 1
  Probe twiceOnce after roughing, once after finishing.
• 2
  Set finish realisticallyRa 0.8–1.6 μm suits unattended cell work.
• 3
  Log the offsetsTrends need a record, not memory.