Mitsubishi CNC Data Acquisition and Mooring at an MES Platform

• 1
  What Mitsubishi CNC data acquisition actually exposes
• 2
  How the mooring layer sits between control and MES
• 3
  Why MES mooring changes what the numbers mean
• 4
  Where Mitsubishi CNC data acquisition stops being useful
• 5
  Step by step: from one machine to a moored MES feed
• 6
  Polling rate and mooring choice by data type

A Mitsubishi CNC control keeps two very different classes of information. The first is machine state: run, stop, alarm, program number, spindle and feed override. The second is process data: spindle load, axis position, feedrate, tool number, alarm code, cycle time. Most people ask for both at the same rate. That is the first mistake, because state changes are events and process values are a stream.

State registers are small and change rarely. Polling them every 500 ms is usually enough to build an accurate timeline of what the machine was doing, and a 500 ms poll costs almost nothing on the control bus. Sampling spindle load at the same 500 ms is a different story: a load spike during a 12 mm roughing pass can last under 100 ms, so a slow poll will average it away and the chart will look calm while the tool is not.

Cycle time is the value everyone argues about. On a Mitsubishi control a cycle can end for several reasons: program end, M00 stop, alarm, or an operator pause. If the mooring layer collapses all four into one "cycle complete" event, your OEE denominator drifts and nobody can explain why the morning shift reports 92% and the afternoon shift reports 71% on the same part.

Tool data is the most underused register set. Tool number, offset number and remaining life are already sitting in the control. Pulling them into the MES removes a manual step at the machine and gives you a tool-life curve that matches real cutting hours instead of the operator's memory.

• 1
  State vs streamPoll state every 500 ms; sample load and feedrate faster than the events you care about.
• 2
  Cycle end has causesProgram end, M00, alarm and operator pause are not the same event.
• 3
  Tool life lives in the controlTool and offset numbers are already there; do not retype them.

A mooring layer is not a protocol converter. It is a small service that owns three jobs: read from the control, hold the data when the network drops, and deliver it to the MES platform in a shape the MES already understands. Vendors often sell this as a gateway box. The box matters less than the buffering logic inside it.

Networks in a machine shop drop. Someone unplugs a switch to move a cart, an IP address changes after a controller swap, or the MES server restarts for a patch. If the edge service writes straight through, every drop becomes a gap in the timeline. A local buffer of 30-60 minutes of sampled data covers almost every realistic outage on a shop floor. Past an hour you are usually dealing with a real fault, and the MES should be told, not fed stale numbers.

Timestamps are where most integrations quietly fail. There are at least three clocks involved: the control clock, the edge device clock and the MES server clock. If the edge device stamps a sample when it is sent instead of when it was read, a 20-minute network outage shifts the whole block forward and the MES will show a machine cutting parts during a shift change. Stamp at the point of read, and keep the control clock disciplined.

Data volume is smaller than people expect. A single Mitsubishi machine at a 500 ms state poll and a 100 ms process sample produces on the order of a few hundred kilobytes per shift once you compress and only send changed values. The hard part is not bandwidth. It is agreeing on what each tag means before the first line of code is written.

• 1
  Buffer 30-60 minutesEnough for normal shop-floor outages without hiding a real fault.
• 2
  Stamp at read timeNot at send time, or outages shift the whole block.
• 3
  Share the tag dictionaryAgree on tag names and units before coding starts.

Once Mitsubishi CNC data acquisition is moored to an MES platform, a machine stops being a standalone asset. Its state, cycle counts and alarms sit next to work orders, material lots and inspection results. That is the point. The same spindle load curve means something different when it is attached to a specific part number and a specific tool offset.

The engineering payoff is traceability at the process level, not the part level. If a batch of parts fails a dimensional check, the MES can show whether the cycle time on that run was shorter than usual, whether a tool change happened mid-batch, or whether the machine was in a degraded state. Those are process signals that a paper traveler never captures.

There is a boundary worth stating. An MES platform records what the control reports. It does not measure the part. A machine can report a clean cycle while the part is drifting toward the tolerance limit, because the control only knows the commanded position and the following error, not the finished dimension. Keep the metrology loop separate and compare the two later.

The same limit applies to our own shop. We run 127 high-precision CNC machines across three wholly-owned plants, and machine data tells us when a spindle is working hard. It does not tell us whether a bore is at Ø25.000 mm or Ø25.012 mm. That still comes from inspection, which is why every part is measured before shipment.

• 1
  Process-level traceabilityMachine state next to work order, lot and inspection result.
• 2
  MES records, it does not measureControl data is not metrology data.
• 3
  Two loops, compared laterKeep machine monitoring and dimensional inspection separate.

Not every machine belongs on the MES. A manual mill, a wire EDM used twice a month, or a machine that only runs one long job has little to gain from a 500 ms feed. The integration cost is real: a gateway, a network drop, a tag map, and someone to maintain it. If the machine does not drive a scheduling or quality decision, skip it.

Older controls are the second boundary. Some Mitsubishi models expose a full register set over Ethernet. Others only give you a serial port, or require an option board that is no longer sold. Before promising a customer a live dashboard, read the control model and firmware version and confirm which interface is actually enabled. This one step prevents most failed projects.

The third boundary is culture, not hardware. If no one on the floor is going to look at downtime reason codes, the MES will fill with "other" and the data will be worthless within a month. Start with two or three reasons that operators can pick in one tap, and let the list grow only when the floor asks for it.

• 1
  Skip low-value machinesIf it does not drive a decision, do not instrument it.
• 2
  Check firmware firstConfirm the interface is enabled before promising a dashboard.
• 3
  Start with two reason codesA short list operators will actually use beats a complete one they ignore.