How Heidenhain machine tools are collected in the MES system via the 4G network

• 1
  Key takeaways
• 2
  Where a Heidenhain machine tools MES system actually gets its numbers
• 3
  What the 4G leg adds to the Heidenhain machine tools MES system
• 4
  The edge buffer is the part that makes the Heidenhain machine tools MES system survive
• 5
  Antenna placement and the physical limits of a 4G link
• 6
  When the Heidenhain machine tools MES system should stay wired
• 7
  Transport choice by data type

Every Heidenhain machine tools MES system starts at the same place: the control. A TNC 640 or TNC 7 keeps a state model in memory. It knows whether the spindle is turning, which program block is active, how many parts have run, and what the active tool offset is. None of that is written for your MES. It is written for the operator panel.

The MES gets data through one of three doors. The first is the OPC UA server built into the newer controls, which exposes a structured address space of nodes. The second is MTConnect, which presents the same information as a read-only HTTP stream of XML or JSON. The third is a vendor gateway or a small edge box that polls the control and republishes the values.

The 4G network does not change any of this. It is a transport. A router with a SIM card takes the Ethernet frames from the control or the edge box and puts them on the cellular network. On the other end, your MES or an IoT broker receives them. If the protocol layer is wrong, faster radio will not fix it.

This distinction matters when you size the link. People ask about 4G speed. The real question is how many nodes you poll and how often. A TNC pushing 200 nodes every second is a different load from one pushing 40 nodes every ten seconds. The radio rarely decides whether the system works. The polling design usually does.

• 1
  Control-level dataProgram name, block number, feed override, spindle speed, active tool, alarms.
• 2
  Counter dataPart count, good count, scrap count, cycle time per program run.
• 3
  State dataRunning, idle, setup, alarm, maintenance. The four states MES cares about most.
• 4
  Not from the controlSpindle vibration, coolant concentration, air pressure. Those need separate sensors.

A wired link between the control cabinet and the MES server is a known quantity. You run Cat 6, you test it once, and the round-trip is under 2 ms for the rest of the decade. A 4G leg is not that. It is a shared radio resource, and your packets compete with whatever else is on the cell.

In a quiet plant, a 4G round trip to a cloud broker usually lands between 30 ms and 80 ms. At shift change, or when a nearby site is doing something heavy, spikes of 300–800 ms appear. Very rarely you get a multi-second stall while the modem reattaches to a different cell. That is normal cellular behavior, not a fault.

For MES collection this is mostly fine, because most MES data is not urgent. A part count that arrives 400 ms late is still a correct part count. The data has to be complete and in order, not instantaneous. That is a different requirement from motion control, and it is why 4G works here at all.

The exception is alarm handling. If a tool breaks and you want the MES to stop the next job before it starts, the whole chain — detection, publish, broker, MES rule, response — has to fit inside the time it takes to load the next part. On a 40-second cycle that is tight. On a 6-minute cycle there is room.

• 1
  Typical quiet-cell RTT30–80 ms to a regional cloud endpoint.
• 2
  Busy-cell RTT300–800 ms spikes, still usable for counters.
• 3
  Worst caseMulti-second stall during cell reattach. Buffer for it.

No cellular link is continuous. Handovers happen, signal dips behind a steel wall, a tower gets maintenance. If your edge software writes directly to a remote database on every event, a 30-second outage becomes a hole in the production record that nobody can reconstruct.

The fix is boring and effective. Run a small service on the edge box that subscribes to the control and appends every event to a local store with a timestamp and a sequence number. A separate sender drains that store to the MES whenever the link is up. The MES sees a pause, then a burst, and the record is complete.

Size the buffer by the worst outage you are willing to tolerate. A 60-minute buffer on a machine that produces one event per second needs about 3,600 records. That is a few hundred kilobytes. Storage is not the constraint; the constraint is that you must test the replay path before you trust it.

Sequence numbers matter more than timestamps here. Machine clocks drift, and cellular time sync is not always applied. A monotonic counter lets the MES detect a gap and request a resend. Without it, you cannot tell a quiet machine from a lost packet.

• 1
  Local store firstAppend events to disk before any network call.
• 2
  Monotonic sequenceLets the MES detect and request missing records.
• 3
  Replay testPull the antenna for 10 minutes. Check the record closes.

Most failed installations we hear about are radio problems, not software problems. A modem in a metal cabinet behind a closed door is a poor receiver. Machining centers are large steel structures, and the shop floor is full of coolant mist and chips that do not help.

Put the antenna outside the cabinet, on a bracket, with a clear view of the ceiling or a window. Keep the coax run short. If the modem reports a signal strength below about -100 dBm RSRP, expect retransmits and longer stalls. An external antenna with 5 dBi of gain often moves that number by 10 dB or more.

Multi-path is the other issue. A signal that bounces off a steel wall and arrives slightly late will corrupt the symbol. Moving the antenna half a meter can change the link more than any setting in the router. If you have two operators complaining about the same machine at the same time of day, look at the antenna before you open a ticket with the carrier.

For a plant with several machines, one modem per machine is simpler to reason about than a shared modem with a switch. Machines have different data rates and different criticality. Sharing a link means one noisy machine can delay the others. The cost of a second modem is small next to the cost of explaining a gap in the production record.

• 1
  RSRP targetBetter than -100 dBm. Below that, expect retransmits.
• 2
  Antenna locationOutside the cabinet, clear of the machine frame.
• 3
  One modem per machineSimpler isolation when one link misbehaves.

4G is the right answer when the machine is hard to cable. A rented bay, a temporary cell, a machine that moves between two lines, a building where the owner will not let you cut a trench. In those cases the 4G leg turns a two-week cabling project into a one-day install.

It is the wrong answer when the MES drives closed-loop decisions at cycle speed. If your rule engine adjusts feed or stops a toolpath based on data that must arrive in under 20 ms, do not put a radio in that path. Run fiber or copper, and keep the 4G link as a monitoring channel only.

It is also a poor fit when the data volume is large and continuous. Streaming high-frequency vibration or full servo traces over cellular is expensive and fragile. The right pattern is to compute a feature at the edge — RMS, peak, kurtosis — and send the small result. Raw traces stay local.

A practical middle ground is a hybrid. Wire the machines that sit near the cabinet room. Use 4G for the outliers. Both feed the same MES with the same schema, so the plant view is uniform even though the transport is not.

• 1
  Good fitCounters, states, alarms, OEE, remote monitoring.
• 2
  Poor fitClosed-loop control, servo traces, high-rate waveform streaming.
• 3
  HybridWired where possible, 4G for the machines that cannot be reached.