Mazak Machine Tool Data: How Acquisition Actually Works

• 1
  Key takeaways
• 2
  What Mazak Machine Tool Data Actually Contains
• 3
  MTConnect, MT-LINKi and FOCAS Compared
• 4
  Sample Rate, Timestamps and the Cost of Polling
• 5
  Where Acquisition Stops Being Useful
• 6
  Which Acquisition Path Fits Which Shop

A Mazak machining center produces two very different classes of signal. The first is controller state: run mode, program number, tool number, spindle speed, feed override, alarm code. The second is physical state: spindle load, axis position, servo current, coolant pressure, pallet status. Controllers publish the first class natively. The second class is usually derived, and how it is derived depends on the control generation.

On SmoothX and SmoothG controls, Mazak exposes an MTConnect agent and a REST interface. On Matrix and Matrix 2 controls, MT-LINKi is the normal route, with MTConnect available on later firmware. On older Fusion 640 controls, you are typically limited to FOCAS, serial output, or I/O relay taps. This is the single biggest constraint on any project. Three machines on the same floor can require three different acquisition methods.

The practical question is not what the control can emit. It is what your system can use. A CMM program with 200 tags sounds impressive until the MES only consumes 12 of them. Start from the downstream use case — OEE, tool-life tracking, alarm response, energy reporting — and work backward to the tag list.

• 1
  Native tagsRun state, program, tool, spindle speed, alarm, override values.
• 2
  Derived tagsCutting time, cycle count, utilization, load trend, energy estimate.
• 3
  Manual-only tagsSetup notes, fixture ID, operator comments, inspection results.

MTConnect is an open, read-only, XML-based standard. It is vendor-neutral, so one collector can read a Mazak, a Haas and a Brother without three separate drivers. The trade-off is tag coverage. MTConnect exposes what the control chooses to model, and Mazak's agent does not surface every internal register.

MT-LINKi is Mazak's own monitoring platform. It reaches deeper into the control, including maintenance counters, spindle and axis load history, and some servo data that MTConnect does not map. The cost is vendor lock-in. Moving to another machine brand means a second system, and the data model is Mazak-specific.

FOCAS is the older Fanuc-based route, still relevant on legacy Mazak turning and milling machines. It reads registers directly, which gives flexibility and also risk. FOCAS libraries change between control versions, and a firmware update can break a working collector. Firmware change control matters as much as the software itself.

• 1
  Choose MTConnectMixed-brand floor, standard tags, no vendor preference.
• 2
  Choose MT-LINKiAll-Mazak floor, deep maintenance and load data needed.
• 3
  Choose FOCASLegacy controls, specific registers, in-house software team.

Most Mazak collectors poll the control once per second. That is fine for OEE, because utilization is a slow number. It is not fine for stop analysis. A 1 s poll can miss a 700 ms tool-change stall completely, and those stalls are exactly what a cycle-time project is trying to find.

Pushing to 100 ms changes the architecture. You need a buffer at the machine or in the cell, because a network hiccup at that rate loses data rather than delaying it. Store-and-forward on an edge device is the usual answer. It also means your database write path must handle 10× the volume for the same number of machines.

Timestamps are the quiet failure mode. If the control clock drifts from the server clock, events land in the wrong order and every downtime report is wrong. Run NTP on the control network and check drift monthly. A 2 s offset is enough to misattribute a stop to the wrong shift.

• 1
  1 s pollEnough for OEE, utilization and shift reporting.
• 2
  100 ms pollNeeded for short stops, tool change and load spikes.
• 3
  Clock syncNTP on the control VLAN; verify drift every month.

Data acquisition does not fix a process that was never measured properly. If your fixture is wrong or your offsets drift, more tags will just describe the problem faster. Fix the machining first, then instrument it.

There is also a hard limit on what a controller knows. It cannot tell you whether a bore is on size, only that the cycle finished. Metrology data has to come from a CMM, a gauge, or an in-process probe, and it has to be joined to the machine record by part serial or work order. Without that join, you have two disconnected data sets.

Security is the other boundary. Opening a control network to a collector creates a path into production equipment. Keep the acquisition VLAN separate from the office network, allow only outbound connections from the edge device, and log every write attempt. The reading side is usually safe. The configuration side is where risk lives.

Finally, consider who maintains it. A dashboard nobody owns decays within a quarter. Assign an owner, define the tags that matter, and review the tag list twice a year. Fewer, trusted tags beat a large set that nobody has validated.

• 1
  Fix the process firstInstrumentation amplifies whatever is already happening.
• 2
  Join to metrologyLink machine records to inspection by serial or work order.
• 3
  Segment the networkAcquisition VLAN, outbound-only edge, logged changes.
• 4
  Name an ownerOne person accountable for tag accuracy and review.