Connecting a CNC to an OEE system is a sequence, not an installation.
The six steps in the order they actually happen, the questions to settle at each, and the four mistakes that turn a two-week project into a two-month one.
Connecting a CNC to an OEE system is usually described as a software task. In practice it is a sequence of small decisions — which machines can be reached, what counts as a cycle on each of them, who enters a reason when one stops — and the projects that stall are the ones that treated it as an installation rather than a sequence.
This article walks that sequence in order, with the questions to settle at each step and the mistakes that cost time.
Six steps, in the order they actually happen.
Read left to right: inventory precedes reach, reach precedes acquisition, and verification closes the loop. Skipping step 04 is the single most common cause of an OEE figure nobody believes, because the plant and the software are counting different things.
Walk the floor with a list and mark every machine into one of three groups: machines with a control that can be reached over the network, machines with a PLC or instrument that can be read, and machines with nothing readable at all. That single count determines the shape and cost of the whole project, and it is a half-day of work that saves weeks.
Record the control family and generation while you are there. Two machines of the same make bought six years apart can differ substantially in what they expose, and finding that out during commissioning rather than during planning is how schedules slip.
Do not assume the newest machines are the easiest. A machine bought with a minimal option package can expose less than an older one that was specified generously.
Networked controls need an IP address on a network the monitoring server can reach. In most plants this means an existing switch and a cable; occasionally it means a small amount of cabling work, and where that is impractical the machine is treated as though it had no control at all.
Machines with no readable control are reached with a gateway device wired to whatever signals exist — a contactor, a stack light, an existing counter. Uplink can be Ethernet, Wi-Fi or 4G, which matters because structured cabling rarely reaches every corner of an Indian plant. The engineering detail is in digital I/O machine monitoring.
This is the step vendors talk about most and the customer needs to think about least. MachineWise engineering configures acquisition for the specific control in front of it and confirms what that machine actually returns; integration is read-only, no control parameter is modified, and nothing is written to the machine.
What matters at this step from your side is one thing: insist that the confirmation is done on your machines rather than accepted from a compatibility list. A datasheet describes a control family. Your machine has a generation, an option package and a service history.
The step that decides whether anyone believes the number.
An OEE system has to know what one part looks like. On some machines that is obvious. On many it is not: a program that makes four parts per cycle, a machine that dwells mid-program while an operator gauges a feature, a cycle interrupted for chip clearing and resumed. If the definition is wrong, availability and performance are both wrong, and the resulting OEE is precise and useless.
Settle it explicitly, machine by machine, and write it down. The failure modes and how each is handled are covered in automatic cycle detection explained.
A machine can report that it stopped and for how long. It cannot report why. That single input is the difference between a downtime record and a downtime analysis, and it is where most projects lose the floor.
The design rule is that it must take two taps at the machine and use words the operators already use. Fifteen fields, free text or a tablet in the supervisor's office all produce the same outcome within a month, which is nothing. Where reason coverage genuinely matters, enforced capture holds cycle-start until a reason is entered.
Before the data is presented to anyone as fact, watch one machine and its dashboard side by side for a full shift. States that disagree with reality get corrected then, not after a supervisor has publicly disputed a number.
Expect the first honest measurement to be lower than the plant's assumption. That is not a fault in the system; it is the point of installing one. The useful output of week one is not the score but the shape of the gap — which machines, which times of day, which repeating causes.
Four failure modes, all avoidable.
Buying before inventorying
A quotation built on an assumed machine mix is renegotiated during commissioning. Count first.
Leaving cycle definition to the software
Defaults are guesses. On multi-part or interrupted cycles they are wrong, and the OEE is wrong with them.
Making reason entry a form
Anything longer than two taps stops being used, and a partial Pareto is worse than none because it looks complete.
Going plant-wide first
Two machines for thirty days answers every assumption in the plan at a fraction of the disruption.