In this guide 6 sections
Ask which controllers the vendor has already connected, how data leaves each machine, what happens when a link drops, and how run hours land on the job cost. Insist on a manual booking fallback for older machines. Request a demonstration on your own controller, not a slide.
Which controllers will it actually talk to?
List every machine on the floor with its make, controller and year. A Fanuc-controlled VMC, a Siemens lathe and a grinder with a relay panel are three different connection jobs. Hand the vendor that list before any quote arrives.
Ask for reference installs on the same controller families. A vendor who has connected Fanuc but never Heidenhain will learn on your floor, and that learning costs you weeks. Say so plainly.
Then ask what the connection reads. Some options catch only cycle start and stop, while others read spindle load, program name, part counter and alarm codes straight from the controller.
- Make, model and controller of each machine
- Protocol or adapter used, such as MTConnect, OPC UA or a signal tower tap
- Fields read: cycle state, program, part count, alarms
- Machines that need extra hardware and who supplies it
What happens when the link drops?
Shop networks fail. A switch gets unplugged, a machine is rebooted, or a Wi-Fi bridge loses signal behind a haze of coolant mist. Ask whether the connector stores readings locally and sends them later.
Without buffering, the gap turns into silent idle time on the utilisation report, and you end up blaming the operator for a fault in a cable. Have the vendor stage a gap during the demo, and look at how the report marks it.
Ask who is alerted. An alarm nobody sees is no better than none. Also ask how clocks are kept in step. If the machine, the server and the tablet disagree by minutes, a cycle can land on the wrong shift or the wrong job.
Keep a manual booking path
Some machines will never connect cleanly. A manual lathe, a bench grinder or a machine with a dead controller board still earns money. The software must let the operator book start, stop and quantity on a tablet.
Check that manual and automatic entries land in the same job record. Two screens mean two versions of the truth. Ask how the system flags a manual entry that overlaps a machine reading from the same hour.
Also ask what an operator does when the tablet is busy. A queue at the terminal brings the paper slip straight back.
- Operator login by badge or PIN
- Reason codes for stops, set by you
- Edit rights and an audit trail for corrected times
- Handling of overlapping manual and machine entries
How do run hours reach the job cost?
This is the question the sales deck skips. Machine data only matters if each cycle is tied to a work order, an operation and a machine rate. Ask exactly how the system knows which job a cycle belongs to.
Usually the program name or an operator scan links the two. Find out how the system copes if a program is reused across jobs, or if a proving run uses the same code as production. Ask where setup time sits against run time.
Then request one finished job on the demo. Planned hours, actual hours, scrap and cost per piece should sit together on a single screen, with no export to a sheet.
Ask for a trial on your own machine
A pilot on two machines of different brands tells you more than any reference call. Keep it running through a full month, and compare the readings with the operators’ own booking sheets and the shift supervisor’s memory of the week.
Agree before it starts what success means. Gaps under a stated length, correct job linking and a report your planner opens every morning are fair tests. Vague hopes are not.
Write down who owns the cabling, the network ports and the late-night support call. Settle that before you sign.
Machine data vendor question sheet
One printed page that gives each question its own row, with four columns: question, vendor answer, proof shown in the demo, and a mark out of five. Rows are grouped under controllers, link failures, manual booking, job costing and support. A header block holds your machine list, with the controller type and the connection method agreed for each. Fill it in during the demo, not afterwards.
Open a blank worksheet to printQuestions owners ask
Do older machines need new controllers to be monitored?
Often not. A signal tower tap or a current sensor can read running and stopped states. You will not get program names or alarm codes that way, so job linking may rely on operator booking.
Who should be in the vendor demo?
Bring the planner, one setter and whoever owns your network. The planner checks job linking, the setter checks the tablet at the machine, and the network owner checks ports, firewalls and who can reach the data.
Can machine data replace operator booking?
Not fully. The machine knows when it ran. It does not know why it stopped or which job a proving cycle served. Keep short reason codes for stops.
How long should a pilot run?
Long enough to include a changeover, a night shift and a network hiccup. A few weeks on two machines usually shows all three.