← THYNKTA
NOTES FROM THE MILL

What broke last time

Why I stopped building a work log and started building a medical record for machines.

A kilang is about a hundred machines that must all run together. Every one of them will break. That part was never the question.

The question that actually costs you comes after: what broke last time, and what did we do about it?

At our kilang that answer sat in three places. The logbook in the bengkel, if you could still read the handwriting. A WhatsApp group, four hundred messages back. And somebody's memory — usually the one who transferred out in March.

None of it can be searched. That is the whole problem in one sentence, and that is why I started building.

What I got wrong first

My first instinct was to build a system for recording kerja. Log the fault, assign it, close it. A ticketing system wearing maintenance vocabulary.

That was wrong, and it took me a while to see why. Recording kerja tells you what your people did. It tells you nothing about your machines. And the machine is what stops production.

The change came when I stopped treating it as a work log and started treating it as a medical record.

Every machine should have a chart. You open it and you see what has gone wrong with this one, how often, and what was done — like a doctor pulling a patient's history before deciding anything.

Once you have that, something useful comes out for free. A screw press that used to fail every 38 days starts failing every 20. Nobody reports that — no single work order looks alarming on its own. But the chart says it plainly, months before the machine strands you in the middle of peak crop.

The constraints came from the kilang, not the code

Almost every decision that mattered came from how the kerja actually happens, not from anything technical.

Electrical faults must go to a certified chargeman. That is not preference, that is competency, and the system should not allow a wrong assignment in the first place. So when a fault is tagged electrical, only chargemen appear. The mistake stops being possible instead of being caught later.

We work berdua on hot jobs, for safety. So a job can be assigned to two people, and when it is assigned to one, that shows. I did not build it to check on anybody. If you are already recording who did the kerja, you may as well be able to answer whether the buddy system was followed.

And a juruteknik will not use anything that takes more than a minute. That one fact killed more of my ideas than any technical limit.

Building for people who never asked for an app

Twenty people use it now. Not one of them asked for it. That changes what you build.

It runs on the phones they already carry, because telling a fitter to install something from an app store is telling him not to bother. Bahasa for the floor, English for the office, because otherwise half the crew quietly works around it. Everyone signs in once and stays signed in, because a password prompt at 2am is a reason to write on paper instead.

None of that is clever engineering. It is just paying attention to who is holding the phone.

What it is worth

Every maintenance argument comes back to one number, and most kilang have never actually put it on paper. Move the sliders to your own plant:

What an hour of downtime costs

Crude oil throughput value, at your own figures.

45 t/hr
20.0%
RM 4,000/t
Per hour stoppedRM 36,000
A four-hour stoppageRM 144,000

Crude oil only — kernel is not counted, so the real figure is higher. This is throughput value at risk, not profit lost: a mill with spare capacity catches some back later, one running flat out in peak crop does not. Either way it is the number worth knowing before deciding whether a maintenance record is worth keeping.

To another kilang engineer

You already know the constraints that matter. That is not a small thing. It is exactly the part generic maintenance software gets wrong, because it was written by people who have never had to stand in front of a manager and explain why the press is down and nobody can say whether the same thing happened last month.

The software you can learn. The kilang knowledge is the hard part, and you already have it.

MaintenXLive

The system this is about — maintenance and per-machine repair history.

READ MORE →