Retrofitting: the sensor decides the question you can ask

The situation

The pilot ran in November on 24 machines, chosen because they were the quickest to fit. Talvin Systems clamped a current sensor round the motor lead of each one. No bracket, no magnet, no downtime, an hour for the lot.

The pilot report landed on the owner's desk with two findings.

1. The 18 servo machines were running at 9.3% utilisation. The report called this "a critical under-use of assets".
2. The 6 clutch machines were running at 97.8%. The report called these "the benchmark for the floor".

Weighted together the pilot said the floor was at 31.4%, and it recommended an urgent study of the servo lines.

Both findings are wrong, in opposite directions, and the cause of both is the sensor. Two weeks and eleven management-days went into investigating a servo problem that did not exist. That is the cheap version of this mistake. The expensive version is buying machines on it.

Why one clamp gave two different lies

A current sensor measures the current in the motor lead. What that means depends entirely on what the motor does when the operator is not sewing.

• A servo motor stops when the treadle is released. Current falls to near nothing between seams. So current-on follows the needle. And the needle on a sewing machine is genuinely moving for only a small part of the day: 9.3% on machine 214, and there is nothing wrong with that. The rest of the operation is picking up, aligning, turning corners, trimming and disposing, and all of it is inside the standard minute. A sewing machine showing 80% needle time would mean the operator had stopped handling the work, which is impossible.
• A clutch motor spins from the moment the machine is switched on until the moment it is switched off. A mechanical clutch couples it to the head when the treadle goes down. Current is high all day, whether or not a stitch is formed. So a current clamp on a clutch machine reports something very close to "the switch is on", which on a manned line is very close to 100%.

The pilot's servo number was a real measurement of a thing nobody should manage on. The pilot's clutch number was not a measurement of production at all.

Retrofit is the normal case, and it constrains everything

A garment factory is not a new plant with machines wired to a common data bus. Sumilon's 412 machines were bought over nineteen years from seven makers. Most have no data port of any kind. Several of the older heads have no electronics at all. And no two generations offer the same thing. So the practical question is never "what does the machine publish?" It is "what can I attach to it without changing it?"

395 machines instrumented, 17 blind, USD 48,514 fitted. Note what the table settles. The pickup was chosen for both the servo and the clutch machines not because it is better in general, but because it is the only instrument that counts cycles on a clutch machine. The sensor followed the question.

Choose the sensor by the question, and write down what you have given up

This is the table to build before anybody quotes anything.

Half the rows say nothing here, and that is the honest answer rather than a gap in the table. It is also the reason lesson 4 exists. The most valuable machine data in a garment factory is downtime with a cause attached, and no sensor on this list can supply the cause.

The 17 blind machines, and the shadow they cast

Sumilon's blind machines are 8 bartacks, 4 eyelet presses, 3 heat-transfer label presses and 2 drawcord tippers. Instrumenting them meant a proximity sensor on a moving ram and a guarding review, at USD 310 each, or USD 5,270 for the set. Arnel Bacalso deferred it, and on the arithmetic of the day he was right: they are 4.1% of the fleet, and they were the most awkward 4.1%.

Then look at what happened. The 4 eyelet presses are operation 88, hood-cord insertion, and in April they starved the finishing end for three days. The programme found it anyway. Not from the presses, which reported nothing, but from the eleven instrumented machines downstream of them, which spent the afternoon logging long gaps with nothing to sew.

That is the general rule about a hole in the coverage. A blind machine is still visible in the shadow it casts on the instrumented ones. It follows that the machines worth instrumenting first are not the ones you most want to watch. They are the ones with the most instrumented neighbours downstream. A machine that is the only one of its kind, sitting at the end of the line, is nearly invisible. The same machine in the middle of the line reports itself through everything behind it.