Year three, and the part that stops it

The hours the arithmetic quietly assumed

Every figure in lesson 3 rested on an assumption that was never written down: that the cell would be there when the work arrived.

work the cell must do   78,000 x 1.16 = 90,480 machine-minutes
style changes           31 x 300      =  9,300 machine-minutes
total                                   99,780 = 1,663.00 hours

Against a machine year of 1,880 hours that is comfortable. Now put availability into it. Availability is the share of the year the machine is actually able to run.

Availability enters the payback through hours, and hours are the one place where a shortfall turns straight into money. At 86.4% the cell is short 38.68 hours, which is 2,001 garments prepared by hand at no saving, or USD 1,092.35 a year off a result that was already negative. Nothing else in the model moved. The machine did not get slower. It was available less.

That is why availability is worth more attention than cycle time in every year after the first. It is also why a proposal that quotes a cycle time and no availability is quoting the easier half.

Where the downtime went, and it is not where anybody looks

Demirkol keeps a log. Year two's downtime was 255.68 hours, and he splits every entry into the time a repair actually took, and the time spent waiting for the means to do it — waiting for a part, waiting for a person, waiting for somebody to decide.

repair                              73.00 hours
waiting                            182.68 hours
                                   255.68 hours

71.4% of the cell's downtime was waiting. Not diagnosing, not repairing, not a design weakness in the machine: waiting. That split is why maintenance regimes are argued about at the wrong level. Course 20.6 prices three maintenance regimes properly and is where to go for that argument. This lesson is about the half that no regime touches, because a preventive schedule does not shorten a nine-week lead time.

The kit that was chosen correctly and was wrong

The supplier's recommended spares list came to USD 3,940, which purchasing thought excessive. They cut it, and they cut it on a principle you can defend: hold the parts that fail most often. Demirkol's own failure log was the evidence, and the ranking was correct.

Read the Failures a year column, and then the Days lost a year if not held column. By failure rate the sensor is fifth of six and was correctly dropped. By days lost it is second, because its lead time is nine weeks from one source. The quantity that matters is not how often a part fails, and it is not what the part costs. It is failures a year multiplied by the time it takes to get another one, and that product is in neither of the columns anybody reads.

A day of the cell being down is 480 minutes at 1.16 minutes a garment, 413.79 garments at USD 0.5459, or USD 225.89.

In March of year two the sensor failed. Airfreight brought the replacement in 11 working days instead of forty-five, at USD 310.

11 days x USD 225.89 = USD 2,484.79 of benefit lost
against a part costing USD 41

60.6 times the price of the part, and the whole of it could have been avoided for USD 82 — two sensors on a shelf.

Rebuild the kit on days lost rather than failure rate, and it holds two drive belts, one gripper cup set, one valve and fuse set, one vacuum generator and two sensors: USD 890, against 55.5 days of expected annual loss at USD 225.89, which is USD 12,536.90 a year. The vision camera stays off the list at USD 1,340 for three expected days, and the right answer there is a written promise from the supplier to lend one, not a purchase.

The four things that decide whether it is still running in year seven

Lesson 3 charged the cell over seven years. Whether it survives seven years turns on four questions, and none of them is about the machine.

Who owns the programmes and the drawings. Each of Akbulut's pocket shapes exists as a taught programme and a machined former. If the programme files live only on the cell's controller, and the former drawings live only at the supplier, then a controller failure or a change of ownership at the supplier costs the factory every shape it has built up. Lesson 3 showed that a new shape costs USD 469.46 to set up again. Nineteen of them is USD 8,919.74 of work sitting on one hard disk. Ask for the files, ask for the drawings, and keep a copy off the machine.

Whether the controller's software is licensed or owned. A yearly licence is a running cost that was not in lesson 3's USD 6,400, and it is also a lever the supplier holds for the whole life of the asset. Find out before you sign, not in year four.

Whether a second person can run it. Lesson 5 priced one technician leaving at USD 6,164.53. A second trained person is not a luxury. It is the cheapest availability insurance on the site, and it costs the training time and the grade.

Whether the local agent is still an agent. The nine-week lead time on the sensor was nine weeks because there was one source. That is a commercial fact about the supplier's distribution, not a technical one about the part, and it can change in either direction. Ask, in writing, who supplies parts in your country and what happens if they stop.

What to put in the order, before it is placed

Aydemir's honest summary to Bayraktar was two sentences. The cell is the right machine for the wrong plant, and lesson 3 said so with a number before anybody signed. If it is bought anyway, the five clauses above are worth more than the discount that will be offered instead of them.

Act as an engineering manager placing an order for an automated cell in a garment factory, who knows that the machine's speed is the least important thing in the contract. Draft the terms. Facts: machine or cell [DESCRIBE], supplier [NAME OR LEAVE BLANK], price [AMOUNT], the fabrics I actually run with weights [LIST], the number of different shapes it must handle in a year [NUMBER], my working hours a year [NUMBER], the hours of work I need it to do [NUMBER], and my cost of one day of the cell being down [AMOUNT, or ask me how to work it out]. Produce five things. First, an acceptance test written on MY materials: how many panels and blanks in each fabric, what is measured, how it is reported, and what counts as a pass — including first-attempt pick rate and the mean and spread of placement error, reported separately by fabric. Second, a ramp clause with a number in it: what the supplier must demonstrate on new shapes, how the pieces to steady state are counted, and what happens if the figure is missed. Third, an availability definition that says exactly what counts as down time, including waiting for a part, waiting for a person and waiting for a decision. Then show me what the same machine's availability looks like under a definition that counts only repair time, so I can see the size of the loophole. Fourth, a spares list ranked by expected days lost a year — failures a year multiplied by working days to obtain the part — rather than by price or by failure rate, with a recommended holding and a total. Fifth, the deliverables I must receive with the machine: programme files, fixture and former drawings, calibration procedure, training with a named competence test, and a written answer on who supplies parts in my country. Ask me for anything you need rather than assuming it, and flag any clause a supplier is likely to refuse.