Skip to main content
Captia Technology
Captia ConsultingCase study

Article

Case study: eight machines, one bottleneck and a 1,900 euro shelf

A real, anonymised plant case: nine dry contacts on eight machines, two weeks of baseline, the availability percentage that misled everyone, the effective-cycle-versus-takt analysis that found the real constraint, a 1,900 euro countermeasure, the control chart that proved the gain and a payback calculator with the case numbers.

Published
September 6, 2026
Updated
September 6, 2026
Format
Case study
Reading
14 min

A plant with eight machines, a product that could not keep up with demand and a management team convinced it had to buy a machine. Nine dry contacts, three weeks of data and a €1,900 shelf were enough to go from 566 to 635 frames a week. This is the full journey: what was measured, what was misleading, what was decided and what did not work first time.

How many hours did this machine work yesterday?

It is the question every serious operational diagnostic starts with, and in most plants it has three answers. None of them is data.

  • The paper log. Filled in by the operator at the end of the shift, from memory. Rounded to the half hour.
  • The supervisor. “Fine, as usual.” Knows it stopped, not how many times or for how long.
  • The ERP. The order was closed three days later. It says how much was produced, not when production stopped.

The line below is data: a dry contact in the tube bender’s cabinet and an NTP-synchronised clock. Every blue stretch is time in cycle; every gap, a stop.

06:0007:0008:0009:0010:0011:0012:0013:0014:00
A dry contact and a clock. Filled: the machine is cycling. Gap: stopped. Hatched: planned break. The position of each stop is illustrative; the total respects the availability in the case.

What was installed: eight machines, nine dry contacts

Not a single PLC modified. Each machine reports its running or stopped state through a volt-free contact; an input module publishes it over MQTT with TLS to a Captia Connect server on site, which buffers locally if the network drops and forwards to Captia.ai. The welding cell has two cradles and gives two signals; hence nine contacts for eight machines.

MachineWhere the signal comes from
M01 Manual cuttingAuxiliary contacts on the motor protection switch
M02 CNC milling machineOperating-state signal from the cabinet
M03 Deep-drawing pressRelay on the green running beacon
M04 Robotic welding cellCycle-lamp digital output, one per cradle
M05 Tube benderRelay on the active-cycle lamp in the cabinet
M06 Pallet wrapperRunning signal from the cabinet
M07 Laser cutterCommissioned in week 5, outside the measurement window
M08 Automatic cuttingVolt-free relay at 230 V

Investment, excluding VAT

€31,125.28

Hardware, electrical installation, on-site server, MES integration and commissioning. A case figure, altered like the rest.

Per connected machine

€1,508.41

Plant fixed cost of €19,058 plus this amount for each additional machine.

Continuity

3 outages · 0 losses

Three Wi-Fi drops in bay 2 of 4, 11, 37 minutes during measurement. The on-site server resent all 3,482 events with their original timestamps.

Two machines at 77%: one stops 23 times, the other 214

Baseline: two weeks, two shifts, 150 planned hours per machine. The availability percentage is misleading. Automatic cutting and the tube bender both sit at 77%, and they have nothing in common: one stops 23 times in two weeks, the other 214. Counting every stop changes the diagnosis.

Tube bender76.8 % · 214 stopsAutomatic cutting76.9 % · 23 stopsManual cutting68.2 % · 96 stopsCNC milling machine81.5 % · 41 stopsDeep-drawing press74.0 % · 63 stopsRobotic welding cell77.9 % · 52 stopsPallet wrapper58.6 % · 37 stops06:0010:0014:00
Two machines at 77% with opposite stories: automatic cutting stops 23 times; the tube bender, 214. Stop criterion: state 0 for 30 seconds or more within planned time, with a 2-second debounce.

41% of the plant’s stops, ten minutes each

A 10-minute stop never appears on a paper log. Added up, the bender’s 214 stops are 35 hours in two weeks: one stop every 32 minutes of running time, with a mean duration of 9.8 minutes.

Tube bender214 · 41 %Manual cutting96 · 18 %Deep-drawing press63 · 12 %Robotic welding cell52 · 10 %CNC milling machine41 · 8 %Pallet wrapper37 · 7 %Automatic cutting23 · 4 %
Stops per machine over two weeks of baseline, highest first. By stopped minutes, the wrapper would come first with 22% of stopped time.

The Pareto trap. Sorted by stopped minutes rather than by number of stops, the top of the list would be the pallet wrapper, sitting at 58.6% with only 37 long stops. With that Pareto in hand a second wrapper at €48,000 would have been approved. The wrapper is not broken: it is waiting for product. And product does not arrive because upstream there is a machine stopping every half hour.

Everyone watched the robot. The robot was waiting for tube

Demand is 620 frames a week in 4,500 planned minutes: a takt of 7.26 minutes per frame. To find which machine limits the line, nominal cycle time is useless; what counts is effective cycle time, nominal divided by availability. Only one machine sits above the takt, and it is not the welding cell, the usual suspect because it is the most expensive.

takt 7.26 minAutomatic cutting5.20Tube bender7.94Deep-drawing press6.22CNC milling machine6.75Robotic welding cell7.06Pallet wrapper5.46

Line capacity

566 of 620 a week

Shortfall of 54 frames a week. The backlog grows and management hears "we need to buy a machine".

How to read it

Effective cycle = nominal cycle ÷ availability. The machine whose bar crosses the takt line is the one limiting the line. Only one does, and it is not the robot.

Demand of 620 frames a week in 4,500 planned minutes.

A second data point confirms it. Cross-correlating the bender and robot signals gives a lag of 18 minutes: when the bender stops, 18 minutes later the robot stops. That lag is the size of the intermediate buffer, 3 frames. The robot was at 77.9% not because of its own faults, but because it ran out of bent tube.

On the floor: ten waits per shift for the forklift

With the timeline in hand, week three was spent at the foot of the bender. Tube arrives in packs of six. When the pack runs out, the operator calls. There is one forklift driver for both bays and he is usually loading trucks at the wrapper. Mean wait: ten minutes. Exactly the mean stop duration Captia.ai had been counting for two weeks.

BAY 1BAY 2STOREtube in packs of 6M05benderM06wrapper and dockM04welding
Urgent trips to the tube store, always in a hurry and always interrupting truck loading.

Before · urgent trips per shift

10.7

62 metres between the tube store and the bender. 1.33 km per shift, always in a hurry.

After · fixed rounds per shift

4

Replenishment every two hours. 0.50 km per shift, planned.

Cost of the countermeasure

€1,900

A shelf for four packs and a card that triggers replenishment. Captia.ai does not draw this. Captia.ai says which machine to go to.

The bender stops stopping. The robot rises untouched

Three weeks later, 225 planned hours per machine, the same signal compared machine by machine against the baseline.

MachineActionBeforeAfterΔ
Tube benderTube supermarket next to the machine76.8 % · 21486.1 % · 71+9.3
Robotic welding cellNone. It stops waiting for tube77.9 % · 5282.3 % · 63+4.4
Manual cuttingOrder and material ready before the shift change68.2 % · 9673.0 % · 121+4.8
Pallet wrapperNone. The second wrapper is dropped58.6 % · 3759.1 % · 55+0.5
CNC milling machineNone81.5 % · 4181.7 % · 60+0.2
Deep-drawing pressNone74.0 % · 6375.0 % · 93+1.0
Automatic cuttingNone76.9 % · 2377.6 % · 34+0.7
Plant73.4 %76.4 %+3.0
UNL 86.4baseline mean 76.8LNL 67.2countermeasureday 1day 2596 %60 %
Natural limits computed on the baseline (mean ± 2.66 × mean moving range). The fifteen later days sit above the centre line; it is not noise and it does not fade: 85.4%, 86.3%, 86.6% in weeks 4, 5 and 6. The daily series is illustrative; the means are those of the case.

Plant average rose 3.0 points. A normal dashboard shows that, and anyone who looks only at that misses the point: what changed is line capacity, from 566 to 635 frames a week. Manual cutting also improved, with order and material prepared before the shift change, and the three remaining machines were left alone: their stop counts rose because they produced more, not because they failed more.

Four decisions and the data point behind each

None required technology. All required knowing which machine to look at.

DecisionCostEffectData point behind it
Tube supermarket at the bender€1,900Bender from 76.8% to 86.1%. The line goes from 566 to 635 frames a week.214 stops in two weeks, mean time between stops 32 min.
Leave the robot alone€0Welding from 77.9% to 82.3% without intervention.Its stops follow the bender’s with a 18-minute lag.
Do not buy the second wrapper€48,000 deferredNot a saving: deferred investment. If it is never needed, it is €48,000.Wrapper at 58.6% with 37 long stops: waiting for product, not failing.
Next: SMED on the bender, not on cuttingTo be decidedShortening changeovers on automatic cutting would give availability points and zero frames, because it is not the constraint.Effective cycle vs takt, after: bender 7.08, cutting 5.15.

The installation pays for itself in weeks, not years

Payback is valued at the constraint: every frame the bender did not bend was a frame that was not sold. Recovered: 54 a week, capped by demand. At €28 contribution margin per frame and 45 weeks, that is €68,040 a year, and the installation pays for itself in 24 weeks. With your numbers it may come out differently; that is what the calculator is for.

Payback calculator

Preloaded with the case numbers: 54 frames a week at €28 margin, 8 machines. Move the controls with your own numbers.

54
€28
8
Investment, with 8 machines
€31,125.28
Frames or hours recovered, valued per year
€68,040
The installation pays for itself in
24 weeks
Three-year IRR
211 %
Each month without deciding costs
€5,670
Including the €499 monthly fee, total payback in
26 weeks

Investment = plant fixed cost of €19,058 plus €1,508.41 per machine. Payback = investment ÷ annual value × 52. Arithmetic in the browser; nothing is sent.

What the Captia.ai monthly fee has to beat

Captia.ai is paid per plant, with no per-machine limit: €499 a month, €5,988 a year. That is the bar it has to clear every month to earn its fee.

Bar in frames

4.8 a week

At €28 margin per frame and 45 weeks.

Bar in time

24 min

Per machine per week, with 8 machines at €42 per stopped hour.

What the bender alone recovered

21 h

In the three following weeks: 52 minutes per machine per week, 2 times the bar.

Installation and Captia.ai together, with the case numbers: payback in 26 weeks, and from then on €62,052 net per year after paying the fee. If Captia.ai does not deliver 4.8 frames a week, it does not earn its fee.

Three things that went wrong and one we did not measure

A plant project with no incidents is a project that has not been executed.

  1. Wi-Fi in bay 2. Coverage did not reach two machines. One additional access point, change proposal of €640. Meanwhile, the on-site server’s buffer prevented any lost events.
  2. 230 V on the press. The running beacon is driven by a 230 V AC relay and the input module works at 24 V DC. Interface relay and an afternoon’s work.
  3. The MES API. The MES delivered work orders, but without quantities per shift until the second iteration. That is why payback is calculated with availability and takt, not with OEE.
  4. What we did not measure. Performance and quality. Without quantities and rejects per shift there is no OEE, and Captia.ai does not call it OEE until it is.

Frequently asked questions

Why dry contacts and not the PLC signal?

Because it requires no changes to any machine’s program and no dependency on its manufacturer. A cycle-lamp relay, a beacon or an auxiliary contact on the motor protection switch already says whether the machine is running, and that is enough to count every stop. It is the fastest and least invasive way to get the data; quantities and rejects come later, from the MES or from additional sensors.

Why was the availability percentage not enough to find the problem?

Because two machines at the same 77% can have 23 long stops or 214 ten-minute stops, and the cause and the fix are different. The percentage summarises; the stop count and mean duration diagnose. That is why Captia.ai counts every event with its timestamp instead of delivering a single indicator.

What is effective cycle time and why compare it with takt?

Effective cycle time is nominal cycle time divided by availability: how long a machine really takes to produce a part, stops included. Takt is available time divided by demand. The machine whose effective cycle exceeds the takt is the line’s constraint, and any improvement elsewhere yields availability points but zero units sold.

How do you know the improvement is not noise?

With a control chart: natural limits are computed on the baseline and the later days are checked for a sustained shift. Here the fifteen later days sit above the centre line and weeks 4, 5 and 6 hold at 85.4%, 86.3%, 86.6%. A run like that is not produced by chance.

Are these the client’s real numbers?

No. The project is real and the decisions are the ones that were made, but the name, the location and the figures have been altered for confidentiality while keeping the proportions. The daily series in the charts are illustrative and respect the case means.


We do not sell dashboards. We sell knowing which machine to go to. If in your plant the answer to “how many hours did this machine work yesterday” is still a paper log, an operational diagnostic with each machine’s running signal is the first step. Write to us and we will look at your case.

Author

Written by the Captia Consulting team

Last updated: September 6, 2026