MANUFACTURING · QUALITY · FLOW
20 min readThe fastest machine can make the factory slower.
A production line is not a collection of independent machines. It is a network of queues, constraints, variability, information and quality decisions.
Put a new machine on a line and double its speed. It sounds like an obvious improvement. But if the downstream operation cannot consume the extra output, the result is not automatically more production. It may be more work-in-process, more waiting, more handling and more defects hidden in inventory.
This is the first important idea in manufacturing engineering: local optimisation is not the same as system optimisation.
1. Demand sets the pace
Takt time translates customer demand into a required production rhythm.
If a shift provides 27,000 seconds of net production time and demand is 450 good units, takt is 60 seconds per unit. That number is not a machine parameter. It is a system requirement created by demand.
| Metric | What it means |
|---|---|
| Takt time | required pace to satisfy demand |
| Cycle time | actual processing time for a cycle |
| Lead time | elapsed time from request to completion |
| Throughput | good output completed per unit time |
2. Bottlenecks decide what the factory can deliver
The constraint limits system output. Increasing capacity at a non-constraint can make dashboards look better while finished-product throughput remains unchanged. In the worst case, upstream processes flood the bottleneck with WIP.
Once the constraint is improved, the bottleneck may move. Improvement is therefore iterative.
3. WIP is time stored on the factory floor
For a stable system, increasing WIP without increasing throughput increases flow time. That is why large buffers can make a line feel “safe” while hiding instability, quality problems and poor coordination.
4. Variability turns averages into traps
A station with an average cycle time of 55 seconds is not automatically safe against a 60-second takt. If the distribution has a long tail because of micro-stops, product mix or manual variation, the line can still starve and block frequently.
Capacity studies should therefore look at distributions, percentiles and state-dependent losses—not only means.
5. OEE decomposes equipment loss
Availability captures downtime, performance captures speed loss and quality captures defective output. Three small losses multiply. That makes OEE useful as a loss model, but dangerous as a target if teams manipulate definitions simply to raise the number.
6. Quality cannot be inspected into the product
End-of-line inspection can detect nonconforming output. It does not make the process capable. A capable process must be stable enough and centred enough to produce within specification repeatedly.
These indices only make sense when the process is sufficiently stable and the measurement system is trustworthy.
7. Control limits and specification limits are not the same
Specification limits come from product or customer requirements. Control limits describe the observed behaviour of a process. A process can be statistically stable but centred outside specification. It can also meet specification today while being statistically unstable and drifting toward failure.
8. The measurement system can manufacture false problems
If a gauge has poor repeatability or reproducibility, a good part may appear bad and a bad part may appear good. Operators may then adjust a stable process based on measurement noise.
That is why calibration, uncertainty, resolution and Gauge R&R belong inside production engineering. Measurement is part of the process that makes the decision.
9. FMEA moves failure analysis earlier
FMEA asks what can fail, what happens if it fails, why it could fail and what controls exist to prevent or detect it. The useful output is not the spreadsheet score; it is the engineering conversation that changes the process before the failure reaches production.
| Question | Manufacturing example |
|---|---|
| Failure mode | component inserted in the wrong orientation |
| Effect | assembly cannot complete or fails later |
| Cause | fixture allows ambiguous orientation |
| Prevention | poka-yoke fixture geometry |
| Detection | vision check / sensor confirmation |
10. Digital manufacturing should close a loop
MES, SCADA, IIoT and dashboards are useful only if data improves a decision. A connected factory that collects every signal but has no response mechanism is a data archive, not a smart production system.
11. Sustainability is another system output
Scrap, rework, energy consumption and material loss are not side topics. They are process outputs. Increasing line speed while increasing scrap or energy per good unit may worsen overall performance even if throughput rises.
A modern manufacturing KPI set should therefore connect quality and productivity with energy intensity, yield and lifecycle impact.
12. The factory is a system, not a machine catalogue
The best manufacturing improvements usually come from understanding interactions: demand with capacity, WIP with flow time, equipment loss with quality, measurement with decisions, and energy with production scheduling. The engineering challenge is to optimise the system rather than celebrate one locally impressive number.