Changing one moulding setting at a time is slow and misses interactions. How designed experiments find the settings that matter, quantify their effects and build a robust process window in a handful of runs.
A moulder had been chasing a warpage problem on a large polypropylene panel for weeks. Each change, more cooling, higher hold pressure, a different mould temperature, helped on some days and not others. A designed experiment with four factors and 16 runs, completed in a single day, showed why: the effect of hold pressure depended on mould temperature. At one mould temperature, more hold pressure reduced warpage; at another, it made it worse. One-factor-at-a-time testing could never have found that.
Design of experiments, DOE, is the most efficient way to understand a process with several interacting variables, and injection moulding is full of them.
Why one factor at a time fails
Changing one setting at a time, holding everything else constant, seems logical. It has three problems:
- It needs many runs to cover several factors.
- It misses interactions, where the effect of one factor depends on another.
- It often lands on a local optimum that is not robust.
DOE changes several factors together in a planned pattern, then uses statistics to separate their effects.
A typical moulding DOE
- Define the response: a measurable output such as a critical dimension, warpage, weight or a cosmetic score. Make sure it can be measured reliably, as in Gauge R&R.
- Choose factors: the settings most likely to matter, such as melt temperature, mould temperature, hold pressure, hold time, injection speed and cooling time.
- Set levels: a low and high setting for each, wide enough to show an effect but inside a safe window.
- Choose a design: a full factorial for three or four factors, or a fractional factorial to screen more factors with fewer runs.
- Run it: randomise the order, allow the process to stabilise after each change, and collect several parts per run.
- Analyse: main effect and interaction plots, and a statistical model showing which factors and interactions matter.
- Confirm: run the predicted best settings and verify the result.
What a DOE gives you
- The few factors that drive the response, and the many that do not.
- Interactions you would never find by trial and error.
- A mathematical model of the process, which you can use to find settings that hit the target with minimum variation.
- A documented basis for the process window, valuable for validation and PPAP.
Practical tips for moulders
- Start with scientific moulding. Use viscosity and gate freeze studies to set the basics before running a DOE.
- Screen first, then optimise. A fractional factorial can screen six or seven factors in 16 runs; a smaller follow-up DOE then optimises the important ones.
- Allow time for the process to stabilise, especially after mould temperature changes.
- Measure parts after they have conditioned, if the response changes over time.
- Use machine-independent factors, such as melt temperature measured in the melt, rather than barrel set points where possible.
Beyond troubleshooting
DOE is equally useful for new tools, for setting robust process windows during validation, for material changes and for comparing machines. Many experienced moulders find that a single well-run DOE replaces months of intuition, and makes that knowledge available to everyone on every shift, as our warped panels case study showed.
