How to Reduce Manufacturing Defects with Design of Experiments (DOE)

Every manufacturer knows the pain of a defect rate that will not budge. You adjust the temperature — no improvement. You increase the pressure — a little better, then worse again. Technicians tweak one setting at a time, chasing a moving target, while scrap piles up and margins erode. This intuitive One-Factor-at-a-Time (OFAT) approach feels like common sense. It is also the single most common reason process problems never get solved.

Design of Experiments (DOE) is the statistical method that breaks the cycle. In this article we will explain why OFAT fails, how DOE reduces manufacturing defects, and walk through a real case study where a designed experiment slashed defects by 68%. If your team is serious about process improvement, this is the tool that turns opinions into evidence.

Why One-Factor-at-a-Time (OFAT) testing fails

OFAT means changing a single input variable while holding all others constant. You test Factor A, lock in its "best" setting, then move on to Factor B — assuming the ideal setting for A is independent of B. In the real world of chemistry, polymers and machinery, that assumption is usually wrong.

The fatal flaw of OFAT is its inability to detect interactions. In complex systems, the effect of one factor often depends on the level of another. By keeping other variables fixed, OFAT sails straight past the "sweet spot" where variables combine to produce a superior result. You can run dozens of trials, feel busy and productive, and still miss the answer entirely — because it lives in a combination you never tested.

What is Design of Experiments (DOE)?

Design of Experiments is a branch of applied statistics for planning experiments so that valid and objective conclusions are drawn effectively and economically. Instead of moving one lever at a time, DOE changes multiple factors simultaneously in a structured, orthogonal pattern. That lets you map the entire experimental space — main effects and interactions — with far fewer runs than brute-force testing.

Three terms anchor every DOE:

• Factors — the controllable inputs you manipulate (e.g. temperature, pressure, speed).
• Levels — the specific settings assigned to each factor (e.g. 220°C vs 260°C).
• Responses — the measurable outputs that reveal the effect of the changes (e.g. shrinkage, yield, strength).

Two principles keep the conclusions honest: replication (repeating runs to estimate experimental error and increase precision) and randomisation (running trials in random order so unknown "lurking" variables cannot bias the result).

Case study: cutting injection-moulding defects by 68%

Here is the DOE method applied to a real problem. A precision electronics manufacturer was reporting a 14% rejection rate on plastic housing components due to excessive part shrinkage. Manual, one-factor-at-a-time adjustments to the moulding machines never stabilised the process, producing inconsistent dimensions and wasted material.

Step 1 — Define the objective

The goal was specific and measurable: reduce shrinkage variation to within ±0.05 mm of the design specification while keeping cycle time under 30 seconds. A clear objective is non-negotiable — everyone must agree what "success" looks like before any data is collected.

Step 2 — Select factors and levels

Three controllable factors were chosen, each at a low (−1) and high (+1) level:

• A: Melt Temperature — 220°C to 260°C
• B: Injection Pressure — 80 to 120 Bar
• C: Cooling Time — 15 to 25 seconds

Step 3 & 4 — Design and run a 2³ full factorial

Three factors at two levels gives a 2³ = 8-run full factorial design — every possible combination, each replicated five times for statistical significance. Eight structured runs replaced the endless, unstructured tweaking that had failed for years.

Step 5 — Analyse the data

A Pareto analysis of effects revealed that injection pressure and cooling time were the dominant drivers of shrinkage. More importantly, the analysis exposed a strong temperature × pressure interaction: at high melt temperatures, injection pressure mattered far more. This was the exact effect OFAT could never see — and precisely why the old trial-and-error approach kept failing. Technicians were adjusting temperature without compensating with the pressure shift the data demanded.

Step 6 & 7 — Optimise, verify and implement

Using the factorial data, a 3D Response Surface Methodology (RSM) model located the optimal window: melt temperature 235°C, injection pressure 115 Bar, cooling time 22 seconds. Confirmation runs validated the prediction, and the settings were written into the Standard Operating Procedures.

The rejection rate dropped from 14% to 4.5%, with estimated annual savings from reduced waste and re-work exceeding six figures per production line. One designed experiment paid for itself many times over.

Choosing the right type of DOE

DOE is not a single technique but a toolbox. Matching the design to your situation is what keeps it efficient:

• Full Factorial — studies all factor combinations and interactions in depth. Ideal for 2–4 well-understood factors.
• Fractional Factorial — a cost-effective subset of runs for screening many factors down to the "vital few".
• Plackett-Burman — highly efficient screening for 10+ candidate factors with minimal runs.
• Randomised Block Design (RBD) — controls known nuisance variation (e.g. batch or shift differences).
• Taguchi Methods — robust design that minimises sensitivity to noise using orthogonal arrays.
• Response Surface Methodology (RSM) — maps process curvature to pinpoint the optimal settings, as in the case study above.

A common, powerful sequence is to screen with a fractional or Plackett-Burman design, characterise with a full factorial, then optimise with RSM.

The business case: why DOE beats intuition

In complex processes, statistical significance beats expert intuition — consistently. DOE lets you design for the whole process window rather than a single fragile setpoint, so your process stays robust even when raw-material viscosity or ambient conditions drift. Fewer defects, less scrap, lower re-work, faster root-cause analysis and defensible, data-backed decisions: that is the return on learning to design experiments properly.

Get your team trained — HRD Corp claimable

Reading about DOE builds awareness; doing it builds capability. Megabyte Consultancy's Design of Experiments (DOE) training takes your engineers from OFAT pitfalls all the way through factorial, Taguchi and RSM designs — working through this exact injection-moulding case study hands-on. It is available as in-house onsite training at your premises, live online sessions, or self-paced e-learning, and is HRD Corp claimable for eligible Malaysian employers.

Frequently asked questions

Do participants need a statistics background?

No. A good DOE program starts from first principles — cause and effect, factors and levels — before introducing factorial and response-surface designs, so engineers without a statistics background can follow along and apply it.

What software is used for DOE?

DOE can be run in Minitab, JMP, Design-Expert or even structured Excel templates. The thinking matters more than the tool; training focuses on designing and interpreting experiments correctly regardless of software.

Is Design of Experiments training HRD Corp claimable in Malaysia?

Yes. Employers registered with HRD Corp can claim eligible DOE training under the HRD Corp Claimable Courses scheme. Submit the grant application on eTRiS at least one day before training begins; our TTT-certified trainer profile, content and schedule are ready for your application.