车间里调工艺，常见的做法是"一次只动一个旋钮"：温度试了试压力，压力试了试时间。看起来稳，其实这种单因子试验把最关键的东西漏掉了——因子之间的交互作用。DOE（试验设计）听起来高深，其实搞懂三个概念就够入门：因子、主效应、交互作用。
▌ 因子：影响结果的输入变量
因子就是你能控制、并且怀疑会影响结果的工艺参数。以注塑为例：料温、注射压力、保压时间、模具温度，都是候选因子。
- 选因子不是拍脑袋，优先从 PFMEA 的失效原因和鱼骨图里挑；
- 每个因子定两个水平（高低各一），范围要覆盖正常工艺波动之外、又不至于打出废品；
- 一次试验因子不宜贪多，筛选阶段 5~7 个以内为宜。
▌ 主效应：改一个参数，结果变多少
主效应回答的问题很直接：把某个因子从低水平调到高水平，结果平均变化多少。比如料温从 180℃ 提到 220℃，产品强度平均提升了一截，这个差值就是料温的主效应。主效应大的因子，是调参时最该先管的因子。
▌ 交互作用：1+1≠2
这才是 DOE 的灵魂。交互作用是指一个因子的效果，取决于另一个因子处在什么水平：料温升高对强度的提升，在高压下明显、在低压下几乎为零——温度和压力之间存在交互作用。
单因子试验为什么注定漏掉它？因为它每次只动一个因子、把其余全部固定，看到的永远是"其他条件不变"时的主效应。而真实工艺里，因子很少各管各的——交互作用普遍存在，而且往往决定工艺窗口的边界。
▌ 实操路线：先筛再挖，两轮锁定工艺窗口
- 第一轮筛选：用筛选试验（部分因子设计）从多个候选因子中找出主效应显著的 2~3 个关键因子，其余固定在经济水平；
- 第二轮深挖：对关键因子做全因子试验，把两两交互作用全部暴露出来，画出完整的工艺窗口；
- 确认的窗口写进控制计划和作业指导书，参数调整才算有了依据。
记一句：单因子试验看的是"局部坡度"，DOE 看的是"整张地形图"——交互作用不在图纸上，就在你的废品里。
想让团队把 DOE 从概念落到工艺改进？续创的在线实操课边学边练，用真实工艺数据做试验设计。欢迎联系续创信息技术咨询。
The common way to tune a process on the shop floor is to turn one knob at a time: try temperature, then pressure, then holding time. It feels safe, but this one-factor-at-a-time approach misses exactly what matters most — interactions between factors. Design of Experiments (DOE) sounds intimidating, yet three concepts are enough to get you started: factors, main effects and interactions.
▌ Factors: The Input Variables That Drive the Result
A factor is a process parameter you can control and suspect of influencing the outcome. In injection moulding, melt temperature, injection pressure, holding time and mould temperature are all candidate factors.
- Don't pick factors on a hunch — start from the failure causes in your PFMEA or a fishbone diagram;
- Set two levels (low and high) for each factor, spanning beyond normal process variation but short of producing scrap;
- Resist the urge to test too many at once — keep it to 5–7 candidates in the screening phase.
▌ Main Effects: How Much the Result Moves When One Factor Changes
A main effect answers a simple question: when a factor moves from its low to its high level, how much does the response change on average? If raising melt temperature from 180°C to 220°C lifts part strength noticeably, that difference is the temperature's main effect. Factors with large main effects are the ones to manage first when tuning a process.
▌ Interactions: When 1 + 1 ≠ 2
This is the soul of DOE. An interaction means the effect of one factor depends on the level of another: the strength gained by raising melt temperature is significant at high pressure but almost zero at low pressure — temperature and pressure interact.
Why do one-factor experiments inevitably miss this? Because they move one factor while holding all the others fixed, so they only ever see the main effect "with everything else constant". In real processes, factors rarely act independently — interactions are everywhere, and they often define the boundaries of the process window.
▌ The Practical Route: Screen First, Then Dig Deep — Lock the Window in Two Rounds
- Round one, screening: use a screening design (fractional factorial) to identify the 2–3 factors with significant main effects out of many candidates, and fix the rest at economical levels;
- Round two, digging deep: run a full factorial on the key factors to expose every two-way interaction and map the complete process window;
- Write the confirmed window into the control plan and work instructions — parameter adjustments are now evidence-based.
One line to remember: a one-factor experiment shows the local slope, while DOE shows the whole terrain — if interactions are not on your map, they are in your scrap bin.
Want your team to move DOE from concept to process improvement? CINO's hands-on online courses let trainees practise experiment design on real process data. Contact CINO Information Technology Consulting.