Single Factor Rotation Always Fails? —— Five-Step Practical Approach to Full Factorial DOE

Many quality professionals get stuck on model selection when they start learning DOE: they know how to handle a few factors, but when the number of factors is small, they are unsure. Full Factorial Design (Full Factorial) is a method that tests all combinations of factor levels—4 trials for 2 factors at 2 levels, 8 for 3 factors, and 16 for 4 factors. Although the number of trials seems "luxurious," it allows you to see both main effects and interactions simultaneously, which single factor rotation and partial factorial designs cannot achieve. Today, we will use a case study of plastic part warping to explain when to use full factorial design, how to arrange it, and how to analyze it.

1. When to Use Full Factorial Design

It is worth using full factorial design when all three conditions are met:

1. The number of factors is small, generally no more than 4 to 5. For 4 factors at 2 levels, 16 trials can usually be scheduled by most production lines. However, for 6 or more factors, 64 trials should be carefully considered, and partial factorial screening should be considered first.
2. Interactions may play a key role. Adjusting Factor A alone has no effect, but adjusting A and B together doubles the effect—this "1+1 greater than 2" phenomenon can never be detected by single factor rotation, only by full factorial trials.
3. The trial cost is manageable, and the conclusion must be definitive in one go. Compared to partial factorial designs, full factorial designs do not make "alias" compromises; each main effect and second-order interaction can be independently estimated. This is suitable for the "final round of confirmation" before fine-tuning optimization after a few key factors have been screened out.

Conversely, if the goal is to quickly screen out 2 to 3 factors from 8, directly using full factorial design is wasteful—256 trials are too many and unnecessary. Full factorial and partial factorial designs are not substitutes but sequential steps: screen first, then full factorial, and finally response surface optimization.

2. Case Study: 16 Full Factorial Trials for Plastic Part Warping

A supplier of automotive interior components was producing dashboard skeletons, but the warping deformation after injection molding exceeded the standard, causing misalignment of the snap-fit clips during assembly and batch returns from the client. The project team brainstormed and listed 6 candidate factors, which were scored using a cause-and-effect matrix, and 4 factors were selected for full factorial trials, each with two levels:

A Mold Temperature: 60°C / 90°C; B Material Temperature: 230°C / 250°C; C Holding Pressure: 60MPa / 80MPa; D Holding Time: 4 seconds / 8 seconds.

The response variable was the warping amount, and for each combination, 3 products were measured using a coordinate measuring machine (CMM) to take the average value. Before the trials, a measurement system analysis was conducted to ensure that the CMM repeatability met the requirements—otherwise, 16 trials would be in vain.

With 4 factors and 2 levels, there are 16 combinations, which is the essence of "full factorial": no level combination is omitted.

3. Trial Implementation: Randomization, Replication, and Data Recording

Once the plan is set, there are three ironclad rules for implementation:

1. Randomization: The order of the 16 combinations should be randomized before production to avoid systematic bias due to factors like mold temperature machine warming up or operator fatigue over time.
2. Replication: Each combination should be replicated 3 times to estimate the trial error from piece-to-piece variation—without replication, there is no way to discuss "significance."
3. Recording: Record the environmental temperature and humidity, equipment status, and raw material batch for each trial. If anomalies are found during analysis, it will be possible to trace whether they are due to trial issues or recording issues.

The 16 combinations were completed over two shifts, with shift differences included in the analysis as blocks to separate the "shift change" effect from the error.

4. Three-Step Analysis: Main Effects, Interactions, and ANOVA

After data collection, follow these three steps for analysis:

1. Review the main effects plot: Among the four factors, mold temperature has the greatest impact on warping, followed by material temperature. Holding pressure and holding time are not significant when considered individually.
2. Review the interaction plots—this is the exclusive benefit of full factorial design: plot the interaction between mold temperature and holding time, and the two lines clearly cross. When the mold temperature is 60°C, increasing the holding time from 4 seconds to 8 seconds has almost no effect on warping. However, when the mold temperature is 90°C, the warping at 8 seconds is significantly less than at 4 seconds. In other words, holding time is "useless alone but effective with high mold temperature." Single factor rotation would fail here: it only changes one factor at a time and would never discover this "team."
3. Perform ANOVA: The model is overall significant, with an R² of about 0.9, and the residuals are randomly distributed without a funnel shape. The interaction term A×D is significant, corroborating the interaction plot.

Conclusion: Mold temperature is the key factor, and holding time must be set in conjunction with mold temperature. Material temperature should be set to the higher level, and holding pressure can be freely chosen based on the lowest cost principle.

5. Confirmation Trials and Implementation

Five batches of confirmation trials were conducted using the optimal combination—mold temperature 90°C, material temperature 250°C, holding time 8 seconds, and holding pressure 60MPa. The average warping amount decreased from 2.1mm to 0.6mm, and the first-pass rate of snap-fit clip assembly increased from 82% to 99.5%. Confirmation trials cannot be skipped: before scaling up the small sample conclusion to the batch level, it must be verified with independent batches.

Subsequently, the parameters were solidified in the injection molding process card: mold temperature and holding time were included in SPC monitoring, holding pressure was set to the optimal cost value, and a parameter reference table was provided to the mold change personnel. Over three months of tracking, the warping amount remained stable with no rebound.

6. Three Common Misconceptions

Misconception 1: Hardcoding full factorial design when there are many factors. For more than 5 factors, start with partial factorial screening to narrow down the candidates to 3 to 4 before using full factorial design. The order cannot be reversed.

Misconception 2: Only looking at the main effects plot and not drawing the interaction plots. In this case, holding time appeared "useless" when considered alone, but it was actually a key lever under high mold temperature—missing the interaction can lead to completely opposite conclusions.

Misconception 3: Not performing randomization and replication. Conducting trials in sequence to save effort can mix noise into the effect estimation, leading to incorrect conclusions. Without replication, there is no way to discuss "significance."

The value of full factorial design is not in "doing a few more trials," but in clearly understanding the entanglements between effects in one go. When combined with partial factorial and response surface methods, it forms a complete DOE advancement path: screen first, then full, and finally optimize. For every quality professional who has been misled by interactions, 16 trials to achieve a "no-fail" conclusion is always worth it.

Main effects show the direction, interactions reveal the truth—each trial in full factorial design helps to clear the blind spots in the conclusion.

Knowledge code: 6.4.1

Version: v20260817

Author: Quality Think Tank Quality Think Tank is dedicated to providing quality management practitioners with systematic professional knowledge, methodologies, and practical tools to continuously enhance corporate quality capabilities.