Introduction to Design of Experiments (DOE): Finding the Optimal Parameter Combination with the Fewest Trials

Abstract: Process parameters are often "tested" one by one—adjust one parameter, observe the result, and then move on to the next. This method is not only time-consuming and labor-intensive but also often overlooks the interactions between parameters. Design of Experiments (DOE) uses statistical methods to arrange trials for multiple factors simultaneously, obtaining the most information with the fewest trials, making it a shortcut to finding the optimal process parameters. This article explains the complete path from designing to implementing a DOE using a five-step process and an example of injection molding.

1. Why Process Optimization Cannot Do Without DOE

The most common method of parameter adjustment on the shop floor is "one factor at a time": keeping all other conditions constant while changing only one parameter to observe the response. This method has two major drawbacks.

First, it fails to identify interactions. Two parameters may work fine when adjusted individually, but when combined, they may cause defects—this "1+1≠2" phenomenon can never be discovered with the one-factor-at-a-time approach. Second, it requires many trials and yields unreliable conclusions. Each adjustment of a factor necessitates a new round of trials, and the number of trials can skyrocket with multiple factors. Without statistical support, it is difficult to distinguish whether changes are due to the parameters or noise.

DOE takes the opposite approach: simultaneously changing multiple factors according to a pre-designed matrix, and then using statistical methods to separate the contribution of each factor and the interactions between factors. The same amount of information can often be obtained with only a fraction of the trials compared to the trial-and-error method. For this reason, DOE is widely used in process parameter optimization, formulation design, and reliability improvement, and is also a core tool in the Improve phase of Six Sigma DMAIC.

2. Remember Four Basic Concepts

• Factors and Levels: Factors are the parameters you want to adjust (such as temperature, pressure), and levels are the set values for each factor (for example, 220°C and 240°C are two levels).
• Response: The result metric you want to optimize (such as warpage, strength) must be measurable and repeatable.
• Main Effect: The average change in the response when a single factor is changed from its low level to its high level.
• Interaction: The phenomenon where the effect of one factor changes depending on the level of another factor, which is a unique "discovery capability" of DOE.

In terms of design types, mastering two types is sufficient for beginners: Full Factorial Design combines every level of each factor, providing the most comprehensive information and is suitable for scenarios with fewer factors (usually no more than 4); Fractional Factorial Design performs only a subset of the full factorial combinations, suitable for scenarios with more factors where initial screening is needed. Two practical concepts are Replication and Center Points: Replication is used to estimate experimental error and determine whether the effects are significant; Center Points (where each factor is set to its midpoint) are used to check if the response is linear, helping to decide whether a higher-order design is necessary.

3. Five-Step Implementation Process

Step One: Define the Response and Measurement Method. First, answer "What to optimize" — the response should be singular, quantifiable, and directly related to customer requirements, and the measurement system should be stable (it is recommended to perform an MSA first to confirm).

Step Two: Select Factors and Levels. Use a fishbone diagram and PFMEA to collect candidate factors, and select 3 to 6 key factors that are most likely to influence the response; for each factor, choose 2 to 3 levels, ensuring that the level intervals are wide enough and fall within the operable range.

Step Three: Choose the Design and Arrange the Trial Sequence. For fewer factors, select a full factorial design; for more factors, start with a fractional factorial screening. The trial sequence must be randomized to avoid environmental drift (such as temperature, shift, material batch) being mixed into the results.

Step Four: Conduct the Trials and Record Data. Strictly follow the design matrix, do not arbitrarily change parameters; record any abnormal conditions during the trials (such as material change, machine stoppage) as reference information for analysis.

Step Five: Analyze, Confirm, and Validate. Use main effect plots, interaction plots, and ANOVA to identify significant factors and the optimal combination, then validate the results under the optimal conditions to ensure the improvement is reproducible before incorporating it into the procedure document and control plan.

4. A Simple Example of Injection Molding

An injection-molded part has excessive warpage, and the quality team suspects it is related to barrel temperature, holding pressure, and holding time. They conduct a full factorial design with 3 factors and 2 levels, totaling 8 trials, with the warpage (mm) results as follows:

Calculate the main effects: increasing the temperature from 220°C to 240°C reduces the warpage by an average of 0.5mm, making it the most influential factor; increasing the pressure from 60 to 80MPa reduces the warpage by an average of 0.4mm; the effect of time is minimal. Interaction analysis shows a synergistic effect between temperature and pressure—warpage is lowest (1.1mm) at high temperature and high pressure. Based on this, the optimal combination is set to 240°C, 80MPa, and 5s. The warpage is verified three times, yielding results of 1.1mm, 1.2mm, and 1.1mm, all meeting the standard. The control plan is then updated and the parameters are solidified.

5. Common Pitfalls and Implementation Suggestions

• Pitfall One: More Factors Mean More Comprehensive. Too many factors can lead to an excessive number of trials or overly fragmented designs. Use fractional factorial or screening trials to lock in a few key factors first.
• Pitfall Two: Level Intervals Too Small. Levels that are too close together can result in effects being drowned out by noise, leading to "insignificant" conclusions that are actually due to insufficient level intervals.
• Pitfall Three: Not Randomizing. Conducting trials in a convenient sequence can lead to environmental changes being misinterpreted as factor effects.
• Pitfall Four: Only One Round Without Validation. DOE conclusions must be validated with independent trials to avoid mistaking coincidence for a rule.

When implementing, start small: choose a long-standing, cost-controllable process issue and run the entire process with 8 to 16 trials. Once you see the benefits, expand to more processes. Before each trial, ensure the measurement system is stable to avoid mistaking measurement errors for process effects.

6. Conclusion

The value of DOE lies not in statistical formulas but in systematically arranging trials instead of blindly trial-and-erroring: first, clearly define the response and factors, then design the most efficient trial plan, and finally, let the data speak and validate the results. It naturally complements FMEA and control plans—FMEA identifies risk factors, DOE finds the optimal parameters, and control plans solidify these parameters into daily management. Mastering this method transforms process optimization from a matter of "luck" to a matter of "planned results."

The essence of DOE is to use the fewest trials and the most scientific arrangement to turn parameter optimization from "trial and error" into "calculated conclusions."

Knowledge code: 6.4.1

Version: v20260805

Author: Quality Think Tank The Quality Think Tank is dedicated to providing systematic knowledge, methodologies, and practical tools for quality management professionals, helping companies continuously improve their quality capabilities.