Design of Experiments explained

Design of Experiments (DoE) is a structured, statistical methodology used to plan, conduct, analyse, and interpret controlled tests to evaluate the factors that may influence a particular outcome or process.

A visual summary of the Design of Experiments (DoE) cycle: plan the experiment, conduct trials, analyse data, and interpret results to optimise processes or products

It is widely used in engineering, manufacturing, scientific research, and quality improvement. Read more about how DoE fits in to our portfolio of developmental activities here: Development Services.

🔧 In Simple Terms:

Design of Experiments helps you determine which inputs (factors) have the most significant effect on an output (response) — and how to optimise that output through systematic testing rather than trial and error.

📌 Core Principles

Factors: Inputs or variables you can control (e.g. temperature, pressure, speed).
Levels: The values at which you test each factor (e.g. high/low or several steps).
Responses: The outcomes you measure (e.g. strength, yield, surface finish).
Interactions: When two or more factors affect the response jointly.
Replication: Repeating tests to estimate variability and improve confidence.
Randomisation: Running trials in a random order to reduce bias.
Blocking: Grouping similar experimental conditions to reduce noise from uncontrolled variables.

🧪 Example Use Case (Engineering)

Suppose you are optimising the surface finish of a machined part.

High-precision CNC machining with coolant delivery to optimise surface finish — a typical application area for Design of Experiments (DoE) in process optimisation

Factors may include:

Cutting speed
Feed rate
Tool type

Using DoE, you would plan a matrix of experiments that varies these factors systematically, rather than changing one at a time, allowing you to:

Quantify main effects and interactions
Model the process mathematically
Find the optimal settings

Call us

Use our contact form

📈 Common DoE Methods

DoE Type	Description	Use Case
Full Factorial	Tests all combinations of all factors and levels	Small experiments with few factors
Fractional Factorial	Uses a subset of combinations to reduce cost/time	Screening large numbers of factors
Taguchi Methods	Emphasises robustness against variation	Process control and reliability
Response Surface Methodology (RSM)	Fits a surface to explore optima	Precise optimisation after screening
Mixture Designs	Used when factors are proportions (sum to 100%)	Formulation problems (e.g. polymers, fuels)

✅ Benefits

Faster and cheaper than one-factor-at-a-time testing
Reveals interactions and nonlinearities
Improves process understanding and control
Leads to better quality and reliability

📚 Related Concepts

ANOVA (Analysis of Variance)
Regression modelling
Statistical process control (SPC)
Six Sigma

✅ When Are Full Factorial Designs Generally Used?

Full Factorial Designs are applied when you want to investigate all possible combinations of factor levels to:

Fully understand main effects and interactions
Develop accurate, predictive models
Avoid missing important relationships between variables

🔬 1. Early-Stage Product or Process Understanding

Ideal when you’re uncertain how different factors influence a response and want to evaluate all interactions.

Example: Studying the influence of temperature, pressure, and dwell time on adhesive bond strength.

🧪 2. Small Number of Factors (Typically ≤ 4)

The number of runs increases exponentially with the number of factors. For a design with $k$ factors, each at 2 levels, the number of experiments is:

$\text{Number of runs} = 2^k$

For $k = 3$ : $2^3 = 8$ runs
For $k = 4$ : $2^4 = 16$ runs
For $k = 5$ : $2^5 = 32$ runs (still manageable)

🧯 3. When Factor Interactions Are Important

Full factorials detect all main effects, two-factor, and higher-order interactions.

Example: In chemical processes, the effect of $\text{Temperature}$ may depend on $\text{Pressure}$ — a two-factor interaction.

📏 4. For Calibration or Precise Modelling

Full factorials provide a robust data set for:

Regression modelling
Response prediction
Understanding linear and interaction effects

✅ 5. Verification or Validation Studies

Used to validate critical systems where all combinations of settings must be evaluated to ensure performance or compliance.

🚫 When Not to Use Full Factorial Designs

Situation	Recommended Alternative
$k > 5$	Fractional Factorial or Screening Design
Limited time/resources	Taguchi or Plackett-Burman Design
Curved/nonlinear responses	Response Surface Methodology (RSM)
Mixture components	Mixture Designs (e.g., Simplex Centroid)

🔁 Summary – Full Factorial designs

Use full factorial designs when:

You have $2 \leq k \leq 5$ factors
All main effects and interactions matter
You need detailed models or high-resolution results
You can afford the time and cost of a full test matrix

Call us

Use our contact form

✅ When Are Fractional Factorial Designs Generally Used?

Fractional Factorial Designs are used when you want to study many factors efficiently, without testing all possible combinations. They let you:

Screen important factors quickly
Reduce the number of experiments
Still estimate main effects and selected interactions

These designs are ideal when resources are limited or only a subset of effects are expected to be significant.

🧪 1. Large Number of Factors (Typically $k \geq 5$ )

The number of runs in a full factorial grows exponentially:
$2^k$ runs for $k$ two-level factors.
For $k = 6$ : $2^6 = 64$ runs
For $k = 7$ : $2^7 = 128$ runs

A fractional factorial design (e.g. $2^{6-2} = 16$ runs) tests only a fraction (e.g. $\frac{1}{4}$ ) of the full matrix — saving time and cost.

🔍 2. Factor Screening Studies

Use fractional designs to identify the most influential factors from a larger pool.

Example: A design team investigates 7 factors affecting injection moulding quality using a $2^{7-4} = 8$ run design.

⚠️ 3. When Higher-Order Interactions Are Negligible

Fractional designs assume that three-way and higher interactions are insignificant — so these can be aliased (confounded) with main effects or two-way interactions.

This trade-off is acceptable in early-stage experiments or when domain knowledge supports the assumption.

💡 4. Efficient Experimentation with Limited Resources

Ideal for:

Academic research with limited test time
Prototyping phases
Early process development

They provide actionable insight while controlling costs.

🧠 How They Work: Resolution Concept

Fractional designs are classified by resolution, which defines what effects are aliased:

Resolution	Meaning
$\text{III}$	Main effects may be aliased with two-factor interactions
$\text{IV}$	Main effects clear of two-factor interactions; two-factors may be aliased with each other
$\text{V}$	Main and two-factor effects are unconfounded; suitable for detailed modelling

Higher resolution = less aliasing = more reliable effect estimates.

🚫 When Not to Use Fractional Factorial Designs

Situation	Use Instead
$k < 4$ and cost is not prohibitive	Full Factorial
High-order interactions are important	Full or Augmented Design
Model curvature is expected	Response Surface Methodology (RSM)
Strong regulatory/validation needs	Full Factorial or Central Composite Design

🔁 Summary – Fractional Factorial designs

Use fractional factorial designs when:

You need to evaluate $k \geq 5$ factors efficiently
Only main effects and low-order interactions are important
You want to screen variables before optimisation
You have resource or time constraints
Some aliasing is acceptable for the problem stage

Call us

Use our contact form

✅ When Are Taguchi Designs Generally Used?

Taguchi Designs are used when the goal is to improve robustness and optimise quality, especially in manufacturing and product design.

Two engineers evaluating design robustness in an industrial setting using experimental data — a practical application of Design of Experiments (DoE) for engineering development

They focus on identifying control factor settings that minimise variation due to noise factors (i.e. uncontrollable external conditions).

🧰 1. When You Want to Improve Robustness

Taguchi designs are ideal when the priority is not just optimising the mean response, but also reducing variability and making the system insensitive to external disturbances.

Example: Designing a car seat adjuster that works consistently across temperature extremes, voltage fluctuations, and wear.

📉 2. When Signal-to-Noise Ratio Matters

Instead of just analysing the response mean, Taguchi methods use signal-to-noise (S/N) ratios to quantify robustness:

Higher S/N = More Robust Performance

Common S/N formulations:

“Larger-the-better”:
$S/N = -10 \cdot \log_{10} \left( \frac{1}{n} \sum_{i=1}^{n} \frac{1}{y_i^2} \right)$
“Smaller-the-better”:
$S/N = -10 \cdot \log_{10} \left( \frac{1}{n} \sum_{i=1}^{n} y_i^2 \right)$
“Nominal-the-best”:
$S/N = 10 \cdot \log_{10} \left( \frac{\bar{y}^2}{s^2} \right)$

🧪 3. When You Have Many Factors and Limited Runs

Taguchi designs use orthogonal arrays (OAs) to test many factors with a small number of experiments.

Example: $L_9$ array → 4 factors at 3 levels in just 9 runs
Example: $L_{16}$ array → up to 15 factors at 2 levels in 16 runs

These arrays maintain balance and orthogonality, ensuring that factor effects are statistically independent.

🧬 4. When You Can Distinguish Control vs. Noise Factors

Taguchi’s outer-inner array strategy allows you to:

Test control factors in the inner array
Simulate real-world variation via noise factors in the outer array

This method is central to Robust Parameter Design.

🛠️ 5. When You Want a Simplified Yet Structured Approach

Taguchi methods are often used in:

Six Sigma DMAIC projects
Lean manufacturing environments
Industrial R&D

They are easier to apply than full regression modelling, making them accessible to teams with limited statistical training.

🚫 When Not to Use Taguchi Designs

Situation	Use Instead
You need detailed interaction modelling	Full or Fractional Factorial
You need to model nonlinear curvature	Response Surface Methodology (RSM)
You require regression-based predictive models	RSM or DOE regression
You must estimate all two-way interactions independently	Factorial Designs with Resolution V

🔁 Summary – Taguchi designs

Use Taguchi designs when:

You want to optimise robustness and reduce variability
You need a small, structured design with many factors
You want to improve quality and consistency
You’re focused on practical industrial optimisation
You’re using S/N ratios to measure performance quality

Call us

Use our contact form

✅ When Are Response Surface Methodology (RSM) Designs Generally Used?

Response Surface Methodology (RSM) is used when you want to optimise a process or refine a model after key factors have been identified.

Illustration of an engineer optimising predictive models through data analysis and process refinement — a key stage in advanced Design of Experiments (DoE) and Response Surface Methodology (RSM)

RSM explores curvature in the response surface, allowing you to find maximum, minimum, or target values of the response.

🧪 1. When the System is Nonlinear

RSM assumes that the response depends on factors through a curved surface — typically modelled as a second-order (quadratic) polynomial:

$Y = \beta_0 + \sum \beta_i X_i + \sum \beta_{ii} X_i^2 + \sum \beta_{ij} X_i X_j + \varepsilon$

$X_i$ : coded input variables
$Y$ : predicted response
$\beta_{ii}$ : curvature (quadratic) terms
$\beta_{ij}$ : interaction terms

📈 2. When Optimising a Response

RSM is ideal for locating:

Peaks (maximum output)
Valleys (minimum cost, error, etc.)
Stationary points (e.g., product meets a target specification)

🧰 3. After Screening is Complete

RSM is typically applied after a screening design, like a fractional factorial, has identified the important factors. These are then studied in more depth using RSM.

🧮 4. When You Want to Fit a Predictive Model

RSM allows you to construct accurate regression-based models that can:

Predict response for any setting within the factor region
Explore interactions and curvature
Support multi-objective trade-offs

📊 Common RSM Designs

Design	Description	Number of Levels
Central Composite Design (CCD): see here	Augments a factorial with centre and axial points	5 levels
Box-Behnken Design (BBD): see here	Balanced design using midpoints of edges	3 levels
Face-Centered CCD (CCF)	A practical alternative with axial points on cube faces	3 levels

🧬 5. When You Want Model-Based Optimisation Tools

RSM enables advanced tools like:

Contour plots and 3D surface plots
Canonical analysis of stationary points
Desirability functions for multi-response optimisation

🚫 When Not to Use RSM

Situation	Better Alternative
Many unknown or unimportant factors	Start with a screening design (e.g., fractional factorial)
High-order nonlinearities (> quadratic)	Use neural nets or machine learning approaches
Few factors and no curvature expected	Use full factorial or Taguchi designs
Response is binary or categorical	Use logistic regression or classification models

🔁 Summary – RSM designs

Use RSM designs when:

You’ve already identified key factors ( $\leq 5$ )
The response surface is curved
You want to optimise or refine a product or process
You need accurate, interpretable models
You’re seeking the best combination of inputs for performance

Call us

Use our contact form

✅ When Are Mixture Designs Generally Used?

Mixture Designs are used when the factors are components of a mixture, and their proportions — rather than absolute amounts — affect the response. A key constraint is that the sum of all component proportions must equal 1 (or 100%).

🧪 1. When Inputs Are Proportional and Constrained

In a mixture experiment, the variables represent fractions or percentages of components. Thus, the defining constraint is:

$x_1 + x_2 + \dots + x_q = 1$

Where:

$x_i$ : proportion of the ithi^\text{th}ith component
$q$ : number of mixture components

This constraint introduces dependency between variables — increasing one component must decrease at least one other.

🧬 2. When the Product is a Blend or Formulation

Common in:

Food science (e.g. proportions of fat, sugar, protein)
Pharmaceuticals (e.g. excipient mixtures)
Polymers and composite materials
Paints, lubricants, fuels, adhesives

Example: Optimising a fuel blend with components A, B, and C such that $x_A + x_B + x_C = 1$

📐 3. When Factorial or RSM Designs Are Inappropriate

Traditional factorial or response surface designs do not respect the constraint $\sum x_i = 1$ , and can generate invalid formulations.

Mixture designs ensure all experimental runs fall within the feasible simplex (i.e. inside the triangle, tetrahedron, etc. of valid mixtures).

📊 Common Mixture Design Types

Design	Description	Suitable When
Simplex Lattice	Regular grid of mixtures over the simplex	All combinations are equally spaced
Simplex Centroid	Uses all pure blends, midpoints, and overall centroid	Useful for main effect estimation
Constrained Mixture	Boundaries or ratios are imposed on component levels	For formulations with min/max limits
D-Optimal Mixture	Custom design optimised for model terms	For irregular or costly experiments

🧮 4. When the Response is a Function of Proportions

The canonical mixture model is:

$Y = \sum \beta_i x_i + \sum \beta_{ij} x_i x_j + \sum \beta_{ijk} x_i x_j x_k + \dots + \varepsilon$

Where:

$Y$ : response
$\beta_i$ , $\beta_{ij}$ : blending coefficients
Only cross-product terms appear — there are no $x_i^2$ terms because of the constraint

🔁 5. When You Want to Optimise a Formulation

Mixture designs allow you to:

Map the response surface over the mixture space
Use contour plots to visualise optimal blend regions
Apply desirability functions for multi-response optimisation

🚫 When Not to Use Mixture Designs

Situation	Use Instead
Inputs are independent (not part of a mixture)	Use Factorial or RSM designs
Total amount, not proportion, affects response	Consider factorial with quantitative factors
Mixtures are nested within process variables	Use combined mixture-process designs
Categorical inputs are involved	Use mixture with categorical blocks or a hybrid approach

🔁 Summary – Mixtures designs

Use mixture designs when:

The inputs are proportions of components and must sum to 1
You’re optimising a blend, recipe, or formulation
Traditional DOE methods violate mixture constraints
You need to model interactions between mixture components
You want a structured, efficient way to explore the mixture space

Call us

Use our contact form

✅ When Are Mixture Designs with Process Variables Used?

These designs are used when the final response depends on both:

The composition of a mixture (e.g. percentages of ingredients)
One or more external process settings (e.g. curing temperature, time, stirring speed)

They enable simultaneous optimisation of formulation and operating conditions.

A temperature-controlled mixer blending a complex liquid formulation — illustrating the role of mixture designs with process variables in experimental optimisation of product performance

🧪 1. When You Need to Optimise Both Formula and Process

Mixture-process designs let you quantify how process variables affect the performance of different blends.

Example: A polymer adhesive is made from components A, B, and C, but its final bond strength also depends on curing temperature and pressure.

🔬 2. How the Design Structure Works

You define:

Mixture components x1,x2,…,xqx_1, x_2, \dots, x_qx1,x2,…,xq, such that: $\sum_{i=1}^{q} x_i = 1$
Process variables z1,z2,…,zpz_1, z_2, \dots, z_pz1,z2,…,zp which are unconstrained.

The model may include:

Mixture terms: $x_1, x_2, x_1 x_2, \dots$
Process terms: $z_1, z_2, z_1^2, \dots$
Interaction terms: $x_i z_j$ , i.e. mixture-by-process interactions

🧮 3. Typical Model Form

$Y = \sum \beta_i x_i + \sum \beta_{ij} x_i x_j + \sum \gamma_k z_k + \sum \delta_{ik} x_i z_k + \varepsilon$

Where:

$x_i$ = mixture proportions
$z_k$ = process variables
$\delta_{ik}$ = interaction terms capturing how process variables influence each mixture component

📊 4. Common Use Cases

Application Area	Mixture Components	Process Variables
Food technology	Ingredients (e.g. flour, fat, sugar)	Baking temperature, time
Pharmaceuticals	Active & excipient mix	Granulation speed, humidity
Polymer formulation	Resins, fillers, hardeners	Curing pressure, heat rate
Coatings	Solvent blend	Spray rate, ambient conditions

📈 5. Design Options

Designs can be built as:

Combined Mixture-Process Designs (crossed designs)
Split-plot Designs (when process variables are harder to change)
Custom Designs (D-Optimal) for irregular regions or constraints

Design software like Design-Expert, JMP, and MODDE support these complex layouts.

🚫 When Not to Use Mixture-Process Designs

Situation	Better Approach
Only mixture composition matters	Use standard Mixture Design
Only process variables matter	Use RSM (e.g., Central Composite Design)
Too many variables for available runs	Use screening designs first
Process variables are categorical	Use blocking or mixed-model approaches

🔁 Summary – Mixtures designs with process variables

Use mixture designs with process variables when:

The response is influenced by both formulation and process conditions
You want to optimise both simultaneously
You’re exploring interactions between composition and operating settings
Standard mixture or RSM designs are too limited for your system

Call us

Use our contact form

👥 Who Uses DoE and Why

Design of Experiments (DoE) is used by professionals and researchers across many disciplines who need to understand, improve, or optimise a process, product, or system. Here’s a breakdown of who uses DoE and why:

🧑‍🔬 1. Engineers

Mechanical, Electrical, Chemical, and Civil Engineers
Use DoE to:
- Optimise design parameters
- Improve product performance
- Identify critical tolerances
- Reduce failure rates
Example: A mechanical engineer tests combinations of material thickness and heat treatment temperature to maximise fatigue strength.

🏭 2. Manufacturing & Process Engineers

Use DoE to:
- Minimise scrap and defects
- Improve process yield and cycle time
- Reduce variability and cost
Example: A production line engineer designs an experiment to find the best combination of curing time, pressure, and temperature for moulded parts.

⚗️ 3. Scientists & Researchers

Fields: Chemistry, Biology, Materials Science, Environmental Science, etc.
Use DoE to:
- Plan efficient laboratory experiments
- Explore interactions between multiple reagents or conditions
- Optimise reactions or processes
Example: A chemist uses DoE to find the optimal catalyst concentration, pH, and reaction time for maximum yield.

🔬 4. Quality Professionals

Roles: Quality Engineers, Six Sigma Black Belts, QA/QC Analysts
Use DoE to:
- Identify root causes of defects
- Improve product reliability
- Implement robust processes
Example: A Six Sigma project team uses a factorial design to investigate the cause of variation in product dimensions.

💊 5. Pharmaceutical & Medical Device Developers

Use DoE to:
- Validate manufacturing processes (per FDA or EMA guidelines)
- Optimise drug formulations
- Ensure consistent product performance
Example: A pharmaceutical company uses response surface methodology (RSM) to optimise tablet disintegration time.

📊 6. Data Scientists & Statisticians

Use DoE to:
- Structure controlled testing environments
- Create interpretable models with experimental data
- Support data-driven decisions
Example: A statistician applies DoE to an A/B/n test involving different interface layouts in software development.

🚗 7. Automotive & Aerospace Developers

Use DoE to:
- Tune control systems
- Reduce NVH (noise, vibration, harshness)
- Develop lightweight, high-strength structures
Example: An aerospace design team tests wing structures with different rib spacing and material layups for maximum stiffness-to-weight ratio.

🧪 8. Food, Cosmetics, and Consumer Product Scientists

Use DoE to:
- Optimise sensory qualities and shelf life
- Improve formulation stability
- Reduce costs without compromising quality
Example: A food technologist designs a mixture experiment to balance taste, texture, and nutrition in a new product.

🎯 Summary Table – Users of DoE

User Type	Typical Goal	Example
Engineer	Optimise product or system performance	Maximise heat exchanger efficiency
Manufacturer	Improve process efficiency and quality	Reduce defect rate in casting
Scientist	Run informative experiments efficiently	Maximise enzyme yield
Quality Professional	Reduce variation, ensure compliance	Root cause analysis in Six Sigma
Pharma/Medical	Validate processes, optimise formulations	Minimise dissolution time
Data Scientist	Structure controlled testing	Website feature testing
Aerospace/Auto	Improve reliability and weight	Optimise crash resistance
Consumer Product Dev	Improve sensory and stability properties	Optimise lotion formulation

💻Which Software Should I Use for DoE?

There are several software platforms available for planning, conducting, recording, and analysing Design of Experiments (DoE). These range from specialist statistical tools to more general-purpose data science platforms, and their capabilities vary depending on the depth of analysis and regulatory needs.

✅ 1. Specialist DoE Software (Purpose-Built)

These tools are specifically designed for experimental design and analysis:

🧪 Design-Expert (Stat-Ease)

Strengths: Comprehensive DoE capabilities including factorial, response surface, mixture, Taguchi, and robust designs.
Features:
- Graphical interface for planning
- Powerful diagnostics (ANOVA, residuals)
- Contour and 3D surface plots
Best for: Engineers and scientists focused solely on experimentation.
OS: Windows
Website: www.statease.com

📊 JMP (SAS Institute)

Strengths: Visual, interactive analysis environment with deep statistical tools.
Features:
- Drag-and-drop DoE planning
- Interactive dashboards
- Data exploration and modelling
Best for: R&D and Six Sigma teams needing visual exploration.
OS: Windows and Mac
Website: www.jmp.com

🧬 Minitab

Strengths: Widely used in quality engineering and Six Sigma.
Features:
- DoE wizard
- Built-in templates for factorial, Taguchi, and RSM
- Extensive statistical tests and control charts
Best for: Quality improvement professionals.
OS: Windows (with web version)
Website: www.minitab.com

🧰 2. General Data Science Platforms (With DoE Plugins/Scripting)

These are more flexible but require scripting or setup:

🐍 Python (via libraries like `pyDOE`, `statsmodels`, `scikit-learn`)

Strengths: Open-source and scriptable.
Use Cases:
- Generate design matrices (full/fractional factorials, central composite, Latin hypercube)
- Analyse via regression or ANOVA
- Automate result plotting (e.g. with matplotlib or plotly)
Best for: Developers and technical engineers comfortable with code.
Limitations: No GUI; steep learning curve for non-programmers.

🟨 R (via `DoE.base`, `rsm`, `FrF2`)

Strengths: Rich in statistical depth and free to use.
Use Cases:
- All classical and modern experimental designs
- Sophisticated model fitting and diagnostics
- Publication-quality graphs
Best for: Statisticians and academic researchers.

📊 Excel (with add-ins like XLSTAT, QI Macros, or JMP Add-In)

Strengths: Ubiquitous and familiar.
Use Cases:
- Basic factorial designs
- Taguchi L4–L32 orthogonal arrays
- Simple ANOVA and trend analysis
Best for: Beginners or teams without specialist software.

📁 3. Enterprise & Regulatory Platforms

Used in regulated industries like pharmaceuticals and medtech.

💊 MODDE (Sartorius)

Strengths: Targeted for QbD (Quality by Design) in pharma and biotech.
Features:
- Process understanding & DoE
- Regulatory compliance (FDA/EMA)
- Data visualisation
Best for: Regulated industries requiring validated tools.

📦 Fusion QbD / DryLab

Strengths: Focused on chromatography and formulation optimisation.
Best for: Analytical chemists and regulatory R&D.

🧮 Summary Table – DoE Software

Software	DoE Design	Analysis	Visualisation	Best For
Design-Expert	✓✓✓	✓✓✓	✓✓	Engineers, R&D
JMP	✓✓✓	✓✓✓	✓✓✓	Visual analysts, R&D
Minitab	✓✓	✓✓✓	✓✓	Quality teams
Python	✓✓✓	✓✓✓	✓✓✓	Programmers, automation
R	✓✓✓	✓✓✓	✓✓✓	Statisticians
Excel (+Add-ins)	✓	✓	✓	Beginners
MODDE	✓✓✓	✓✓✓	✓✓	Pharma, biotech
Fusion QbD	✓✓	✓✓	✓✓	Analytical chemists

🔧 In Simple Terms:

📌 Core Principles

🧪 Example Use Case (Engineering)

📈 Common DoE Methods

✅ Benefits

📚 Related Concepts

✅ When Are Full Factorial Designs Generally Used?

🔬 1. Early-Stage Product or Process Understanding

🧪 2. Small Number of Factors (Typically ≤ 4)

🧯 3. When Factor Interactions Are Important

📏 4. For Calibration or Precise Modelling

✅ 5. Verification or Validation Studies

🚫 When Not to Use Full Factorial Designs

🔁 Summary – Full Factorial designs

✅ When Are Fractional Factorial Designs Generally Used?

🧪 1. Large Number of Factors (Typically )

🔍 2. Factor Screening Studies

⚠️ 3. When Higher-Order Interactions Are Negligible

💡 4. Efficient Experimentation with Limited Resources

🧠 How They Work: Resolution Concept

🚫 When Not to Use Fractional Factorial Designs

🔁 Summary – Fractional Factorial designs

✅ When Are Taguchi Designs Generally Used?

🧰 1. When You Want to Improve Robustness

📉 2. When Signal-to-Noise Ratio Matters

🧪 3. When You Have Many Factors and Limited Runs

🧬 4. When You Can Distinguish Control vs. Noise Factors

🛠️ 5. When You Want a Simplified Yet Structured Approach

🚫 When Not to Use Taguchi Designs

🔁 Summary – Taguchi designs

✅ When Are Response Surface Methodology (RSM) Designs Generally Used?

🧪 1. When the System is Nonlinear

📈 2. When Optimising a Response

🧰 3. After Screening is Complete

🧮 4. When You Want to Fit a Predictive Model

📊 Common RSM Designs

🧬 5. When You Want Model-Based Optimisation Tools

🚫 When Not to Use RSM

🔁 Summary – RSM designs

✅ When Are Mixture Designs Generally Used?

🧪 1. When Inputs Are Proportional and Constrained

🧬 2. When the Product is a Blend or Formulation

📐 3. When Factorial or RSM Designs Are Inappropriate

📊 Common Mixture Design Types

🧮 4. When the Response is a Function of Proportions

🔁 5. When You Want to Optimise a Formulation

🚫 When Not to Use Mixture Designs

🔁 Summary – Mixtures designs

✅ When Are Mixture Designs with Process Variables Used?

🧪 1. When You Need to Optimise Both Formula and Process

🔬 2. How the Design Structure Works

🧮 3. Typical Model Form

📊 4. Common Use Cases

📈 5. Design Options

🚫 When Not to Use Mixture-Process Designs

🔁 Summary – Mixtures designs with process variables

👥 Who Uses DoE and Why

🧑‍🔬 1. Engineers

🏭 2. Manufacturing & Process Engineers

⚗️ 3. Scientists & Researchers

🔬 4. Quality Professionals

💊 5. Pharmaceutical & Medical Device Developers

📊 6. Data Scientists & Statisticians

🚗 7. Automotive & Aerospace Developers

🧪 8. Food, Cosmetics, and Consumer Product Scientists

🎯 Summary Table – Users of DoE

💻Which Software Should I Use for DoE?

✅ 1. Specialist DoE Software (Purpose-Built)

🧪 Design-Expert (Stat-Ease)

📊 JMP (SAS Institute)

🧬 Minitab

🧰 2. General Data Science Platforms (With DoE Plugins/Scripting)

🐍 Python (via libraries like pyDOE, statsmodels, scikit-learn)

🟨 R (via DoE.base, rsm, FrF2)

📊 Excel (with add-ins like XLSTAT, QI Macros, or JMP Add-In)

📁 3. Enterprise & Regulatory Platforms

💊 MODDE (Sartorius)

📦 Fusion QbD / DryLab

🧮 Summary Table – DoE Software

🧪 1. Large Number of Factors (Typically $k \geq 5$ )

🐍 Python (via libraries like `pyDOE`, `statsmodels`, `scikit-learn`)

🟨 R (via `DoE.base`, `rsm`, `FrF2`)