Introduction of Dr. George Box:
Dr. George Box is a renowned expert in experimental design and quality improvement. He holds a Ph.D. and a Doctor of Science degree in mathematical statistics from the University of London. Dr. Box is a professor emeritus and director of research at the Center for Quality and Productivity Improvement at the University of Wisconsin at Madison. He is well-known for developing techniques like response surface methodology (RSM) and evolutionary operation (EVOP).

Experimental Design in the Quality Movement:
Experimental design is not a cure-all for quality improvement. It fits into the context of the quality movement, which focuses on the entire organization. Statistical design is used to investigate and understand the relationship between various factors and the quality of the product or service. It helps identify the key factors that impact quality and optimize them for better results.

Experimental Design for Quality Improvement:
Experimental design is a systematic approach to investigating the relationship between factors and their effects on a response. It involves careful planning and conducting experiments to collect data and analyze the results. By understanding these relationships, manufacturers can determine which factors are most important and how to adjust them to achieve the desired outcome. Experimental design enables efficient and effective quality improvement by minimizing trial and error and optimizing processes.

Applications of Experimental Design:
Experimental design is used in various industries, including manufacturing, healthcare, and engineering, to improve quality. It helps optimize processes, reduce defects, and enhance product or service performance. Examples include: Optimizing a chemical process to increase yield Determining the best combination of ingredients for a new product Designing a manufacturing process to minimize defects

Scientific Method for Continuous Quality Improvement:
Quality movement is about democratizing scientific method for continuous quality improvement. Quality is about learning and knowledge generation.

Pre-Science to Democratized Science:
Pre-science phase: Scientific approach less evident, knowledge and learning limited. Science phase: Science regarded as specialized, accessible only to select few. Democratized science (quality movement): Science simplified, accessible to everyone. Rapid rate of change in environment due to democratized science.

Science as a Catalyst for Learning:
Sharing knowledge: Isaac Newton’s theory of gravitation, combining ideas from various sources. Informed observation: Charles Darwin’s theory of evolution, observing natural world without experimentation. Directed experimentation: Benjamin Franklin’s kite experiment, testing hypothesis through controlled experiment.

Democratization of Science in Quality Movement:
Sharing knowledge: Knowledge sharing and collaboration emphasized in quality improvement. Informed observation: Data collection and analysis to understand processes and identify improvement opportunities. Directed experimentation: Controlled experiments and pilot studies to test improvement interventions.

Conclusion:
Democratization of science in quality movement enables continuous learning and improvement. Sharing knowledge, informed observation, and directed experimentation are key methods for driving quality improvement.

Shared Knowledge:
George Box emphasizes the significance of shared knowledge in quality improvement, citing examples such as libraries, brainstorming, and teams. Collaboration, rather than isolated efforts, is essential for effective problem-solving and innovation. Breaking down barriers between departments and functions is crucial for seamless processes and better outcomes.

Informed Observation:
Quality tools and methods, such as defeating Murphy, are examples of informed observation in action. Paying attention to details, identifying anomalies, and investigating causes are vital aspects of process improvement. A famous example is the analysis of bullet holes in B-17s during World War II to determine where armor plates should be placed.

Iterative Learning:
Continuous quality improvement is not a one-time optimization but an ongoing process of learning and refinement. Assumptions about variables, measurements, and relationships must be continuously challenged and refined. The process involves moving from one idea to another, modifying hypotheses based on data, and gradually converging towards better solutions.

Inductive-Deductive Iteration:
Box introduces the concept of inductive-deductive iteration, a cycle of formulating hypotheses, gathering data, and refining theories. This iterative process is essential for understanding and improving complex systems. The Schuart-Deming cycle, a popular representation of this process, emphasizes the importance of continuous improvement through iterative learning.

Left Brain vs. Right Brain:
Recent discoveries have confirmed the existence of two distinct brain hemispheres, each with its own functions. The left brain is primarily responsible for analysis and deduction, while the right brain is associated with creativity and induction. These hemispheres communicate through the corpus callosum, facilitating the iterative process of learning and problem-solving.

Data and Facts:
Box highlights the importance of data and facts in the learning process, emphasizing that the process is applicable across various fields of research. Data should not be limited to quantitative measurements but can include observations, qualitative insights, and historical information.

Design of Experiments:
George Box emphasizes the importance of experimental design to avoid arguments and ensure valid comparisons. R.E. Fisher, a genius statistician, developed the principles of experimental design in the 1920s. The goal is to eliminate biases and confounding factors, leading to reliable conclusions.

Paired Comparison Test:
In a paired comparison test, two treatments are compared simultaneously by the same person under the same conditions. This method eliminates the influence of extraneous factors and provides a valid comparison. Randomization of the order of treatments further enhances the validity of the comparison.

Factorial Experimentation:
Factorial experimentation allows for the simultaneous investigation of multiple factors and their interactions. This approach provides a more comprehensive understanding of the effects of different factors and their combinations. An example from SKF illustrates how factorial experimentation can be applied in a manufacturing setting to optimize production processes.

Avoiding Arguments in Experiments:
Box stresses the importance of considering all potential sources of variation and controlling for them in the experimental design. He cautions against drawing conclusions from experiments without proper design and analysis, which can lead to inconclusive or misleading results. Randomization and blocking are essential techniques for eliminating biases and ensuring valid comparisons.

Qualitative vs. Quantitative Data:
Box highlights the value of qualitative data, which is often overlooked in favor of quantitative data. Qualitative data can provide valuable insights and context for understanding experimental results. He encourages researchers to consider both quantitative and qualitative data to gain a more comprehensive understanding of the studied phenomena.

Overview:
This section introduces the concept of factorial experiments, highlighting their advantages over one-factor-at-a-time experiments. It also discusses the origin of the term “factorial experiment” and presents a real-life example demonstrating the significant cost savings achieved by using a factorial experiment.

Benefits of Factorial Experiments:
Factorial experiments allow researchers to study the effects of multiple factors simultaneously, providing a more comprehensive understanding of their interactions. They are more efficient than one-factor-at-a-time experiments, requiring fewer experiments and resources. Factorial experiments can reveal important interactions between factors that would be missed in one-factor-at-a-time experiments.

Origin of the Term “Factorial Experiment”:
The term “factorial experiment” was coined by Frank Yates, Ronald Fisher’s first deputy assistant, due to the involvement of multiple factors in the experiment.

Case Study: Bearing Wear Experiment:
A real-life example of a factorial experiment is presented, involving the testing of different factors affecting the wear of bearings. The experiment demonstrated that changing two factors (osculation and heat) together resulted in a significant increase in bearing wear compared to changing each factor individually. This finding led to substantial cost savings in the manufacturing process.

Significance of Interactions:
The experiment highlights the importance of considering interactions between factors, as they can have a significant impact on the results. Many real-world phenomena, such as sexual reproduction, depend on interactions between factors.

Untapped Potential of Factorial Experiments:
Due to the prevalence of one-factor-at-a-time experiments, there are numerous opportunities to discover new insights and potential cost savings by using factorial experiments.

Call to Action:
Viewers are encouraged to call or fax in with questions or comments during the break. The phone number for calling in by voice is 800-442-4613, and the fax number is 800-760-4242.

Background of LHC Tippett:
Tippett, a statistician, was highly regarded by George Box. He emphasized the value of simplicity in experimental design. Tippett’s early work in the 1930s involved using orthogonal arrays for experiments.

Applications of Experimental Designs:
Designs can examine changes in mean and variation. Quality control charts can visually illustrate these changes.

Tippett’s Work on Variation:
Tippett used experimental designs to identify factors causing variation in a spinning machine. Changes in factors could lead to decreased variation, improving process stability.

Identifying and Reducing Cycles:
Box highlighted an example where a cycling issue was discovered in a process. Factorial experiments helped determine how to reduce the cycle. Identifying the source of the cycle, such as a belt, led to a solution.

Responsive Methods for Yield Improvement:
Box introduced responsive methods to enhance yields. These methods involve graphically representing data in multiple dimensions. By visualizing the relationship between factors and yield, optimal conditions can be identified.

Visualizing Multi-Dimensional Data:
Box explained the concept of visualizing data in more than one dimension. Imagine a series of yield vs. temperature graphs at different concentration levels. Connecting these graphs creates a surface that represents the relationship between factors and yield.

Response Surfaces and Contour Graphs:
Response surfaces are a graphical representation of the relationship between multiple independent variables and a single dependent variable. Contour graphs are a series of lines connecting points of equal value on a response surface. They provide a visual representation of the relationship between variables and help identify optimal conditions.

Three-Factor Experiments and Special Response Surface Designs:
Three-factor experiments involve three independent variables and are often used to create response surfaces. Special response surface designs can be used to reduce the number of experiments required to create a response surface.

Multi-Response Experiments:
Multi-response experiments involve measuring multiple dependent variables. It’s important to consider the relationships between these responses when designing and analyzing experiments.

Example: Dog Food Experiment:
An experiment was conducted to optimize the manufacturing process of dog food by reducing the amount of powder in the final product. Three factors (temperature, flow, and compression zone) were varied in a 2 cubed factorial design. The results showed a significant reduction in powder content under certain conditions.

The Current State:
The initial results show a significant decrease in power in the product, from 132 to 92, which is a one-third decrease. However, the goal is to minimize the power in the product, ideally reaching zero.

Contours and Interpolation:
To understand the direction of improvement, it is helpful to visualize the data using contours. Contours are lines connecting points with the same value, similar to isothermal lines on a weather map. By interpolating between the data points, one can estimate the contour lines for different values, such as 120, 110, and 100.

The Trend and the Economic Aspect:
The general trend of the data suggests that the power in the product decreases as one moves towards a particular corner. However, economic considerations also play a role in determining the optimal path for improvement.

Conclusion:
The analysis of experimental data using contours and economic considerations helps determine the direction for further improvement and optimization.

Defining the Problem:
Researchers were conducting experiments to optimize a manufacturing process, but faced conflicting objectives, such as maximizing yield while minimizing energy usage and powder content in the product.

Cost Considerations:
The cost of each objective was assessed, enabling the researchers to quantify the trade-offs between different outcomes.

Break-even Point:
A break-even point was identified, representing the optimal balance where all objectives are met at the same cost.

Further Experimentation:
Areas for further experimentation were identified, considering limitations such as maximum temperature and equipment constraints.

The 25% Rule:
The 25% rule emphasizes the importance of allocating a limited portion of the budget (about 25%) to the first experiment, acknowledging that multiple experiments will be necessary.

Fisher’s Insight:
The quote “the best time to plan an experiment is after you’ve done it” highlights the iterative nature of experimentation and the importance of learning from initial results.

Screening Designs:
Screening designs, including fractional designs and orthogonal arrays, were developed during the war in England and are useful for identifying important factors and interactions in complex experiments.

Design of Experiments: Factorial, Fractional Factorial, and Orthogonal Array
Factorial designs involve testing all possible combinations of factors, while fractional factorial designs involve testing only a subset of combinations. Orthogonal arrays are used in screening experiments to identify the most important factors affecting a process or system.

Screening Designs and Orthogonal Arrays
Screening designs are used in the early stages of an investigation to identify the important factors affecting a process or system. Orthogonal arrays are useful for screening experiments because they allow for the efficient estimation of the effects of multiple factors.

Example of an Extrusion Process Experiment
An example is given of an extrusion process experiment with eight factors being screened: mold temperature, moisture content, holding pressure, cavity thickness, booster pressure, cycle time, gate size, and screw speed.

Fractional Factorial Designs:
In experimental design, fractional factorials are used to reduce the number of experimental runs required to study the effects of multiple factors. This is particularly useful when the number of factors is large, as it can significantly reduce the time and resources needed to conduct the experiment. Fractional factorials involve running a carefully selected subset of the total possible combinations of factor levels.

Orthogonal Arrays:
Orthogonal arrays are a specific type of fractional factorial design where the columns of the design matrix have a special balance property. This balance ensures that the effects of different factors are not correlated, making it easier to separate and analyze the individual effects of each factor. Orthogonal arrays are commonly used in experimental design for their efficiency and ability to provide reliable results with a reduced number of runs.

L16 Orthogonal Array Example:
The L16 orthogonal array is a specific design with 16 runs and 8 factors. It is a two-level orthogonal array, meaning that each factor has two levels (e.g., low and high). The L16 orthogonal array allows researchers to study the effects of 8 factors with only 16 experimental runs, which is significantly less than the 256 runs required for a full factorial design.

Advantages of Fractional Factorials and Orthogonal Arrays:
Reduced number of experimental runs: Fractional factorials and orthogonal arrays significantly reduce the number of runs required to study the effects of multiple factors, saving time and resources. Efficient use of resources: By using a carefully selected subset of the total possible combinations, these designs make efficient use of experimental resources. Ability to separate factor effects: Orthogonal arrays, in particular, provide a balanced design where the effects of different factors are not correlated, making it easier to analyze and interpret the results.

Availability of Design Tables and Computer Programs:
Design tables and computer programs are available to help researchers easily select and implement fractional factorial and orthogonal array designs. These tools can generate appropriate designs based on the number of factors and the number of runs the researcher is willing to make.

Factorial Design Experiments:
Utilize multiple factors and their combinations to analyze their effects on a response variable. Include both low and high levels for each factor, resulting in a balanced design. Commonly used in engineering and scientific experiments.

Calculating Factor Effects:
The average response at a low factor level is compared to the average response at a high factor level. The difference between these averages indicates the effect of that factor on the response variable.

Cuthbert Daniel’s Normal Plot:
A graphical representation of the effects of different factors in a factorial design experiment. Each point on the plot corresponds to a factor, and its position indicates its effect on the response variable. Points that deviate significantly from the central line indicate factors with a significant effect.

Advantages of Daniel’s Normal Plot:
Provides a simple and intuitive visualization of factor effects. Facilitates the identification of significant factors influencing the response variable. Aids in understanding the relative importance of different factors.

Data Analysis for Identifying Important Factors:
The differences between the highs and lows of the data were calculated to identify the effects of each factor. A normal probability plot was used to identify the real effects, which deviated from a straight line. Three factors (holding pressure, booster pressure, and screw speed) were identified as having significant effects.

Experimental Design and Results:
A complete factorial design was used for the three identified factors, resulting in a cube of data points. The average values at each corner of the cube were calculated to create a clear representation of the effects of each factor. The results showed that the three identified factors were the most important, while the other factors had minimal effects.

Practical Applications:
These types of experiments are valuable for screening out unimportant factors, especially in complex experiments. Complex experiments with many variables can be challenging to conduct, especially when physical adjustments are required. The use of paper helicopters is a practical method for teaching these concepts.

Objective:
Illustrating the concept of experimental design using the example of designing paper helicopters.

Experiment Design:
Eight factors identified as potentially influential: wing length, body length, body width, wing width, wing fold, paper clip attachment, etc. Sixteen helicopters created using a recipe based on these factors.

Data Collection:
Experiment conducted by dropping the 16 helicopters and recording their flight times.

Results:
Analysis revealed that wing length and body length were the most significant factors affecting flight time. Longer wings and shorter bodies resulted in longer flight times.

Additional Considerations:
Thickness of the body later identified as another important factor. Encouragement to explore variations and experiment with different factors.

Benefits:
Experimental design provides a practical and engaging way to introduce statistical concepts to engineers. Hands-on approach allows engineers to grasp the principles without the need for prior statistical knowledge.

Challenges:
Some engineers may have negative preconceptions about statistics due to previous unsatisfactory experiences.

Educational Methods:
Traditional teaching methods focus on memorization and information retrieval, which is outdated and ineffective in the modern information era. Modern teaching should focus on teaching broad outlines, problem solving, and information retrieval skills.

Unstructured Problem Solving:
Teaching should emphasize unstructured problem solving, enabling students to apply knowledge to new situations and develop critical thinking skills.

Hands-On Examples:
Starting with hands-on examples helps students understand concepts better and makes learning more engaging.

Design of Experiments:
Design of experiments is a valuable tool for understanding cause-and-effect relationships and making informed decisions. Design of experiments can be used in various fields, including science, engineering, and business. A creative example involving beer, glass types, pouring angles, and foam height demonstrates the practical application of design of experiments.

Resources:
George Box mentions a book and a website as sources for further information on the topics discussed.

Nonlinear Effects and Screening Experiments:
Most factors have nonlinear effects, but many exhibit linear principal effects. Running a few experiments may not yield a magic formula for every problem. Center-of-cube experiments allow for overall testing of nonlinearity. Composite designs can be used to address nonlinearity along specific axes. Measuring in the wrong metric can lead to apparent nonlinear effects.

Optimal Design vs. Orthogonal Designs:
Optimal designs require prior knowledge of factors, regions, and functions, which is often unavailable. Factorial experiments are powerful tools for exploring unknown systems. Factorial designs allow for comparisons and identification of significant differences. Orthogonality in factorial designs facilitates the isolation and interpretation of effects.

Creativity and De-Optimality:
De-optimal designs assume knowledge that is often lacking, leading to potential pitfalls. Factorial designs foster creativity in exploring unknown systems. De-optimality is often overemphasized, and factorial designs offer advantages in understanding complex systems.

Design Considerations for Missing Data:
Missing data can be addressed by considering the region of experimentation. The region of experimentation is often localized around a known working point. Optimal designs may not be suitable when missing data is a concern.

Factorial Experiments as a Teaching Tool:
Factorial experiments can be used to teach basic concepts of experimental design. They provide a hands-on approach to understanding experimental factors and their interactions. Factorial experiments help students develop critical thinking and problem-solving skills.

D-Optimal Designs and Region Sensitivity:
D-optimal designs are sensitive to the region of experimentation. Increasing the size of a factorial design by a small percentage can result in a design with the same D value as a d-optimal design. The application of d-optimal designs is limited due to their sensitivity to the region.

Creative Nature of Factorial Designs:
Factorial designs offer a creative approach to experimentation. They allow for the exploration of interactions between factors. The creative aspect of factorial designs is often overlooked.

Further Information and Resources:
Statistics for Experimenters and Applications of Statistics to Industrial Experimentation are recommended books for further information. Cuthbert Daniel, an engineer and private consultant, came up with many clever ideas in statistics.

Explanation of Plotting Data in Daniel or Normal Plots:
The average difference between the low and high levels of a factor is called the effect due to that factor. Plotting the effects of all factors, including groups of interactions, provides insights into the significance of factors and interactions.

Break-Even Plane and Cost Considerations:
The break-even plane represents the point where two measured responses, such as yield and energy consumption, are balanced. Determining the break-even point helps in optimizing the process without incurring additional costs.

Interaction Example: The Story of the Rabbit:
The story of breeding rabbits illustrates the importance of interactions. The presence or absence of both a doe and a buck in a hutch determines whether rabbits are produced. Interactions are often missed when experiments are conducted with a single factor at a time.

Conclusion:
Dr. George Box’s presentation emphasizes the importance of considering the region of experimentation, the creative nature of factorial designs, and the significance of interactions. He recommends further reading and provides an insightful example to illustrate the concept of interactions.

Osculation:
Osculation refers to the amount of contact between ball bearings and the surface they run on. Changing the osculation can significantly impact the performance of ball bearings. It can be adjusted by modifying the type of ball bearings used or by altering the heat treatment process.

Publications and Media:
The Center for Quality and Productivity (CQPR) at the University of Wisconsin-Madison offers various resources on quality and productivity. These resources include 126 research reports covering topics such as management, experimental design, and control problems. Interested individuals can contact CQPR to obtain a summary of available reports.

Videotape on Helicopter Experiment:
A videotape titled “Designing Industrial Experiments” is available from Scientific Computing Associates. The videotape is a comprehensive six-tape series that covers various aspects of quality and productivity. It includes discussions on basic ideas like the seven tools and progresses to advanced topics like experimental design and robust design.

Courses Offered by CQPR:
CQPR offers courses on quality and productivity taught by experts from the Center for Industry. These courses attract participants from across the country and even abroad.

Course Teachings:
George Box teaches two courses: Designing Industrial Experiment and Engineers’ Key to Quality. A new course, Designing Experiments for Discovery, Improvement, and Robustness, has been added, which goes beyond basic principles. A potential third course on control is being considered, exploring the connection between SPC and engineering automatic control.

Availability of On-Site Courses:
On-site presentations for organizations are possible, but it’s preferred for attendees to come to Madison for convenience. Cost-effectiveness may justify sending an instructor on-site for larger groups.

Conclusion of Telecast:
The telecast on June 12, 1995, with George Box discussing the use of design of experiments for continuous improvement, marks the end of the first half of the 1995 series. Previous months’ topics included TRIZ, teamwork, robust design, and the loss function.

Availability of Videotapes:
Videotapes of the telecasts and the series from the previous year are available for use through National Technological University.

Gratitude and Appreciation:
Jack Revell expresses gratitude to George Box for his eloquent and insightful contributions, recognizing him as a living legend in the field.