Skip to main content

Design of Experiments for Mixtures Using Machine Learning

Design of Experiments for Mixtures Using Machine Learning

Marie Gaudard

Marie Gaudard, Course Author

Senior Data Scientist and Statistical Consultant at Predictum Inc.

Phil Ramsey

Philip Ramsey, Course Author

Senior Data Scientist and Statistical Consultant at Predictum Inc.

About this Course

Mixture systems — in which the components of interest are mixed and their proportions are optimized — involve inherent constraints and complex kinetics, making their behavior difficult to predict.

Traditional methods are not up to the challenge of today’s complex needs. They overburden users, assume inadequate models, and require too many experimental runs.

This course presents a revolutionary approach to the design of experiments for mixtures. It is based on modern machine learning methods that reduce sample size requirements, streamline the process of analyzing and modeling, improve accuracy, and provide deep insight for experiments of high complexity.

Applications pertain to a wide variety of industrial sciences. We have had success in:

  • Pharmaceutical and bio-engineering, conducting media or buffer optimization experiments for increased protein yields from bacteria or mammalian cells
  • Semi-conductors, modeling yield on wafers, identifying yield loss mechanics, and investigating new wafer substrates
  • Asphalt, improving asphalt mixtures for critical process attributes
  • Metals manufacturing, optimizing critical characteristics in metallurgy

There are applications in almost all industries, including chemical, specialty chemical, pharmaceutical, biotechnology, semiconductor, solar, consumer products, and automotive.

This course will help scientists and engineers of many backgrounds predict and optimize system characteristics in a way that was not possible until now.

Who Should Attend

Engineers, scientists, and researchers who work directly with mixtures, mixture designs, and mixture processes in various industries.

Duration

3 days, with ½ day add-on optional

Delivery Methods

  • On-site classroom: 3 full days, with ½ day add on
  • Virtual live: 6 half-day sessions, with ½ day add on
  • Virtual on-demand (coming soon)

Prerequisites

No prior, or formal training in design of experiments is assumed.

Pricing
Individual and corporate pricing available on request

Course PDF
Download here

Instructors

Phil Ramsey
Sr. Data Scientist

Philip Ramsey

Specialist in data analytics, empirical modeling and optimization.

Phil is a Senior Data Scientist and Statistical Consultant at Predictum. He provides consulting services in data science, statistics, and machine learning for integrated analytical systems, custom projects, and training. He specializes in modern experimental design and analysis strategies and the use of statistics, data science, and machine learning in engineering and science.

In addition, Phil is a Professor in the Department of Mathematics and Statistics at the University of New Hampshire (UNH) where he teaches courses at the undergraduate and graduate levels in design of experiments, machine learning, and statistical methods for quality improvement. He has held the following relevant industrial positions: Senior Engineer for Materials and Processes Development, McDonnell Douglas, St. Louis, MO; Staff Scientist/Statistician, Alcoa Technical Center, Pittsburgh, PA; and Statistician/Senior Engineer, Rohm & Haas Electronic Materials (now Dow), Marlboro, MA.

Phil holds a Ph.D. in statistics from Virginia Polytechnic Institute and State University.

Marie Gaudard
Sr. Data Scientist

Marie Gaudard

Specialist in data science, predictive modeling, design of experiments, and machine learning.

Marie is a Senior Data Scientist at Predictum. She specializes in predictive modeling, design of experiments, statistics, and machine learning. She has extensive experience in consulting and statistical training in a wide variety of industries.
Marie is Professor Emerita of Statistics at the University of New Hampshire (UNH), where she has worked extensively with students and companies on the practical application of statistics. She is also a co-author of two books, one of which is about the use of JMP software and statistical methods to improve quality and the other is about the partial least squares technique. She was also a statistical writer for several years as a member of the JMP documentation team at SAS Institute.

Marie holds a Ph.D. in Statistics from the University of Massachusetts at Amherst.

Course Syllabus

Block 1 - (8 hours)

  • About the Modern Mixture Experimentation Course 
  • Factorial versus Mixture Designs 
  • Systems Thinking 
  • Mixture Spaces, Components, Blending
  •  Modeling Mixture Experiments 
    • Case Study: Etch Rate
  • Types of Mixture Designs 
    • Case Study: Yarn 
    • Case Study: Etch Rate (Space-Filling Design) 
  • Exercise: Construct a Space-Filling Design Modern Machine Learning

Block 2 - (8 hours)

  • Modern Machine Learning Example 
  • Self-Validating Ensemble Modeling 
  • Self-Validating Ensemble Modeling (SVEM) Add-In Saving Profiler Scripts 
  • Exercise: Conduct a SVEM Analysis Bounds on Components 
    • Case Study: Flare (Bounds) 
    • Case Study: Flare (Linear Constraints) 
  • Classical vs Modern Mixture Experimentation 
    • Example: Waste Glass Responses, Factors, and Constraints Tables

Block 3 - (8 hours)

  • Mixture of Mixtures Experiments 
    • Case Study: Photoresist Blending 
    • Case Study: Harvey Wallbanger 
    • Exercise: Superfood Drink 
  • Optimization and Multiple Responses 
    • Case Study: Detergent 
    • Case Study: High Alloy Steel 
  • Mixture-Process Factor Experiments 
    • Case Study: Fly Ash 
    • Case Study: Fabric Finish

Additional Block Covering Functional Responses (4 hours)

  • Modeling Functional Responses 
    • Case Study: Fly Ash Curves (Fit Curve) 
  • Functional Responses and Functional Data Explorer (FDE)* 
    • Case Study: Fly Ash Curves (FDE)*

Knowledge you can put to work