Optimization Designs and Response Surface Models

DOE6

Duration : 2.0 day(s)

:: Course Summary

This 2-day course covers the construction of advanced experimental designs specially designed to optimize a process or formulation according to the study objectives and using a minimum number of runs.

:: Learning Objectives

Upon completion of this training, participants will be able to:

Select the levels or categories of a factor to be tested

Construct an optimization design

Analyze and interpret the results

Determine what the optimal conditions are

Construct a response surface model et analyze its results

Make predictions and optimize the process using response surfaces

:: Target Audience

This training is intended for researchers who conduct experiments or studies, who wish to optimize their model, process or formulation and who wish to determine optimal conditions using an appropriate optimization design.

:: Prerequisite

Participants should be familiar with the construction of basic factorial designs and the Analysis of Variance (ANOVA) method, or have followed the following courses or an equivalent:

Fundamental Tools in Statistics

Introduction to the Design of Experiments

Screening Designs

:: Topics Covered

Introduction
- Objectives and Context of Use
- Issues involved in Optimization
- Pertinent Statistical Concepts
- Need for an Alternative to the ANOVA
Construction of Optimization Designs
- Selecting the Factors and Factor Levels to be Tested
- Notion of Central Point and Repetitions
- Properties of Optimization Designs
- Central Composite Designs (CCD) and Box-Behnken Designs
- Use of Blocks
Methodology of Response Surfaces
- Construction of Response Surface Models
- Analysis of Results and Selection of Regression Model
- Use of Models for Prediction and Optimization
Advanced Methods
- Dynamic Experimental Designs: Simplex Method and Steepest Ascent Method
- Optimal Designs
- Incorporating Covariates
Summary

:: Course Content

Finding the optimal conditions of a process is a challenge faced in R&D as well as in industry: developing new products or processes, optimizing quality and performance of a process, etc. Given time and cost constraints, efficient designs are essential to ensure the true optimum is found using a minimum number of runs. Methods where factors are tested one at time can lead to concluding a false optimum as possible interactions between factors are not taken into account.
Several optimization designs exist to answer these needs: response surface models are used to find optimal conditions at the final stage of the optimization of a process and when we wish to develop a model relating influential factors on response variables of interest to quantify performance, quality, yield, etc. Simplex and steepest ascent methods are also used to determine optimal conditions in situations where experiments must be carried one at a time.
This training reviews the principles of constructing optimization designs derived from complete factorial designs. The properties, strengths and weaknesses of Central Composite Designs (CCD) and the Box-Behnken design are outlined.
Response surface methods are discussed next. The modeling of results using linear regression and the use of models for prediction and optimization are explained in detail. The simplex and steepest ascent methods are also reviewed: context of use, underlying principles, assumptions, strengths and weaknesses.
Throughout the training, the emphasis will be put on the application of these methods through examples and case studies to ensure participants understand the construction of optimization designs and how to analyze and interpret the results.