Dr. Montgomery is a well known authority in the field of Industrial Statistics with a special emphasis on Design of Experiments. He is the author of the best selling book Design and Analysis of Experiments, 5th Edition (2000, John Wiley and Sons, Inc.). Dr. Montgomery has been working closely with S-Matrix for the past two years on our development program for Fusion Pro. He had substantial input into the product specification and continues to guide the development of both the expert-system overlay and the statistical routines underlying the software.

Software Validation Statement by Douglas C. Montgomery

"I believe that Fusion Pro represents state-of-the-art in Design of Experiments software. In addition, the Technical Data Mining capabilities are excellently integrated with D.o.E., and represent sound statistical thinking in both the feature content and tailoring of the underlying statistical engines. Some of the major elements of the Software Validation Program I have undertaken with S-Matrix on behalf of Fusion Pro include:"

1. Model-Robust Designs
2. Experiment Planner Wizard
3. Custom Model Designs
4. Historical Data Mining and Matrix Master
5. Repair Design Generation
6. Design Blocking
7. Response Data Transformations
8. Trellis Graphics Using Conditioning Variables
9. Multiple Response Optimization

Dr. Cornell is considered the foremost authority on mixture experiment design. He is the author of the best selling book Experiments With Mixtures, 2nd Edition (1990, John Wiley and Sons, Inc.).

Software Validation Program

Dr. Cornell is working closely with S-Matrix on development of the following mixture design and analysis capabilities for Fusion Pro. The specific capabilities that he has guided and will be validating are listed below. Validation of these capabilities should be completed this summer.

1. Unconstrained and Constrained Mixture Designs
2. Unconstrained and Constrained Combined Mixture-Process Designs
3. Custom Model Combined Mixture-Process Designs
4. Mixture-Process Designs With Multicomponent Constraints
5. Analysis of Mixture-Process Designs

Fusion Pro Pfizer Validation Summary.pdf	Right click on the link at left to download the Fusion Pro Pfizer Validation Summary file. (IE users select the 'Save Target As...' option) (Netscape users select the 'Save Link As...' option).
	Left click on the image at left to get Acrobat Reader.

Pfizer Global R&D has validated Fusion Pro using its Software Development Methodology (SDM) guidelines. Both Fusion One, the "lite" version, and Fusion Pro, the professional edition, have successfully completed the validation program.

The Pfizer SDM, which meets all FDA regulations, covers every aspect of software technology and operation. As an example over 180 software test scripts, each with independent data and expected results obtained from a peer-reviewed journal or a published textbook, have been developed. Pfizer's SDM program also covers supplier validation, including product design control, version control, testing control, release control, disaster recovery control, and Y2K compliance.

Pfizer continues to support further development of Fusion Pro. New technology elements include:

Item 1: Normal Probability Plots of Experiment Variable Effects - Delivered in Q4, 2000

Fusion Pro currently provides mean effects estimation results in the form of tabulated statistics with appropriate hypothesis tests and confidence intervals. Interpretation of these statistics is facilitated by S-Matrix’s PlainTalk Report technology and Variable Effects ranking tables and charts. However, many DOE texts and some software also provide normal probability plotting to aid in identifying and visualizing important effector variables.

This capability enables the experimenter to easily generate normal probability plots of variable effects. The qualitative specification for this capability is based on the following publication:

Montgomery, Douglas C., (1996), Design and Analysis of Experiments, 4th Edition, John Wiley and Sons, New York (specifically pages 318-336).

Item 2: Split-plot Experiment Design and Analysis - Delivered in April, 2001

Current DOE software does not correctly estimate the experimental error in multifactor, blocked experiments in which there is a restriction on randomization within the blocks. This in turn can lead to incorrect hypothesis testing of variable effects.

This capability enables the experimenter to generate experiment designs that correctly address additional restrictions on within-block randomization. The capability extends to correct experimental error and variable-effects estimation. The qualitative specification for this capability is based on the following publications:

1. Montgomery, Douglas C., (1996), Design and Analysis of Experiments, 4th Edition, John Wiley and Sons, New York (specifically pages 521-529).

2. Cornell, John A., (1990), Experiments With Mixtures, 2nd Edition, John Wiley and Sons, New York (specifically pages 365-377).

Item 3: Optimization Overlay Graphics - Delivered in April, 2001

Fusion Pro has a Numerical Optimizer that can identify the optimum settings of the study variables given (1) a goal for each response, (2) a relative importance ranking for each response, and (3) a “desirability profile” that defines how quickly the answer becomes undesirable as it moves away from the goal. Currently the software does not display the “region of overlapping optimum conditions” graphically.

This capability enables the experimenter to create overlay contour graphs that display the multiple response optimization results. This enables the user to visualize the “region of multiple optimum results” and the robustness of the optimum conditions to variation in the study factors. The qualitative specification for this capability is based on the following publication:

1. Montgomery, Douglas C., (1996), Design and Analysis of Experiments, 4th Edition, John Wiley and Sons, New York (specifically pages 596-599).

Item 4: Dispersion Effects Estimation in Factorial Designs - Delivered in November, 2001

Most DOE analyses focus on isolating and quantifying experiment variable mean effects. Software programs use regression or ANOVA analysis, coupled with the appropriate hypothesis tests, to identify important variables in terms of their mean effect on the response. However, these analyses do not estimate the effects of the variable on product stability or quality variation. These effects are called dispersion effects. A variable identified as not having a significant mean effect from a standard regression or ANOVA analysis may still be an important effector in terms of dispersion effects. Without this information, the experimenter may prematurely dismiss a variable that has important effects on product stability or quality.

This capability will enable the experimenter to estimate the effects of experiment variables on product stability and quality variation in replicated and unreplicated two-level factorial designs. The qualitative specification for this capability is based on the following publications:

1. Box, George E. P., and Meyer, Daniel R. (1986), "Dispersion Effects From Fractional Designs," Technometrics, 28, 19-27.

2. Nair, Vijayan N., and Pregibon, Daryl. (1988), "Analyzing Dispersion Effects From Factorial Experiments," Technometrics, 30, 247-257.

Item 5: Sequential Optimization by Method of Steepest Ascent - Delivered in November, 2001

Fusion Pro has a "next experiment" strategy for every design it creates. It can also construct an efficient "next experiment" from historical data. However, The user is currently required to extrapolate the results of a given study to the selection of variables and ranges for the "next" experiment. Although the relevant data and tabulated results required for such an extrapolation are already contained in the software, the extrapolation often requires a DOE subject matter knowledge beyond that of the general user.

This capability will enable the experimenter to obtain a suggested next experiment based on (1) the estimates of variable effects, (2) the multiple optimization results, and (3) the desirability ranking and relative importance weighting of the analyzed responses. The qualitative specification for this capability is based on the following publication:

Montgomery, Douglas C., (1996), Design and Analysis of Experiments, 4th Edition, John Wiley and Sons, New York (specifically pages 578-584).