This article discusses the requirements, deliverables and tasks for the final 3 phases of the IMMSAT process. The first 3 phases, Identify, Model and Measure, were discussed in last month's newsletter (May 2009).
To re-iterate the introduction in the previous article, the IMMSAT Research Process is built on the historic foundations of the scientific method. When this approach is taught in a formal class we call it Research Excellence or REx. The six phases of this process are designed to make it difficult to produce bad science. Each phase builds confidence, develops knowledge and ultimately produces irrefutable empirical data that the science is properly modeled, measurable, stable and adjustable across a bounded range of functional performance.
The six phases of the Research Process are:
- Identify
- Model
- Measure
- Stabilize
- Adjust
- Transfer
Note: TMBP stands for "tools, methods and best practices".
(Editor's note: Phases 1 through 3 were discussed in the May 2009 issue. If you missed it, reply to this email requesting it and I will send it. -Carol)
Phase 4: Stabilize
The fourth phase is called Stabilize. After we can trust the data coming from our measurement systems, we can assess sample data sets from empirical tests performed on proto-type models that are physical expressions of our analytical models. We can then analyze the sample data in the form of descriptive and graphical statistics, tests of comparison and Analysis of Variance (ANOVA) based upon various forms of designed experiments. This is the phase of detailed hypothesis formation, evaluation and acceptance or rejection of each hypothesis. In fact, a system of hypotheses can be established and evaluated during this phase. This is the beginning of critical parameter management for a research team.
The key deliverable is empirical data that proves that the phenomena are statistically repeatable when we make the physical embodiment of our model do what it is intended to do. We can "turn it on", "turn it off" and do it again and again and document the stability of cause and effect relationships. This is fertile ground for the application of Statistical Process Control (SPC) methods. We can make a compelling argument about stability using a simple Individuals and Moving Range chart! We can quantify central tendency and the spread of sample data around central tendency with statistical accuracy and precision. Distributions can be documented as stable, identifiable and able to be categorized under nominal laboratory conditions. If data are not normally distributed - fine, we will use non-parametric statistical characterization to document a median-based distribution and its range characteristics. Confidence in the statistical significance of the models and their terms can be set at any level desired - 95% confidence is typically the standard. This is not so far from what a design engineer would do during the early phase of designing the nominal performance of an element of a product. We are designing an element of science as opposed to an element of a product. The problem DFSS instructors run into when trying to relate to scientists is they fail to change their process context and vocabulary to focus on just those tools and methods that a scientist will value in the design of a new element of science! It makes a big difference.
The Stabilize Gate requirements include:
-
Comparison of empirical models to analytical models.
-
Empirical data and models that statistically prove, with specified confidence, the stability of the phenomena associated with the system of hypotheses underwritten by the analytical models.
The Stabilize Gate deliverables include:
The Stabilize Phase tasks that will produce the deliverables include:
-
Conduct empirical data production in the form of designed experimentation.
-
Construct empirical math models including all fitted model terms: coefficients, higher order terms, interactions and residuals (lack of fit and pure error).
Phase 5: Adjust
The fifth phase is called Adjust. Now that we have stable "science" that is coming from capable and trusted measurement systems, we can introduce change to the stable cause and effect relationships. This phase is all about intentional change under nominal lab conditions and in some induced stress cases. It is here we learn if our stable science is able to be moved around in the embryonic form of "design space". Are there "tuning parameters" that can be identified and used to repeatably change the central tendency and variation around the central tendency of our new science? Does the distribution change when we change its candidate governing parameters? This is useful variation and not considered a form of noise. This is a fore-runner to the ability to do a capability study where one desires to place a mean on a target after reducing sensitivity to sources of variability under both nominal and stressful conditions. We can't do a capability study in IMMSAT because we only have signal and noise and no customer requirement from a market or segment yet! We only have stability requirements and adjustability requirements. Is the phenomena stable and can I adjust it at will under nominal lab conditions? I don't know what a target of customer range is yet - I just want to characterize the dynamics of this science.
The Adjust Gate requirements include:
-
Define the range of functionality of the analytical and empirical models.
-
When appropriate, define functional performance profiles under stress cases.
The Adjust Gate deliverables include:
The Adjust Phase tasks that will produce the deliverables include:
-
Conduct nominal and stress case empirical investigations into the dynamic ranges of the phenomena.
-
Empirically model the terms that control and adjust the performance ranges that bound the new science.
Phase 6: Transfer
The sixth phase is called Transfer. There is a great deal of documentation, physical hardware and equipment transfer of the new science and its attendant measurement systems that deserves a phase unto itself. It is here where the hand-off from people who create science to people who develop technology occurs.
The Transfer Gate requirements include:
- All models must be fully documented.
- All measurement systems must be capable and documented.
- All technical reports and critical parameter databases must be documented.
The Transfer Gate deliverables include:
- Documents: All models, technical summaries, risk analyses, critical parameter databases.
- Hardware, software and firmware: measurement systems, prototypes, files, etc.
The Transfer Gate tasks that will produce the deliverables include:
- Write reports and technical summaries, etc.
- Package and deliver hardware, software and firmware.
