When I first met Dr. Genichi Taguchi in 1991, we discussed how I was developing the new designs I was working on. He then proclaimed - "You focus on the wrong thing... you are not thinking about the energy, you must follow the energy if you want quality! You cannot get quality if you focus on measuring quality; reliability does not come from counting failures!"

Well, that was only the first of many dressings-down I was to receive at his hand - every one of which I absolutely deserved! He rebuilt my thinking process over the two years he was at Kodak, along with Doctors Don Clausing of Quality Function Deployment (QFD) fame, and Stewart Pugh, the inventor of the Pugh Concept Selection Process. These three mentors got hold of me and worked on me until I had a "renewed mind" in the context of engineering thinking about developing new products. Clausing taught me to focus on risk-ranked Functional Requirements that I now call NUD - the New, Unique and Difficult. Pugh taught me to generate numerous concepts that were high in feasibility and low in competitive vulnerability resulting in a hybrid concept that was truly superior. Taguchi taught me to develop this superior "system" by fully understanding its Ideal Function prior to robustness optimization.

What is an Ideal Function?

An Ideal Function is the predicted behavior of a design that is purely due to the desired physics-based (Y as a function of controllable Xs) model that underwrites its ability to work as stated in the final concept description under nominal (noise-less) conditions. Ideal refers to the behavior in the absence of sources of unwanted variation. Function refers to what the design does using the physics-based definition of consuming and transforming energy and mass to produce work that fulfills the design's functional requirements. Here we are in the world of continuous variables that are defined as scalars and vectors. To a statistician the Ideal Function is described by a design's main effects, its interactive factors as well as any non-linear terms and coefficients that mathematically describe it behavior with 95% confidence. When the physics-based description of an Ideal Function are merged with the statisticians model - well, we really have something quite useful as we move on to further develop a "robust" design. But first things first....

Functions and Risk Levels in CPM

CPM focuses on the Functions that a technology, product or manufacturing process must perform to fulfill its requirements. The following diagram (Fig. 1) describes a Function through the parametric variables that enable it to occur based upon physical law.

Fig. 1: The Ideal Function

A Function is described by one verb and one noun that together state the most fundamental action the design must accomplish. Also included are the sources of unwanted variation that can disrupt the function. These are what Taguchi called "noise". The function in a robust design has been made insensitive to noise. Taguchi cautioned that an engineer needed to fully define the Ideal Function of a design before there was any hope of developing robustness once stressful sources of noise were intentionally forced upon the design during the later stages of development and certainly prior to system integration. So the early stages of detailed sub-level design are to be all about converting the concept design into a parametric design that is fully described by the Xs and Ys associated with its intended function(s), including the constraints imposed upon the Ideal Design. Figure 2 illustrates what goes into and what comes out of an Ideal Function.
 

Fig. 2: Inputs and Outputs of Ideal Function

The Function Tree

All the ideal Functions in a design are then listed in a Function Tree, illustrated by Figure 3.

Fig. 3: The Function Tree

Once a Function Tree is fully illustrated and the NUD Functions highlighted, the team constructs a detailed Functional Flow Diagram - listing the serial-parallel flow of the functions over time. Function Trees are the basis for construction of a Functional Flow Diagram (see Figure 4).

Fig. 4: Function Tree is Basis for Functional Flow Diagram

The Functional Flow Diagram and the Function Tree serve two useful purposes:

1. They are essential to constructing a coherent Design Guide so people understand how the design works, and
2. They set the stage for generating base-line Parameter Diagrams for Modeling and Simulation and Designed Experimentation (DOEs).

Stratifying Risk in the Context of Functions and Complexity

To manage risk at a fundamental level, one must look at the differences in functions from simple to complex configurations of a design as it becomes a complete system. (Note: ECO= "Easy, Common and Old")

Fig. 5: NUD vs. ECO Development Risk

The diagram above (Fig. 5) illustrates a rule-of-thumb we use in CPM: Risk increases as one moves from:

1. static parts (items that do not move much on a micro-deflection basis or move very little relative to other parts in a macro-displacement context; no or extremely small mass-energy transfer or transformation, negligible "work" is measurable at this level of a product)
2. to parts and materials that are integrated into functioning sub-assemblies that have some form of internal dynamics (some form of mass-energy transfer or transformation we can classify and measure as "work")
3. to integrated parts and sub-assemblies into functioning sub-systems that have additional, more complex forms of internal and across sub-assembly dynamics (more complex forms of mass-energy transfer or transformation we can classify and measure as "work")
4. to integrated parts, sub-assemblies and sub-systems that form a functioning system that has very high complexity within and across the functioning system. (the most complex forms of mass-energy transfer or transformation we can classify and measure as "work")

Critical parameter development identifies a hierarchy of functions and associated risk due to NUD designations. Once NUD Functions are identified, Design Failure Modes & Effects Analysis (DFMEA) is conducted.

Functions and the risk they carry during development are controlled by five Design Controls as part of the risk mitigation of DFMEA:

1. Computer-Aided Engineering Analysis (CAE)
2. Computer-Aided Design (CAD)
3. Basic Characterization of Y-X Relationships (through Regression Analysis)
4. Designed Experiments (DOEs)
5. Verification & Validation Tests (replicated evaluations against requirements)

The Program Manager must design the appropriate design controls as tasks into the schedule and account for their costs if the project's performance goals are to be met. Functions must be associated with performance in this context - not quality. Quality happens because Functions are measurable, stable, adjustable, low in sensitivity, robust, and capable of meeting Cp & Cpk goals as part of the product's functional requirements. Notice I have closed the loop back to requirements! Quality happens because functions are mastered! This was essentially what Taguchi was trying to teach me. I hand the baton to you....

Once the scalar and vector, parametric variables that govern the physics of a design are well-defined and understood, we can move on to face the "dark side of the force" - the Law of Unintended Consequences. It too obeys the Laws of Physics and does so precisely according to the hidden part of your design's architecture that was not included in the Ideal Function!

Next month's newsletter will address some methods we can employ to minimize the likelihood we will be blind-sided by the Law of Unintended Consequences. This will include Design for Additivity on identifying and learning about the noises that can disrupt our Ideal Functions. See you then! 

Product Development Systems & Solutions Inc.

See the Archive of PDSS Newsletters