2. What is Taguchi’s Concept? Or Taguchi Methods?
In Dutch: Taguchi Methodieken.
This article is Part 2 of the trilogy of Taguchi Methodieken | Taguchi Methods | Taguchi’s Concept:
Part 1. Who is dr. Genichi Taguchi?
Part 2. What is Taguchi’s concept?
Part 3. When to use Taguchi’s Concept?
So, what is Taguchi’s Concept?
There is a lot of information on the internet about Taguchi Methods. And you can buy good books! But you can be puzzled: What is the goal? Obtain a lot of information with less tests? Is it Design of experiments (DOE)? Is it statistics? Is it a quality philosophy? Or a method to determinate and decrease Quality Costs?
In addition, because of the high quantity of research and production-case-studies, you can be mislead to think it is a quality tool only applicable at production stage. Or you can be afraid of statistical issues in literature.
Lets make one thing clear: Taguchi’s Concept is a powerful tool for engineers in Development and Engineering! To make products reliable or robust. Based upon smart testing!
To make this superb tool more accessible for Engineers, I@E developed a 7-step concept based upon Taguchi’s ideas of optimizing Systems (i.e. [simulations of] products, processes, (sub)assemblies) using DOE. Design of Experiments based upon orthogonal arrays.
I will explain this tool in this chapter.
What is I@E’s Taguchi’s Concept?
It is a Concept to get your system on target. Applicable in (pre-)production but excellent tooling for Development and Engineering. On target and with a deviation on target that is as less as possible. In respect to Quality costs! Dr. Taguchi propagates a tool to calculate the break-even point between quality-deviation-reducing- efforts and costs: the “loss-function”.
Meet Target. Meet specifications or meet requirements; we call it “meet Target”. But there is a difference between Development and Engineering.
Imagine we are darters; than our Engineering-goal is: throw “Always 180” in all different kind of situations. See Fig. 1. Think about temperature conditions (e.g. warm or cold fingers), wearing out (e.g. wether we are 20 or we are 80 years old), or the way we feel (e.g. are we impressed of stress of a lot of audience or are we playing alone)?
Can you make a translation to your system? How do you achieve to make your system more immune for wear-out, tolerance of parts and ambient conditions?
In statistical terms: how can we achieve a system-output like the blue line in Fig. 3? (click figure to enlarge)
But imagine our Darting-process output (your system output?!) is like Fig. 2. That’s according the green line in Fig. 3. Dart no 1. on Target, nr. 2 on zero points and the third in Bulls-eye. A big spread. So the Development-goal is: How can we get on target? Move unto the red line?
If we are on target, than Engineering has a challinging goal: how can we tighten the deviation? (From red into blue). In Taguchi terms: how can we stay robust on target?
In a bullet summary, Taguchi’s Concept helps you
- in Development and Engineering stage,
- optimizing your System and make your System more robust;
- by targetting and decreasing deviation
- In the figure: go directly from green to blue by using S/N ratio. So, do not even think to go from green to red first! See Part 3.
I@E helps you to find an answer to these important development and engineering questions with Taguchi’s Concept. How? With an Engineering Concept that contains 7-steps. See Fig. 12. The emphasis of the Concept lies on parameterdesign to decrease deviation.
What is Parameterdesign? Do you recognize the following situation: The production day-shift leaves the factory. The parameter-setting is according to Fig. 4. The night shift starts with re-arrangement of all the parameter-settings of the proces like Fig. 5. Every shift believes in their optimum situation of parametersetting because of the apparent same Quality F of their System output. This example makes clear that there are more parametersettings possible to obtain an optimum. So it is easy to understand that the optimum you have chosen for your system, is not obvious the best engineering-optimum!
Another example: your are washing your clothes at home and you are thinking about the washing proces: could it be possible to use less detergent, less water, less energy to get the same result? White towels? 10 years ago developers of washing machines were really challenged by this goal. They did parameter-design. The parameters S1 -S5 ( “washing time”, “size of drum”, “rotation velocity”, “degree of filling” and “fall-space”) can be set in such a way that we can use less water (N1) and energy (N2) nowadays. The next step could be “no detergent” (N3)!
What is the best optimum? Please remember that the goal of engineering is to obtain a robust system: the highest output F in combination with the “less possible sensitivity for Noise factors” (detoriation, tolerance, wear-out, ambient factors). Maybe the parametersetting of fig. 5 gives a better optimum?
A better optimum? Suppose we have a very good system-description and we know the relation between the output F and different parametersettings. See Fig. 6. Which Parametersetting would you like to install as default for your system? The day-shifts optimum F1 to the left? Or the lower summit night-shift optimum on the right? You want to have a high output F!
Parameter Design. It was dr. G. Taguchi’s radical insight that the choice-proces of parameter-setting in Development and Engineering can be improved in an extreme way! This Choise-proces is called Parameterdesign. There are different systems of parameterdesign. First of all there is the no-structured trial and error way, the testresults are determined by subjective opinions. The second possibility is to change one-factor at a time (OFAT) to find a better performance of your System. The third possibility is ‘structured Design’. That would be great! In combination with a multidisciplinary team and objectivity of parameterchois and an agile team? That would be greater!
Structured design! Let us examine and compare three kinds of approach in an honest way: we do 4 tests with 3 parameters at two levels and we measure the output F pro test. Structured approaches; but which test-approach would you prefer if you look at the validity of the results and the efficiency of the test: Table 1, 2 or Table 3?
How many tests/testresults can you use to predict the relation between a parameter Sx and the ouput Fx? In table 1 and 2 you can only use 2 tests/testresults. In table 3 you can use 4 tests/testresults.
What happens if one factor has an effect upon another factor? And the reprocibility of the test results? When do you think they are highest?
And assume that we would use a full-factorial test: 2**3 = 8 tests with 8 testresults and that testnr. 7 (in binary code S1= 2, S2 = 2 and S3 = 2) gives the best results. In a full-factorial test we would choose this parametersetting. Which test-matrix could you use best to reach that optimum with 4 tests? 1,2, or 3?
The example above shows you the advantages of orthogonal arrays with the smallest possible array: the L4 (2**3).
But how to deal with 5 parameters like our example in Fig. 4. ? Or think at the ‘washing machine example’. Let us make a comparision again. Table 4 shows the One Factor at a Time Approach (OFAT). If we choose an orthogonal array and we want to test at 2 levels we have to choose an L8 (2**7) array. That means we have to do 8 tests. But again: “what is the validity of the testresults and the efficiency of the tests and the reprocibility or value of your system-knowledge base on a long term?”
Please Compare Table 4 and 5; both the possibility to test 5 parameters at two levels in a structured way at two levels. If you would do all possibilities to test you had to examine 2**7 = 128 testresults. Which testdesign do you like most to find a testoptimum? Keep in mind Fig. 6!
So there are differences between Parameterdesign-methods to optimize your system. In respect to velocity, the quality of testing, the quality of analysis, the quality of synthese of the optimum and the reprocibility and the value of system knowledge. Parameter-design methods are:
- trial and error
- one factor at a time
- full factorial
- fractional design
- orthogonal design
It will be clear that “Orthogonal Designs” forms the heart of ‘I@E’s Taguchi’s 7 steps Concept’. In which body is that heart beating? Let us have a closer look at Fig. 12.
In stage (1) it starts with the forming of a multidisciplinary agile team to insure objectivity. We describe your System, Parameters and F in a Functional Value way. We use internal and external knowledge.
(2) Parameter Choise. Inventarisation, Priority and Significancy are the most important words in this phase. Define upon the integrated knowledge the significant parameters in the system (see Fig.4). “Significant” also means the differentiation between “signal parameter or a noise parameter”. Sometimes this excercise leads to new system-outlines or even other business models.
(3)How to measure your output (F)? Use adequate parameters (2) and Quality Performance Targets (1) to avoid interactions.
(4) Choose an adequate orthogonal testdesign primary based on ‘less excecuting time’ (TTM!) and secondly on high knowledge base! After excecuting the test you use S/N ratio to deal with complex but still accessible optimization. “S/N” is no statistically stuff but an engineering-hammer-tool!
(5) Robust Performance Design is focused upon optimum-synthesis instead of analysis. E.g. if you have 4 parameters and you want to test them at three levels, you have 3 ** 4 = 81 possibilities to test in a full-factorial way. An orthogonal array of 9 tests L9 (3**4) enables you to get faster insight in parameter-relation to output and synthesis of an optimum.
(6) The next step is testing your ‘optimum’. Analysis of results ánd accordation with expectation in system-desription and parameter-brainstorm (2) will lead to enhancement of
(7) your knowledge database. The engineer or agile group/team knows this proces on forehand. All steps are known, the proces of analysis ánd synthesis of the results is known and the effect of choosing an optimum is knowing. This makes the generated knowledge to your Knowledge Database (7) about parameters and system performance very sustainable!
Structured Engineering! You want to learn or to implement it in your company?You want to have more information about the application? Then read Part 3: When to use Taguchi Methods.
