Robust Quality
Taguchi uses the QLF in developing the idea of robust quality (Naylor, 1996). The robustness of a product refers not only to its ability to conform to specifications but also to its performance and reliability above and beyond the stated limitations. Dr. Taguchi focuses on examining the robustness of a product under varying conditions in the environment, component parts and manufacturing processes. Engineers designing products for robustness under environmental variations must understand the environment in which the product is to be used. Designers thus attempt to overcome the effects of degradation, use and misuse of a product by building in extra strength at key points. It is also desirable to make products insensitive to variations in their own components and so design engineers seek ways to minimise the effect of inevitable component variation on the assembly. Taguchi also claims that well designed experiments can effectively reduce variability in manufacture and so cut down on waste and scrap.
Stages in the design process
Taguchi et al (1989) define three stages in the design process:

  • System design
  • Parameter design
  • Tolerance design
  • At the systems design stage, the overall form of the system is identified. This stage of the design process requires close co-ordination between marketing, planning and systems engineering. Attribute values are specified at the Parameter design stage and it is very important to ensure robust design at this stage of the process. The allowable ranges of deviations in parameter values are defined in the tolerance design phase. Both parameter and tolerance design require close co-ordination between design engineering and manufacturing. These three stages in the design process form the basis of Taguchiís paradigm for achieving a robust product and process design. Taguchiís methodology suggests that, other things being equal, products developed according to this paradigm will result in the least loss of quality due to variance from specification and also meet cost constraints .

    Noise Taguchi describes variation in a productís design parameters and in the external environment as noise. Noise factors are classified into three groups for example;

    NoiseDescription of Noise
    External NoiseVariation in external conditions e.g. humidity or supply voltages
    Internal NoiseDeterioration in parts and materials
    Between Product NoiseVariations in products built to the same specifications, caused by differences in materials and manufacturing processes

    Dr. Taguchi proposes two techniques used in the selection of design parameters, which minimise the sensitivity of the design to noise.

  • Signal to Noise ratios (S/N) for parameters of the design.
  • The use of experiments to explore a range of parameter combinations to inform design decisions.
  • In design, S/N ratios are used to measure the effect of noise factors on performance characteristics. S/N ratios take into account both amount of variability in the response data and closeness of the average response to target. The S/N ratios with greatest application are those for;

  • Smaller is better
  • Larger is better
  • Target value is best.
  • In cases where there are many interacting design parameters, it is necessary to carry out experiments to explore the effect of combinations of parameters. Carrying out a full set of experiments for a given set of variables is realistic only in experiments where small numbers of parameters are involved.

    Taguchi, however, has developed a technique to obtain valuable information for design using a limited number of tests. In this technique, tables known as orthogonal arrays are combined, where there are interactions between variables, with charts known as linear graphs. Orthogonal arrays are tables in which the columns show the parameters to be tested and the rows show combinations of parameter values to be used in experimental trials. The parameters are often referred to as factors, and the values that the parameters can take are known as levels. The tables are called orthogonal because the levels of each factor are equally represented in the trials (Browne et al, 1998).

    For example, the strength of a part is a function of three factors x, y and z. To evaluate the strength of the part for two different values (1 and 2) of each of the three factors, four trials are performed.

    A Lumped Parameter model is used when the design factors are independent of each other. In experiments where dependent factors are involved, linear graphs are used to show which factors relate to each other to form new factors.

    Copyright 2008, All rights reserved.
    Designed and Hosted by
    Mirage Solutions