Nutek, Inc.     Quality Engineering Seminar, Software, and Consulting ( Since 1987)                          Site: Examples

Taguchi
Robust Designs

Taguchi DOE
Software

Application
Consulting

Project  Mngmt. &
Quality Trainings

Company Profile &
Clients List

Home & Contact
Information


DOE Application Case Studies (Examples)
Review example case studies online, examine application steps, learn how to plan experiments, and download complete case study reports.

Case Study 1: Vane Pump Cylinder Block Process Parameter Study
Case Study 2: Electrostatic Powder Coating Process Optimization
Case Study Summaries (Brief Descriptions, text only)
Sample Case Study Report (Display csex-105.pdf file):  Clutch Plate Rust Inhibition Process Optimization Study
Sample Case Study Report (Display csex-101.pdf file):   Die Casting Process Parameter Study
Review and print Sample Problem Solving Project Reports: Report-I Report-II

About the Case Studies
The DOE application case studies included above are all actual experiments conducted by industrial practitioners. Because of proprietary nature of the projects, the names of owner companies and the specific objectives have been purposely altered. Each experiment adhered to the application steps and attempted to satisfy the purpose of the studies. The description of the case studies listed above, include project specifics in each of the application steps outlined here. (All experiments were designed and analyzed using Qualitek-4 software)

If you are starting out with your first DOE application, read the case studies to see how it is done typically, but stick to the following guidelines and seek out answers to questions posed in each of the steps. The guidelines presented below are intended for readers who have either completed a course (seminar) on DOE/Taguchi Approach or otherwise familiar with the technique.

Application Steps

I. PLANNING - This is the most important step in the application process. For better results, planning must be done with the project team and all things must be decided by consensus. The project objectives, how the objectives are evaluated, factors involved in the project, etc. are determined in this session. (See below for a detailed agenda of discussions)

II. DESIGNING EXPERIMENTS - Based on the factors and levels identified in the planning session, the experiment is laid out for the project. In the Taguchi approach, most experiments can be designed using the Orthogonal arrays (L-4 through L-64) prescribed Dr. Genechi Taguchi. Some situations with mixed level factors can be designed by minor modification of the standard Orthogonal array.

III. RUNNING EXPERIMENTS - The designed experiments are carried out as per the recipe prescribed by the Orthogonal array. When the project objectives includes seeking ROBUST design, experiments must be carried out by exposing them to the conditions of the Noise factors as specified by the Outer array, In all cases, a random order is preferred for carrying out the experiments.
When experiments are carried out, the performance must be measured (results) as per the established objective (s) and by following the method of evaluation agreed upon in the planning session. When there are multiple criteria of evaluations, measured data under each evaluation for each tested sample must be carefully recorded in the data collection form created for the project.

IV. ANALYSIS OF RESULTS - The first thing to do before analyzing results is to decide what to do about multiple criteria of evaluation. The thing to remember is that, as far as DOE analysis is concerned, you are only allowed one numeric data for each sample. So, in case there is more than one objective, you will have the option to analyze each criteria at a time, or combine them all into one single Overall Evaluation criteria (OEC). You need not worry about this task if you are after only one thing and there is only one type of result you recorded.
The results of the experiments are analyzed to information in the following types.

  • Main Effect: Provides trend of influence of the factors to the performance measured by the results being analyzed. Calculations in this area can easily produce information about PRESENCE OF INTERACTION between any or all possible pairs of 2-Level factors whether they were considered as part of the study or not. Average factor effects calculated here, allows determination of the Optimum condition and the expected performance at the optimum condition.
  • ANOVA: Yields information about the influences of factors and interactions to the variation of the results. ANOVA also allows scrutiny of the factors and interactions in terms of their relative significance and helps determination of the Confidence Interval (C.I.) of the expected performance at the optimum condition and the factor influences.
  • Optimum Condition: From the factor averages calculated as part of Main Effect study, the optimum design condition can be determined by considering the Quality Characteristic. Generally a linear model is used to estimate the performance that can be expected at the optimum condition. The Confidence Interval on the expected performance which now be calculated, establishes the limits within which the improved performance is expected to fall. For confirmation, a set of samples are tested at the optimum condition and their average performance compared with the expected performance and the C.I.
  • Conclusions: This should describe summary findings and potential benefits by implementing the improved design condition. Specifically, conclusion should include: (a) What are important factors, in order of their significance, (b) How do these factors behave, (c) Which interactions are strongly present, (d) Which factors are insignificant (remove tolerances), (e) What is the best design condition and how much improvement can be expected, (f) What is the performance trade-off when the desired factor level is set, (g) What is the $ savings that can be expected from the new design, (h) How much variation is reduced and what is the improvement in terms of increased Cpk, etc.

VI. CONFIRMATION OF IMPROVEMENT - This includes running a number of samples at the optimum condition and comparing the average result with the expected performance. If the mean of the sample performance falls within the C.I. on the expected performance, Running confirmation experiment is a must even if you have a strong background in DOE and have a high degree of confidence in the conclusions. It is a necessary verification exercise that enhances your confidence in the technique, and may even help you earn some credibility with your team members.

  I. PROCEDURE FOR EXPERIMENT PLANNING (Brainstorming)

Experimental designs produce most benefits when they are planned carefully. For proper planning it is necessary that a special session is dedicated to discuss various aspects of the project with the project team and that all decisions are made by the group consensus. The project leader should arrange for the planning session, and when possible, have someone who is not involved in the project, facilitate the session. Following are general guidelines for contents and nature of information sought.

a) Project Objectives ( 2 - 4 hours)

  • What are we after? How many objectives do we wish to satisfy?
  • How do we measure the objectives?
  • What are the criteria of evaluation and their quality characteristic?
  • When there are more than one criterion, would we have a need to combine them?
  • How are the different evaluation criteria weighted?
  • What is the quality characteristic for the Overall Evaluation Criteria (OEC)?

b) Factors (same as Variables, Parameters, or Input, 1 - 2 hours)

  • What are all the possible factors?
  • Which ones are more important than others (Pareto diagram)?
  • How many factors can we include in the study?

c) Levels of the Factors ( 1/2 hours)

  • How are the levels for the factors selected? How many levels?
  • What is the trade off between levels and factors?

d) Interactions (between two 2-level factors, 1/2 hours))

  • Which are the factors most likely to interact?
  • How many interactions can be included?
  • Should we include an interaction or an additional factor?
  • Can we afford to study the interactions?

e) Noise Factors and Robust Design Strategy ( 1/2 - 1 hour)

  • What factors are likely to influence the objective function, but are not controllable?
  • How can the product under study be made insensitive to the noise factors?
  • What are the uncontrollable or Noise factors?
  • Is it possible to conduct experiments by exposing them to the simulated Noise conditions?

f) Experiment and Analysis Tasks Distribution (1/2 hours)

  • What steps are to be followed in combining all the quality criteria into an OEC?
  • What to do with the factors not included in the study?
  • How to simulate the experiments to represent the customer/field applications?
  • How many repetitions and in what order will the experiments be run?
  • Who will do what and when? Who will analyze the data?

(Details of other four steps in applications tend to be specific to the project and can be found in the individual case studies presented earlier)

References