Customer Reviews

World Class Quality: Using Design of Experiments to Make It Happen

5 star:		(3)
4 star:		(11)
3 star:		(2)
2 star:		(4)
1 star:		(2)

 

 

I was somewhat confused when reading this book couple years ago to learn about Dorian Shainin approach. Fortunately, subsequent to reading this book, I read the following two much better books on Shainin's ideas:

Mnufacutring Solutions for Consistent Quality and Reliability by Robert Traver, 1995, amacom

Statistical Engineering by Stefan Steiner and Jock MacKay, 2005, ASQ Press

Excellent material, marred by an attitude towards other quality gurus. The author does a good job of presenting sound and fast methodologies in product quality. These techniques can be of value to every professional engaged in manufacturing and quality fields. A must read, but ignore the pungent remarks on other methods.

Taguchi techniques also work well, I have used them often.

I have used Dorian Shainin's statistical engineering techniques for years and have solved several tough problems in manufacturing environments. I have found the techniques to be quite capable of helping me solve probably 99% of the problems I have faced in manufacturing. Bhote's second edition is a welcome update from his sparse first attempt. This book shows techniques, examples and how they work and don't work. More and newer tools are shown in this edition also. I found it a little difficult to read over and over again Keki's slamming of Taguchi and classical DOE. Shainin has used these tools to develop his own tools, and while they aren't used in his strategies by name they should at least be given polite acknowlegement by the author. I often felt as if Keki was begging the reader to use Shainin's techniques instead of simply writing about them. Overall I appreciate the compilation of almost all of Shainin's techniques even though so many have service marks on them for protection. This is a wonderful reference book and I have found it very useful to use when I explain the tools to others. I did have a hard time trying to merge the message of six sigma with Shainin's statistical engineering. I couldn't figure out if Keki thought six sigma was right or wrong, but he then went on to promote his own new version of everything, the Big Q. A little too much selling.

First, let's get the negative stuff out of the way. The first 1/4 of the book contains a lot of comparisons between the author's methods and everyone else's. Not surprisingly, his consistently rates highest. Included are tables with numerical ratings of the different experimental design strategies but no explanation is given for where the numbers come from. Skip it.
The meat of the book is its exposition of Dorian Shainin's DOE methodology. I am not claiming to have made an exhaustive survey of the literature on this subject, but he is the only one I have read to offer a workable strategy for sorting through a large number of variables in order to arrive at those that truly control the response of interest without the confounding of main and interaction effects that results from the use of fractional factorials, Taguchi orthogonal arrays and other methods. I have just four more words on this book: try it, it works.

Bothe's tools can be useful when something already went wrong and you need a quick fix. So multivari, paired comparision, component search should be in your toolbox of problem solving.

Why do things go wrong?

Because the products or their associated processes are not designed well enough. If they had, how could a problem occur? They are too sensitive to variability, which engineers can not control (like different users, wear and aging etc.: noise). A careful look at Bothe's case studies show: parameters studied and finally changed are often - noise factors.

There is a problem with this. The leakage of a capacitor was identified in the Microcomputer case study (delay time). The supplier of this device was contacted and "(...) leakage was reduced at no extra cost for the company (...)".

That is good. But what happens, when something changes at the supplier? When policy changes? Or raw materials are changed to reduce cost? Or moral changes among its employees? Etc. All of a sudden the delay-time problem can pop up again. So, what is the achievement?

There is an other option: make the microcomputer less sensitive against leakage. Even, if the underlying root cause should change over time (degree of leakage), it can not cause harm.

From a quality point of view you should never be forced to search for trouble-makers. If so, it is already too late. A "Red X" like "no coolant in the machine" should never happen. How could it ever happen? It is a finger-print of insufficient design.

In such a situation it is very easy to achieve a 10:1 or 50:1 improvement: the situation allowed already a worsening of 1:10 or 1:50 ... while it should have been stable.

In my view, time pressure, lack of required specific knowledge, lack of phantasy to figure out available disaster and similar lead to insufficient design (think of accidents). This is a serious problem in product and process design.

This is why I prefer Taguchi's approach of making products insensitive against noise in the research and design phase: find that combination of design parameters, which will minimize the impact of noise factors (usage, aging, information etc.) during its lifetime. No sensitivity, no complaints.

Bothe may be right: if you use the Taguchi approach while lacking specific knowledge on the product, its technology, its later usage etc. you will have a problem. Garbage in, garbage out.

So, how to support engineers in finding the required specific knowledge? How to sharpen their thinking in a short period of time?

There are many ways to do it.

E.g. you can start at the symptom and derive clues by Bothe. This will deepen your understanding of the underlying processes. You can use these findings to find relevant noise factors quickly. How to become robust against these noises without controlling them? You could vary design paramters and subject each design variant to the identified (strong) noises. Evaluate the functionality of your product and select the least sensitive one. Oops, this is almost the Taguchi approach.

Does the final result match your expectations? If so, why? If not, what is wrong in your perception of this specific technology? The product now is less sensitive, isn't it? Why?

In my view we need knowledge creating tools, not counters ;-) to create excellent designs at low total cost, fast....

As a process engineer I had spent a year with problem. There was nobody to solve it (in the multinational company). Together with my collegue, with two experiments (Component Search and Paired Comparision) we found the root cause and proposed the corrective actions - modification on tool. After implementation of our proposal the problem has gone forever!

Despite the fact that there are many mistakes in text, graphs, tables, it's the best book about Shainin DoE.

I'm not an academic in statistics and DoE but an end user to this advanced quality tools so I don't really care that much about the academic part of the theory so long as the methods help me in my work. What I find most interesting of Shainin's method is its practicality and ease of use. Keki has indeed done a great job in explaining the methods from his own industrial experiences without flooding the reader with too much theory. He used real life industrial application examples to illustrate his points which is something very useful to the practitioner like me. Although there're some typo errors in this book and could be frustrating but it makes you really read and understand the principle behind each method more carefully and that's good. After reading this book, I find an urge to share this new practical knowledge with my collegues because it is simply easy to use and yet powerful.
By reading this book and practising it in your work, it will becomes an assest to you. I'm not denying the contribution of the traditional DoE and Taguchi method because I felt that all methods are good as long as it is easy to use and at the end of the day help us to solve the problem.

The Shainan approach is quite practical for industrial use. As many other reviewers have written, the constant slamming of other approaches detracts from the book quality. In the final analysis, the book overlooks that aliaising that will still occur, unless there are 4 or fewer factors, even after the screening methodologies have been used. But they should be small if the screening has really identified the various shades of X's

The major problem with the book is that it is riddled with typos. Some of the typos are worse than in the previous addition. Consequently, one has to read with great care and check every number! Quite frustrating when using it to teach the Shainan approach!

The methods highlighted in this book are cheapened by the constant slamming of other methods. Constant references to fractional factorials leading to "downright wrong results" and similar comments indicate either a profound lack of knowledge (which I doubt), or an ulterior motive. ANY methods that are misused, or used with your brain checked at the door will lead to downright wrong results! Bhote makes contrasts & compares his tool set (all of which he calls "DoE") to the stand-alone classical DOE techniques. This comparison is apples to oranges. It is rarely appropriate to launch into a stand-alone DOE without understanding the process and product through use of other tools. So, many of the conclusions drawn are valid only if you want to compare "apples to oranges". A more appropriate comparison would have been Bhote's methods to the full Six Sigma methodology.

Beyond that, there appear to be numerous typos in the examples which were distracting, and made it a challenge to follow the solutions.

In spite of the one-sided comparisons and poor editing, I enjoyed reading the book and learning about the methods.

The book explains excellent problem techniques in a very lucid manner. A must read for any process engineers, Six Sigma belts and any one in production/operation.

Some of the tools the author explains (like Multivary analysis) can be extended into a statistical study (ANOVA) and more detailed results can be obtained. I wish the author does some of it in any forth coming edition atleast as an appendix (if not in main text, to keep its simplicity).