Locked lesson.
About this lesson
Exercise files
Download this lesson’s related exercise files.
Taguchi DOE.docx61.9 KB Taguchi DOE - Solution.docx
62.2 KB
Quick reference
Taguchi DOE
Taguchi DOE is a special case Fractional Factorial DOE that is used for optimizing process performance. It separates the control factors into two categories and by simultaneously testing each category can minimize the number of test runs.
When to use
Taguchi DOE is a design for process optimization, not product design or technology characterization. Further it must be used with an existing process because it relies on expert process knowledge to separate the control factors from the noise factors. Therefore, the process must exist and someone on the design team must be knowledgeable about the process.
Instructions
The Taguchi DOE requires expert process knowledge and also often requires expert Taguchi DOE knowledge. The Taguchi DOE is a hybrid DOE. It separates the list of traditional DOE control factors into two categories the Taguchi control factor and the Taguchi noise factor.
The Taguchi control factor is a factor that the process operator of the process being studied can control as part of their normal duties. These factors can vary based upon the type of equipment used and the process management systems. For this reason, expert knowledge about the process is needed so that the correct set of factors that the operator controls – not just monitors – are identified. The other factors are referred to as noise factors. This does not mean that they are background factors with minor variations. They could be the most significant factors for the process. But since the operator cannot control them, they are treated as noise for the Taguchi DOE.
These factors are analyzed simultaneously in the Taguchi DOE. The control factors are analyzed in a test matrix known as an inner array and the noise factors are analyzed in a test matrix known as an outer array. The inner array is normally a fractional factorial array and many times these factors are multi-level factors. The outer array is normally a full factorial array and normally these are two-level factors. These arrays are quite complex and there are many books and manuals full of the arrays and their derivation. Fortunately, most of the arrays are already found in statistical software, such as Minitab, so once factor selection has been made, the software will create the test matrix.
The strength of the Taguchi DOE is that it requires very few runs to get excellent insight into process control. The disadvantage is the it requires expert process knowledge to select the correct factors and an analysis for the process at one location may be totally inadequate for another location if they have different equipment or management approach. Another caution with Taguchi DOE is that it uses its own unique terminology. In fact, sometimes it uses the same word as used in other DOE methodologies but with a different meaning. There are three terms in particular that are unique in their usage within Taguchi DOE. Control factors are the factors controlled by the operator. There may be other factors that control the process, but if the operator does not have access to them, they are not Taguchi Control Factors. Noise factors are all factors being analyzed that are not control factors. So, these are factors controlled by other factors that are part of the system, and factors in the environment. They may have a significant effect, but they are “noise” from the operator standpoint. Finally, Taguchi DOE labels a process optimization at robust if the operator is able to control it. The process may actually be very fragile, but if it is fully controlled by the operator, Taguchi calls it robust.
Hints & tips
- Only use this approach with process improvement projects.
- You must have a process expert as an advisor when setting up a Taguchi DOE; otherwise you will not know how to pick the control and noise factors.
- Be very careful with the terminology – Taguchi terms do not carry the normal meaning of the terms.
- Taguchi arrays do not use the plus 1 and minus 1 settings for high and low; they use one and two. However, the Taguchi arrays are still balanced and orthogonal. The use of the different numbers should not matter to you once you put your own values in for high, low and midpoints.
- 00:04 Hi, I'm Ray Sheen.
- 00:05 There's another special case fractional factorial DOE technique that I'd like
- 00:10 to introduce and that's known as the Taguchi DOE.
- 00:14 The Taguchi DOE technique is one of those love it or hate it approaches.
- 00:19 Some people absolutely love it because it helps them quickly to optimize
- 00:23 a manufacturing process,
- 00:24 others absolutely hate it because it lacks some of the statistical purity.
- 00:28 Let's look at it.
- 00:30 Genichi Taguchi is a well known quality guru from the Japanese auto industry.
- 00:35 The Taguchi DOE is a special case DOE that was developed for designing and
- 00:40 optimizing manufacturing processes.
- 00:42 One key aspect of the Taguchi DOE is that the control factors that you use in
- 00:47 a normal DOE are categorized in the Taguchi DOE into either Taguchi control or
- 00:52 Taguchi noise factors.
- 00:54 In the Taguchi approach,
- 00:55 factors that are referred to as control factors are manufacturing process controls
- 01:00 that the operator in the shop floor can control.
- 01:03 All other factors are called noise factors even though they may be factors that
- 01:07 are easily controlled by the business or process management.
- 01:10 But since they're not controlled by the operator,
- 01:13 the operator considers them noise.
- 01:16 The Taguchi DOE will then analyze these control and
- 01:19 noise factors separately rather than analyzing them all together in one system.
- 01:23 The control factors are usually analyzed
- 01:26 with a fractional factorial DOE design called the inner array.
- 01:30 And these runs are normally done within a full factorial analysis of the noise
- 01:34 factors, which is called the outer array.
- 01:37 The goal is to develop a set of operating limits for
- 01:39 the control factors that the operator can use,
- 01:41 that will ensure that the process results meet the business objectives.
- 01:45 When this occurs, the Taguchi analysis says that the process is robust.
- 01:51 Let's look a little deeper at the control factors.
- 01:54 Remember, Taguchi DOE is used for process design optimization,
- 01:59 not product or system.
- 02:01 So the Taguchi control factors are those that are controlled by a process operator.
- 02:05 So one implication is that I have to have a process design with flows, equipment,
- 02:11 and procedures before I can know which factors are controllable by an operator.
- 02:16 That means that whoever creates a Taguchi DOE needs to have in-depth
- 02:20 process knowledge to setup the analysis.
- 02:23 The goal is to achieve robust process performance, meaning that the operator can
- 02:28 always make good parts, despite what is happening in the noise factors.
- 02:32 In this case, robust has nothing to do with fragility of the product or process.
- 02:37 It, instead, is an indication of the level of control available to the operator.
- 02:43 A fractional factorial approach is used with these Taguchi control factors.
- 02:47 And the level of the factors are based upon the levels at which an operator can
- 02:51 control the factors using the current process design.
- 02:55 Now, let's look at the noise factors.
- 02:57 Remember, Taguchi noise factors are not the same as what is traditionally
- 03:01 considered statistical noise.
- 03:03 It's not a low level of uncertainty in some factors.
- 03:07 Rather, Taguchi Noise Factors are the factors that the organization does not
- 03:12 want the operator to control.
- 03:13 So it might include typical noise factors like the ambient temperature fluctuation
- 03:18 but it could also include factors that the business is choosing to control such as
- 03:23 different suppliers for parts of material or different product mix in the schedule.
- 03:28 When conducting a Taguchi DOE determine the min and
- 03:31 max levels that you should use for these factors.
- 03:34 And just like with any other DOE, you'll need to control these factors
- 03:38 to ensure that they are at the appropriate high and low levels during each run.
- 03:43 One of the most confusing aspects of a Taguchi analysis is the test matrix, or
- 03:48 Taguchi array.
- 03:49 The Taguchi array separates the inner array for
- 03:52 Taguchi control factors from the outer array for the Taguchi noise factors.
- 03:56 You can see in this diagram the inner array on the left, and
- 03:59 the outer array on the upper right.
- 04:01 There are many different array configurations, and if we start to try to
- 04:05 explain each of those, we'll be here for about another three hours.
- 04:08 Suffice it to say, the statistical verification of the arrays is solid, and
- 04:13 the selection of different array can be done in your statistical software.
- 04:17 The Taguchi arrays often have mixed levels of both 2-level factors and
- 04:22 multilevel factors in the same array.
- 04:24 The array structure will then have a name
- 04:27 given to them based upon the number of factors and the factor levels.
- 04:31 Before the advent of computers that would do the statistical analysis associated
- 04:35 with a DOE,
- 04:36 there were books published that just contained the different Taguchi arrays.
- 04:40 So the DOE leader did not have to recreate their analysis for each array.
- 04:45 The Taguchi arrays are balanced and orthogonal although that may not be
- 04:49 obvious at first glance.
- 04:50 That's because Taguchi uses levels 1 and 2 instead of -1 and
- 04:55 +1 to indicate the high and low settings.
- 04:58 And if it was a multilevel factor, then the settings would be 1, 2, and 3.
- 05:03 You can probably see how this can get confusing if
- 05:06 you learned how to do DOE one way and then had to switch to the other.
- 05:10 Fortunately, Minitab has it all figured out.
- 05:13 Just tell Minitab you want to do a Taguchi DOE and set up your factors.
- 05:17 Select the level for your factors, then it will create the array for you and
- 05:22 tell you what configurations to test.
- 05:24 Let's wrap this up by looking at strengths and weaknesses.
- 05:28 The best time to use this approach is with a process improvement to an established
- 05:33 process.
- 05:33 You already know what output you want to measure and it's easier to identify
- 05:38 the factors that operators control and all the other factors are then noise factors.
- 05:43 The strength will be low cost in a short time to do the analysis because it reduces
- 05:48 the number of runs as compared to a full factorial DOE.
- 05:52 In fact, it even reduces the numbers compared to a more traditional fractional
- 05:56 factorial DOE, because of the separation of the inner and outer arrays,
- 06:00 essentially, keeps both small and the data points can be used for both arrays.
- 06:05 The weaknesses from a technical point of view is that it requires expert knowledge
- 06:10 to pick the control and noise factors.
- 06:12 So if you don't have an expert available or if the process doesn't yet exist,
- 06:16 you're just guessing.
- 06:18 Another concern that I have is based upon my experience with using Taguchi DOE,
- 06:23 and that's terminology.
- 06:25 Noise factors are not traditional noise factors that are used everywhere else in
- 06:30 statics.
- 06:30 The term control factors only apply to operator control factors not managed by
- 06:35 control factors.
- 06:36 And robust means that the operator is controlling the process despite what's
- 06:41 happening to the noise factor.
- 06:43 But the process itself may still be very fragile.
- 06:47 When used correctly Taguchi DOE is a great and powerful analytical tool.
- 06:52 But it can create a lot of confusion, so manage the process closely.
Lesson notes are only available for subscribers.
PMI, PMP, CAPM and PMBOK are registered marks of the Project Management Institute, Inc.