For any testing that a part is suppose to pass a test, let's say the leakage test of a seal flange.

How should I individualize the cause?

I know there is root cause analysis and for different cases I probably need Design of Experiment (DOE) analysis for considering all cases of test to find the root cause?

I have not seen any papers that talks about doe analysis and root cause analysis at the same time.

  • $\begingroup$ Thank you, sure, yes. One doubt, most of the parameters in DOE are numbers, but RCA talks parametric and or qualitative. Indeed I was thinking that DOE is the input for RCA. maybe you could make an example in an answer. Would be great $\endgroup$ Oct 30 '19 at 11:31
  • $\begingroup$ yes sure, may I write a question ? $\endgroup$ Oct 31 '19 at 17:08
  • $\begingroup$ I think so, this should help you find a good answer. Also make you sure you link this question. $\endgroup$
    – user1586
    Oct 31 '19 at 17:17

I think you might be mixing up the intentions of the two analysis tools you you mentioned - Root Cause Analysis and Design of Experiments. Consider thinking of it this way:

Root Cause Analysis is something you do in order to help identify, well, the "Root Cause" of some issue or problem. There are multiple ways to go about doing a root cause analysis in the same way there are multiple way to clean your house. Over time, industry has developed a number of analysis "tools" that can be used to help in figuring out a root cause. Short of explaining each tool, for now I suggest just reading up a bit on some of the following techniques:

  • 5-Whys: Basically keep asking "why" it happened until you get to an answer that you can control
  • Fishbone Diagram: visual tool to help map out the different sources that can contribute to the failure - machine related, material related, people related, environment related, etc.
  • Pareto Charts, Scatter Charts, or other graphs: help to identify trends or interactions that might signal when or how a problem started.
  • Failure Modes and Effects Analysis (FMEA): A very methodical tool for evaluating and how various root causes/issues drive different failure modes and their effects which also considering their severity and a number of other factors. This is what's considered a "bottom-up" approach to root cause analysis.
  • Fault Tree Analysis (FTA): The "top-down" brother of an FMEA where a fault is traced down through all the possible chains of events that could lead to it.
  • Anticipatory Failure Determination (AFD): A creative approach where you try to figure out HOW to actually MAKE the issue happen. In your case, if you wanted to MAKE seal to leak, how would you do it?

These are all tools that are regularly used in risk based industries (think auto manufacturers, medical devices manufacturers, aerospace, etc.) but certainly in lots of other places as well.

Now, the second topic you mentioned - Design of Experiments - while also an option to help with root cause determination, it more suited to helping test a theory, or develop a design or specification. It's commonly used in determining how variation in different inputs can impact the output. Consider something as simple as boiling an egg. Your inputs could be a number of things:

  1. Amount of Water
  2. Temperature of Water
  3. Time spent boiling
  4. Amount of salt added to water
  5. Age of the egg
  6. Size of the pot

A DOE is performed to analyze the compound interactions between this. In a very high level summary the team would perform the following steps:

  • First decide what they wanted to consider a "correctly" boiled or "good" version of a hard boiled egg WAS. What are our specifications for what we WANT.
  • Then there are some ways to eliminate inputs or "factors" which don't really influence the outcome. The more factors we have to evaluate the more complicated the DOE. In this case, we can probably reasonably eliminate:

  • The size of the pot (unlikely to impact the finished egg - assuming of course its big enough to fit the egg and hold water)

  • Age of the egg maybe? Do we feel that this strongly influences the final product?
  • Amount of water - as long as it's boiling (and we don't break the egg) then this probably doesn't matter.

This leaves us with 3 "factors" that we now want to evaluate (and which also seem reasonable):

  1. Temperature of the Water
  2. Amount of salt added
  3. Time spent boiling.

Once a DOE is set up, we would define a number of different experiments where we boiled an egg while varying those parameters in known amounts. Eventually we'd end up with some data that we could statistically analyze to help us understand what input factors we want to control. We'd end up finding that the temperature of the water is somewhat influenced by the amount of salt added. We'd also learn that the amount of time we needed to boil was correlated with the temperature. Eventually, through our analysis, we'd likely come to the conclusion that while temperature is important, the amount we raise it by adding an acceptable amount of salt, doesn't impact the way our egg end up nearly as much as does the time we actually spend boiling it! Not long enough and we end up with raw egg. Too long and we turn it to dried rubber. Our DOE has helped us identify that, when we now go to boil eggs, controlling the time we boil is the best way to guarantee us an egg made exactly the way we want it!

Now, while this may seem a super trivial example, hopefully you can envision how the concept of a DOE as a "tool" could be extremely useful. And just as there were multiple different tools for performing RCA, the same exits for design analysis!

Hope this helps to better understand how these tools relate.

  • $\begingroup$ thank you so much for the super answers. While your answer is quite clear, there's still one big difficulty. Usually in industrial applications you can't order some tests to get the output to see the effect of varying parameters. The difficulty with doe is that you need to run the tests being able to control everything to then do a sensitivity analysis but if you jist have some data of failure in hand? You might stop in RCA and be forced by some boss to conclude? $\endgroup$ Nov 3 '19 at 14:29

Let assume a leak has been reported in the following pump system. As a result there is 1 inch of water on the floor. To identify the failure a root cause analysis (RCA) needs to be performed. To illustrate RCA few unrealistic conditions such as operating temperature range of -40C to 85C have been specified.

enter image description here

Based on the above image the flange consist of 3 different material with different thermal expansion co-efficient's. The RCA determines the rubber gasket as the cause of the failure.

The corrective action is to find a suitable replacement gasket. There are 20 different alternative gasket options.

Executing a Design of Experiment (DOE) can help determine the best alternative gasket. A list design parameters including the above mentioned unrealistic temperature specification would be input to develop an appropriate DOE. Good design parameters are once that can be measurable. The results DOE will provide a qualitative output to aid the decision making process.

Below is an good example of list of parameter to consider for the DOE.

enter image description here

  • $\begingroup$ Thank you very much, well pump seal was just an example, not my expertise. But the parameters can be : bad assembly by worker. Bad thread. Non concentric holes due to assembly or production. Bad or defected threading. Wrong design. Wrong temperature or wrong seal $\endgroup$ Oct 31 '19 at 5:49
  • $\begingroup$ Thank you for your answer but I'm far from understanding your answer, so pleae help me, maybe if you just put the input example of doe I could understand, because I'm not sure that whether doe wants also the output for each input cause or not. If so, do you need to test esch case? And also how has the RCA procedure resulted in that conclusion? Is it calculation wisely or observation? $\endgroup$ Oct 31 '19 at 5:52
  • $\begingroup$ Bad assembly by worker is not a good parameter. Non-concentric holes are good parameter for the DOE, that is because it is easily measurable. Hole locations is clears specification. $\endgroup$ Oct 31 '19 at 10:47
  • $\begingroup$ thank you very much, but still questions to go : I am not super clear that how does RCA individualized the root cause $\endgroup$ Oct 31 '19 at 11:06
  • 3
    $\begingroup$ You are missing the basic point that RCA is about looking at the evidence from the failure that actually occurred, not doing lots of tests on things that "might be wrong" to see if you get the same type of failure. DOE is about doing tests to demonstrate you have fixed the problem, after you have identified it. Take home message: engineering is mostly about using common sense and logic, not following procedures. That's why engineering is hard - most people don't have much common sense, and aren't logical :) $\endgroup$
    – alephzero
    Oct 31 '19 at 13:58

Is the question to use a system developed by bureaucrats ( DOE) or RCA developed by Deming, Juran , etc , over 70 odd years ? As a lifelong professional failure analyst I always looked for root cause; there is nearly always more than one cause for a significant failure. DOE came up with odd conclusions when doing failure analysis in areas of my experience of creep rupture and high temperature hydrogen attack ; that was years ago , maybe they are smarter now.

  • 2
    $\begingroup$ Sorry, but this just isn’t helpful. Nor is denigrating DOE just because you aren’t good at it. DOE is extremely useful, but much more for optimization, not failure analysis. $\endgroup$
    – Eric S
    Nov 1 '19 at 16:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.