How to Handle Competing Failure Modes [transcript]

What are competing failure modes? They’re not a big deal, but we do need to look for them in our datasets. If not, we can choose the wrong statistical distribution of our data and make the wrong conclusion about product performance. Let’s talk more about what competing failure modes are and how to handle them. After this brief introduction.

Hello and welcome to Quality During Design the place to use quality thinking to create products, others love for less. My name is Dianna. I’m a senior level quality professional and engineer with over 20 years of experience in manufacturing and design. Listen in and then join the conversation at QualityDuringDesign.com.  

Failure modes are the other important half of our results of tests. They are part of our data set. We can’t just ignore them because they provide valuable information about how our product is performing. I know it’s fun to just look at the numbers at the quantitative data that our test produces. And I know that some of us overlook failure modes and just look at the numbers. The reason I know this is because there’s been many-a-time I’ve been asked to help fit a distribution to a data set and I was sent a Minitab file and that’s it! There was no failure modes in the data set to look at. So, first of all, when we test to failure, we always record and examine the failure mode. We want to acknowledge the way our product fails, so discussing failure mode is a standard part of our verification reports (or whatever report we’re summarizing our data in). We can record it in the discussion part of our report.

Competing failure modes can be a headache if we’re trying to model our failure data to a probability distribution. Sometimes, when we’re just looking at the numbers, it’s obvious that there could be competing failure modes: if we plot our data set and there seems to be 2 distinct datasets, there could be a competing failure mode going on. If we ignore this, it can really throw off the integrity of our calculations. But once we know them, they are helpful. It’s more information about our product’s performance. And there’s things we can do to handle them when we calculate life data or reliability.

They’re called competing because our unit under test can’t fail more than once. There can be only one, so they compete: which failure mode is going to be the one to break this sample? Once we recognize that our data set has competing failure modes, we want to analyze the data carefully. We don’t want to break up the sample set to fit a distribution. Even though failure modes are competing, we’re still testing one product. We’re testing it multiple times to be able to apply statistical distribution. But our distribution is of our product failing, not the way it fails. What we want to do is determine a reliability model or equation for all modes of failure.

An example that’s easy to visualize is tensile testing: tensile testing to failure. We’ll consider that we have a simple part: we have a part body, and a hub, and they’re joined. We performed our tensile testing and we have two failure modes: we have a break at the joint (where the body pops out of the hub) and we have the break of the body of the part itself. For the break of joint, when we perform calculations for tensile strength because of a break at the joint, we don’t just omit the data where our body part broke. We keep the body part breaks data but treat them as suspensions, meaning our test was stopped before the break at joint could occur. Conversely, when we’re performing calculations for body part broke, we keep the joint break data but treat them as suspensions, meaning (again) our test was stopped before the body break could occur. Even though they’re competing failure modes, they’re still unit under test of our product and provide useful information. If we just omit the data instead of suspending them, we’re eliminating life data of our product as it relates to that one failure mode. We had units under test that didn’t fail because of a joint break and we can use that data to better model our product performance for a joint break.

Now we need to combine datasets. After all, we’re reporting on our product, not by failure mode. We can use reliability block diagrams to help visualize how we’re handling the data for these different failure modes. If we remember our electrical engineering training, reliability block diagrams look like a series electrical configuration or a parallel configuration. The boxes on a reliability block diagram for our competing failure modes scenario are the different failure modes (they could also be for components for subsystems). We perform calculations like simple probability calculations because reliability is a probability.

How we set up our block diagram is dependent upon if our failure modes are independent or dependent. So first thing we do is identify, “Are the failure modes independent?” meaning that the failure information of one failure mode doesn’t affect the other. If it’s independent, we can build a reliability block diagram to look like a series circuit diagram. If our part doesn’t fail this way, then it will fail later this other way. We can analyze the data of each failure mode to determine their own reliability (while using suspensions of course) and then the combined reliability of our part failing at the joint and by tearing at the body equals the reliability of the failure mode of the joint times the reliability of the failure mode of the body.

By treating these competing failure modes as suspensions, modeling the performance of each failure mode, and then combining them (following the rules of probabilities), we have a more accurate representation of our product performance.

If dependent failure modes: we’ll need to get into reliability block diagrams with parallel configurations. Like electrical circuits, reliability block diagrams can get complex and (honestly) fun to solve. Sometimes we need a software solution to help us out because it can get so complicated. Describing this in a podcast: it’s not efficient, so I’ll include a link to a wiki page on this podcast blog at qualityduringdesign.com. And reliability engineers are available to help with tricky or complex situations. They may have the software you need to calculate complex reliability models with systems that have serial and parallel relationships.

What can you do with what we’ve been talking about today?

  • Well, first we always report our failure modes with our quantitative data. They are the other important half of the data set.
  • Be aware and watch out for competing failure modes and realize that they can affect the statistical models we apply and the conclusions we make from our data.
  • Remember that we don’t separate the data into distinct groups of failure modes, but we do treat competing failure modes as suspensions and then combine reliability like the probabilities that they are.

I am interested to hear how many of those listening have used reliability block diagrams and let me know what you thought about it. Reach out to me on LinkedIn (I’m Dianna Deeney), or you can leave me a voicemail at 484-341-0238. I get all the messages and I might include yours in an upcoming episode. If you like this podcast or have a suggestion for an upcoming episode, let me know and share this podcast with your designing peers. This has been the production of Deeney Enterprises. Thanks for listening!