British psychologist, James Reason, has developed a model to explain accidents and
disasters. This model is referred to as the Swiss
cheese model. The idea of the Swiss Cheese models is as
follows. Think about a slice of Swiss Cheese.
In the slice, we have a couple of holes and we think of a hole as a defect.
Now, the Swiss Cheese model, which doesn't look at one slice of cheese in isolation,
but asks what happens when you stick multiple slices of cheese on top of each
other. With a certain small, but positive
likelihood, you can stack up the slices of cheese and all the defects line up and the
outcome is tragic. This is the idea of redundancy.
As you add multiple layers of cheese on each other.
It is less and less likely that you can see though all the slices at once,
But, again, the outcome probability is still not zero.
So what's the probability of a defect in a situation like this?
Now, if we draw this as a process flow diagram, redundant check typically
corresponds to a parallel path in the process flow diagram.
I've illustrated this here with these three paths that are all happening on the
way of producing this flow unit. Now, the orange boxes here are the
redundant test point. What's the probability if each of them
makes the defect with a one percent likelihood?
Well, the likelihood of us making a defect at the very end is simply 0.1 raised to
the power of three. If every one of them catches the defect,
the redundancy kicks in and the defect is detected.
So in order for the defect to happen here, at the end, all three of them have to go
wrong. We can then define the yield of this
process as thirteen. Minus 0.01 raised to the power of three.
So you notice how the process flow diagram,
A true understanding of what's happening in the process is driving how the
individual defect probabilities get aggregated to an overall defect
probability into the process here. In this session, we have discussed two
examples of defects. In the assembly line example, we saw a
situation in which a defect anywhere in the process would leave to a defective
unit of flow at the end. In the Swiss Cheese situation, we could
afford to have some mistakes in the process,
But due to redundancy, this would not necessarily lead to a bad unit of output.
Multiple things have to stack up in a bad way to lead to that fatal outcome.
We've talked about how you can look at the process flow diagram, and then think about
how to aggregate the individual defects, and compute an overall defect probability,
and that allows you then to compute the process here.
When improving processes like the ones we discussed, especially the Swiss Cheese
situations, it's important to not just go after bad outcomes.
Hopefully, these bad outcomes, at the end of the process, are really rare.
Instead, you want to look at internal process variation.
This is the idea of near misses. It's also an idea we will see in more
detail in the session on sig-, sigma. The worst resources are those that
sometimes work and sometimes they don't. If a resource always works and never does
any defects, wonderful. If it is always broken, and everything the
resource touches gets defective, we'll figure that out pretty quickly.
In this session, we have used simple probability theory to describe the
likelihood of a resource producing a defect.
We can then use defects in our understanding of the process flow diagram
to describe the percentage of flow units that are produced correctly.
We refer to that number, as the yield of an operation.
Now, not every time, a resource does something the wrong way,
We'll get a yield loss at the end of the process.
Some defects and internal variation are absorbed by other activities.
There is redundance oftentimes built into the process,
However, understanding such deviations in the process,
Even if they do not lead to fatal consequences at the end of the process, is
a very important point of a good quality management program.