Defects per Unit

In this chapter, we will discuss the defects per unit or DPU. We will determine the number of defects statistically we get per unit, and the average number of defects in our sample. For example, with a really simple part where we are measuring just one attribute such as the weight. We might find that we have a tiny fraction of a number like 0.001 defects per unit. However, if we have 50 different attributes we are measuring in the part, or if we are delivering a complex product, maybe we find a greater than 5 or even 10 defects per unit. The reason we calculate DPU is to get a baseline. We need to establish where we are, and how that compares to the customer expectations. From there we need to identify our future improvement activity. In order to figure out where to concentrate our efforts we can rely on the customer voice, a cost-benefit analysis of our proposed changes. Or a Pareto analysis to zero in on the vital few areas where we will get the biggest bang for the buck with our improvement activities. We may also use this to identify nonconformance, and where we need to improve our controls to get repeatable results out of this process. Now we’re going to use an example of defects with respect to caps on bottles that won’t seal. We will divide the number of defects by the number of units that we inspect.

Before performing this calculation, it’s important to define exactly what constituted defect for your specific situation. There’s a statistical term called confidence that we need to be familiar with. Confidence is the probability that our sample size properly represents the population that we are testing. The larger our sample size, the more confident we can be. However, since it won’t be cost effective or efficient to test every single unit we produce, we will always select a representative sample. You will work with your team and stakeholders to determine what your level of confidence will be, and how many samples you will test. In this example, let’s say there are 1 million bottles. And if we want to be 95% confident, with a confidence interval of 0.1, we need to take at least 97 samples and inspect them. If we want to get to 99% confident, we would need to increase our sample size and inspect at least 166 samples. Defects per unit will tell us how well we’re performing with respect to the defects found in our sample tests. Let’s do a quick little whiteboard exercise here. Now we are going to do a little calculation here of the defects per unit. And let’s just say that we have 376 defects that we actually find in this batch of parts. Well that’s good, we’ve found 376, but now we need to divide that by how many samples did we actually look at. So if we had 6,000 samples that we inspected in which we found 376 defects. We would divide the 376/6,000, and that would give us the calculation of 0.063 defects per unit.

So what that might suggest within 100 units, you would have 6, and for every 1,000 units you would have 63 defective bottle caps. Now we will cover the concept of parts per million, and how that metric is used to evaluate our process. First, we’ll begin with the formula we discovered for defects per unit or DPU. But now we’re going to multiply that number by 1,000,000. So, for example, let’s say a manager of a fast food restaurant observes 100 customer orders, and identifies 15 defective units. The manager can calculate the parts per million by taking the product of 15 divided by 100 times 1 million. This results in a parts per million of 150,000. Calculating parts per million allows us to consider the big picture and say. If we’re going to see 1 million customers per year, are we happy with 150,000 of them having a poor experience of our quality? It allows us to think about how many customers are impacted. And if that’s an acceptable number, or if we need to target improvement.