The Man of La Mancha never got to the unreachable goal—and if you’re being judged by overall equipment effectiveness (OEE), then your manager may also be dreaming an impossible dream. This column will look at problems associated with the use of OEE values.

OEE is a value often used in lean production. The concepts of lean production are built upon the implicit idea that we know how to operate our processes at their full potential. When this happens, process steps can be synchronized and production runs smoothly. However, processes don’t operate up to their full potential on their own. Full potential requires predictable operation, and predictable operation can only be achieved and maintained by process behavior charts.

One of my clients had a plant that was built for a capacity of 40 units per month. By getting their processes operating predictably and integrating them, they were able to consistently produce 120 units per month. Yet statistical process control (SPC) is seldom incorporated into lean production schemes. As W. Edwards Deming said in 1980, “Managers do not know what questions to ask, neither do they know what to look for.” But they do know how to create report-card values. And OEE is a report-card value from hell.

OEE is the product of three values commonly known as availability, quality, and performance. We begin by considering each of these components in turn.

Availability

Availability is usually taken as that percentage of the scheduled time that the equipment was up and running. Formerly, this simple ratio was called efficiency. Figure 1 shows a sign I found on a machine in 1983.

Figure 1: Dynamometer efficiency

One wonders if a dual-end dynamometer should, perhaps, be 200% efficient? In any case, this value shows that availability values are often little more than data divided by an assumption. While this ratio will tell us where we stand relative to the assumption, it might not tell the whole story. And while availability values may be used to rank machines, processes, and departments, these rankings will tend to shift from day to day, week to week, and month to month. As this happens, the rankings will appear to play musical chairs from one report card to the next. The fact that the sign was still on the dynamometer a year later suggests that 115% was a high-water mark for this machine.

Quality

Another component of OEE is quality. This is usually taken as the first-pass yield, the percentage approved on first test, or some other equivalent measure of quality before rework. This measure is focused on how well your process is meeting the voice of the customer, and just like availability, it will naturally vary from one month to the next. Here, the goal is 100% conforming product.

But simply computing the percent conforming does nothing to improve things. Improvement generally requires the reduction of variation, and the proven technique for doing this is the real-time use of process behavior charts.

So, while the percent conforming is an important and easily understood report-card metric, it doesn’t provide a way of moving a process toward the goal.

Performance

Availability has to do with equipment utilization. Quality has to do with conformance to specifications. If it’s on time and in-spec, what’s left to be concerned about? What’s left to be characterized with the performance component of OEE? It turns out that this component is rather ill-defined in the literature. It’s supposed to characterize how well the equipment is operating compared to its “maximum potential,” and to do this we have to know what maximum potential entails.

A process is operating up to its maximum potential only when it’s operated predictably and on target. Predictable operation is required to achieve minimum variance, and minimum variance is required for lean production. Thus, any credible measure of performance will need to compare current process operation with predictable operation.

Fortunately, we already have the performance ratio, P_p, and the capability ratio, C_p. These ratios are based on sufficient statistics, which means we can’t find any better measures for what they do. When we divide the performance ratio by the capability ratio, we compare the current process operation with predictable operation. Thus, the best metric for process performance per se will be:

As your process is operated closer to its maximum potential, this performance metric will approach 1.00 from below. A little algebra will show that it compares the estimated minimum process standard deviation with the current observed standard deviation. Thus, a value of say, 0.48 would suggest that you could cut the width of your histogram of product values in half by learning how to operate at maximum potential.

Process characterization

Each of the three metrics above make sense in their own right. An availability of 96% means that about 19 minutes per shift were lost to downtime. With 94% conforming, we would expect a production run of 8,000 parts to produce about 480 nonconforming parts. And when P_p = 0.61, and C_p = 1.09, our performance metric of 56% suggests that we can reduce the width of the product histogram, which is currently 22 units, down to about 12 units by learning how to operate with minimum variance.

As illustrated above, each metric is expressed as a proportion or percentage. But each is interpreted in terms of the measurement units used to form that percentage. It’s the units that disappear when we compute the percentage that tell us what the percentage means. The measurement units of the numerator and denominator always define the context for a computed ratio. By using this principle, we obtained a clear idea about the availability, quality, and performance of the process described above.

The OEE multiplies these three metrics together to obtain an aggregate value. For the example above, OEE = 0.50. And the first question has to be, “What does this mean?”

The indictment

The OEE aggregate value has three major flaws: It destroys context, it violates the laws of mathematics, and it creates an unreachable goal. Each count of this indictment will be explained below.

Loss of context

As explained above, the availability metric will be a ratio involving time; the quality metric will be a ratio involving units of production; and the performance metric defined above is a ratio involving measurement units. When we multiply these three metrics together, we’re effectively creating a ratio where both numerator and denominator have units of:

Before we can make sense of any ratio, we have to be able to make sense of the numerator and denominator separately. So, just what is a [minute part ounce], or an [hour batch pound]?

When we divide one number by another related number, the units may cancel out. But nonsense divided by nonsense is still nonsense. OEE values lack meaning because they don’t have a meaningful context. Arithmetic might not create meaning, but it can easily create nonsense. Why this is so is explained next.

Laws of mathematics

Ratio-scale data require a well-defined notion of distance plus an absolute zero. The metrics for availability, quality, and performance are based on ratio-scale data, and as such they create an ordering for each category. However, the notions of distance used with each category differ.

Before we can add or subtract values, they have to use the same notion of distance. For example, we can’t subtract 10°C from 75°F until we convert one value to the other scale.

Multiplication and division require the same notion of distance and the same absolute zero point. While the vast majority of our data will satisfy these basic requirements of the laws of mathematics, we have to be careful when working with values computed from the original data.

The three components of OEE have different notions of distance that measure different things (minutes, parts, and inches). This means that, relative to each other, they are merely ordinal-scale data, and the multiplication of ordinal values is forbidden by the axioms of arithmetic. The multiplication of ordinal values will always result in nonsense.

The unreachable goal

OEE values will always be skewed so that “good” values are rare. The following example illustrates this. Our three metrics of availability, quality, and performance are percentages between zero and 100%. While percentages are generally treated as continuous, we can illustrate what happens using a finite number of values for each metric.

Let’s assume that each of our three metrics can take on any one of the six values:

{0.5, 0.6, 0.7, 0.8, 0.9, and 1.0}

When we multiply these six availability values by the six quality values, and then multiply these 36 results by the six values for performance, we end up with 216 results. However, these 216 results fall on only 44 values spread out between 0.125 and 1.0, as shown in Figure 2.

Figure 2: OEE outcomes for finite example

While half of our availabilities, half of our qualities, and half of our performances are 0.8 or greater, half of the OEE values are 0.39 or smaller.

Three-quarters of these OEE values are less than 0.50.

And only 10 of the 216 OEE values (4.6%) are 0.80 or larger.

Thus, OEE produces an abundance of small values and a scarcity of large, favorable values. The skewness of the histogram in Figure 2 is a characteristic of the way OEE values are defined. The scarcity of values in the upper tail will make virtually any favorable OEE value an unreachable goal.

Although it’s true that an improvement in any one of the three metrics will result in a larger value for OEE, the skewness of the histogram means the changes will be nonlinear. For example, if all three metrics start out at 0.90 and one is improved to 0.95, the OEE will change from 0.729 to 0.770, a gain of 0.041. But if all three metrics start out at 0.70 and one is improved to 0.75, the OEE will change from 0.343 to 0.368, a gain of 0.025. Thus, any improvement in one metric is discounted by the values of the other metrics before it shows up in the OEE value.

So, even if OEE had meaningful units in numerator and denominator to provide a context for interpretation (which it doesn’t), and even if it didn’t violate the axioms of arithmetic (which it does), this nonlinear response to any specific improvement would still undermine its usefulness in practice. Thus, OEE is simply a report-card value that’s skewed so that “the beatings will continue.”

Summary

Meaningful metrics for availability, quality, and performance exist, but the greatest among these, in terms of getting the most out of lean production, is the performance metric.

Both the availability and quality metrics describe the past, and they help us to understand where we are at present. But management requires more than a rearview mirror. It needs prediction, and prediction requires more than a description of the past. You also need to know what’s possible. And of the three metrics considered here, only the performance metric defines and communicates what can be achieved with your current equipment, your current process, and your current personnel. Process behavior charts provide a proven method for getting any process to operate at its maximum potential, but until you learn how to use them effectively your performance will continue to fall far short of your full potential.

So, while you may use the metrics for availability, quality, and performance separately, you should resist the temptation to combine them. For when they are combined into OEE, they create a value that can only be described as numerical jabberwocky.