Papers and Case Studies

Availability vs. Reliability Part 1

One of the most common metrics that I see tracked on a regular basis is Availability vs. reliability tracking. In fact, a popular metric, the Overall Equipment Effectiveness (OEE), uses it in its calculation. The OEE is computed using availability, percentage of design capacity, and the percentage of adequate quality production.

Rarely do I see Reliability tracked. I do see Mean Time Between Failures (MTBF) tracked but without knowledge of the failure distribution, this metric cannot be used to compute Reliability. This is perplexing to me because measuring Reliability is a far more accurate measure of equipment performance.

Availability vs. Reliability

First consider definitions of each. Availability is a measure of the percentage of time that a function is ready to operate. Reliability is a measure of the likelihood of failure of an asset (or function) at any instant in time.

High Availability numbers can be achieved without high Reliability values. For instance if a function has redundancy, high Availability is much easier to achieve. If the repair staff is highly skilled and moves quickly, good Availability is possible without redundancy. This, of course, is only true if an abundance of spares parts are stored for that staff to use.

In fact, two identical assets could have identical Availability values, but may vary vastly in the amount of work required to achieve those values. Obviously as the amount of work increases so does the amount of resources (manpower and materials) required. There is also an increased risk to staff safety and the environment. Finally, the carbon footprint increases as new parts are manufactured and the old ones discarded.

So the question remains, why don't more people track Reliability instead, especially if they are interested in equipment effectiveness? One answer might be that it is more difficult to do. Availability is simple. One simply needs to know the amount of time a function (or asset) is not ready to operate. But Reliability requires one to track failure times and then compute the Reliability either parametrically or non-parametrically.

If one is able to track failure times on particular failure modes on specific components, then a parametric analysis is straight forward using Weibull analysis. If however, the failure times include mixed modes on entire machine assemblies, then non-parametric techniques are available and are relatively simple.

I am certainly not advocating to stop measuring Availability, I am however, encouraging a more detailed look by measuring Reliability.