How to calculate a corrosion rate: formula, unit and pitfalls

The arithmetic is trivial: a difference in wall thickness divided by a period of time. That is not where the difficulty lies. It lies in choosing which values you feed into the formula, and in reading what the result actually means once you have it.

The method, and why two calculations beat one

A corrosion rate is calculated at a given condition monitoring location (CML), never across a whole piece of equipment. That is the first rule, and it is broken constantly: comparing the minimum thickness recorded in one campaign with the minimum recorded in the previous campaign means nothing if the two minima come from different locations. The number looks like a rate, but it describes the migration of a measurement point, not the loss of metal at any single spot.

The long-term rate takes the original reading (the known initial wall thickness or the first reliable measurement) and the most recent reading, divided by the elapsed time. It is robust: the longer the period, the less measurement uncertainty weighs on the result. It is also slow to react, because a recent acceleration is diluted across all the years that came before it.

The short-term rate takes the two most recent campaigns. It is responsive but fragile: over a short interval, measurement uncertainty can account for a large share of the difference observed. It works as an early warning, not as the basis for calculating a remaining-life deadline.

Table scrolls horizontally on small screens.

The sensible practice is to keep the more unfavourable of the two when estimating a deadline, and to look closely at the equipment as soon as the short-term rate clearly exceeds the long-term one. See estimating remaining useful life.

What the result does not tell you

A corrosion rate is an extrapolation. It assumes that what has been happening carries on happening in the same way, which is only true under certain conditions. State those conditions and you know when to trust the figure; ignore them and you are relying on a number that quietly stopped being valid some time ago.

The damage mechanism must be stable. A change of fluid, of temperature, of flow regime or of chemical treatment alters the way the metal degrades. When that happens, the earlier history stops being representative, and the calculation has to start again from the new situation rather than being stitched onto the old trend.

The degradation must be reasonably uniform. General corrosion progresses in a broadly linear fashion, which is what makes a single rate meaningful. Pitting corrosion, under-deposit attack or corrosion under insulation (CUI) advance in bursts, often confined to a small area. An average rate applied to a localised phenomenon produces a figure that is both reassuring and wrong, because it spreads a sharp local loss across a surface that is barely affected.

The locations must be paired correctly. This is the most frequent source of error, and the most silent, because nothing in the resulting number betrays the mistake. It is dealt with in detail in condition monitoring locations: where to place them.

Uncertainty, which you must know before drawing conclusions

Every thickness measurement carries an uncertainty: surface condition, couplant, calibration, operator, temperature. That uncertainty has an order of magnitude you need to know, because it sets the threshold below which a difference means nothing at all. Treat the instrument's resolution as though it were the truth and you will read signals into what is really scatter.

Two practical consequences follow.

First, a variation smaller than the uncertainty should not be interpreted. Converting it into a rate produces a figure that looks precise and describes noise. The right response is to record the reading and wait for a further campaign before drawing any trend from it.

Second, an increase in measured thickness is a signal about the method, not a piece of data. Metal does not grow back. Beyond the measurement uncertainty, a value higher than the previous one points to a different location, a different technique, or a data-entry error. The correct conduct is to flag it and check, never to smooth the curve so that it behaves.

• Calculating on campaign minimathe minima may come from different locations, and the resulting rate describes nothing real.
• Displaying a single ratethe long-term rate alone hides accelerations, the short-term rate alone reacts to noise.
• Extrapolating from two readingstwo points always define a straight line; you need at least three campaigns to tell a genuine trend from an artefact.
• Ignoring a change in service conditionsthe earlier history stops being representative, and the most rigorous calculation becomes meaningless.
• Applying an average rate to localised damagepitting and under-deposit attack cannot be averaged away.
• Deleting outliersthey are the only information you have about the quality of the measurement; erasing them erases the diagnosis.

Presenting a defensible calculation

A rate displayed on its own cannot be discussed: it is either believed or ignored. For it to support a decision, it has to be shown alongside everything needed to judge it.

Why this calculation is so rarely done across a whole plant

Technically, nothing prevents it. In practice, it means recovering the earlier campaigns, identifying the matching locations, extracting the values and reconciling them, location by location, on every piece of equipment.

Done by hand, that work is only feasible on the handful of most closely watched assets. This is why most sites hold precise corrosion rates for ten items and none for the other hundred, even though the readings exist for all of them. The data is not missing; the time to bring it together is.

This is exactly what automatic recovery of the reports changes: applying the same treatment across the whole plant, including the equipment nobody had time to attend to. Any tool that claims to do this should be judged on two points only: whether it attaches each value to an identified condition monitoring location, and whether it keeps the link back to the source document.

For the methodological context, see risk-based inspection and our Maintenance Intelligence page. On the equipment families most concerned: piping and pressure equipment.