Last column, we looked at overcoming bias in pressure gauges (“Do Simple Things Right”). The gauges also may have random errors in readings. When purchased, pressure gauges should come with paperwork showing the expected error of the gauge. It should include both the bias and the random error.
What’s a typical error? A good gauge well suited for most troubleshooting work has an expected error of 1 or 2%. This percentage may relate either to the entire range of the gauge or the particular reading. If given as a percent of reading, the error may apply only to a restricted range on the gauge.
To see what this means for us, let’s go back to last column’s example of using a 0–100-psig gauge to take readings from 81 to 49 psig. Table 1 lists some readings and the expected error for percent-of-range versus percent-of-reading. It shows that percent-of-range gauges really give a fixed error amount and that percent-of-reading gauges tend to be more accurate. However, the latter are more expensive. Unless you’re careful, your purchasing department will get the lower accuracy gauges.
Some larger plants have internal instrument shops that repair out-of-specification or damaged gauges. If that’s the case at your site, do you know the accuracy of the gauge you’re getting back from the shop? Often, the answer is no. Lack of knowledge adds uncertainty. Unless the gauge is tested, assume it has an error of 2% of range or more.
At the start of any pressure survey, put all the gauges on a common point with steady pressure and take readings. Then, don’t use any gauge that is more than the error range away from the average.
Reducing the effect of random errors requires different techniques than those for addressing bias. Using multiple gauges at the same location and at the same time can cut the consequences of random errors. Whether or not you must resort to this depends upon the accuracy and precision necessary to justify a conclusion.
Consider a situation I faced a while ago. It required measuring four points to create a pressure profile. The pressure ranges were from 5 psig to 22 psig. The most important range was 12 psig to 22 psig; that range demanded an accuracy of ±0.25 psi to enable a valid decision.
We had 0–30-psig and 0–60-psig gauges available. At the start, we assumed all gauges had an accuracy of ±2%. This created a problem. Relying on single gauges would cast real doubts on the certainty of our conclusions.
So, we used eight gauges split into two sets. One set contained two 0–30-psig and two 0–60-psig gauges. The second set consisted of four 0–30-psig gauges. During use, we identified one gauge as being damaged and removed it from service. Table 2 shows the variation found among the gauges at common points. Assuming random errors, the accuracy was closer to ±1%.
The eight gauges allowed us to deploy four gauges at each location to average out random errors. Taking two pressure surveys with the sets switched enabled us to remove biases from the readings. Both techniques together generated a usable pressure profile and helped identify the process problem.
Even simple jobs such as taking pressure readings have a right way. Performed correctly, pressure profiles can be an invaluable troubleshooting tool.
ANDREW SLOLEY is a Chemical Processing contributing editor. You can email him at ASloley@putman.net.