Identifying Sources of Error from Electromagnetic Flowmeters


Thread Starter


I'll give a short background on what I'm doing then I'll get to the question.

I'm currently writing my thesis on the subject of quality assurance of water consumption data. The data I'm working with is the hourly average (l/s) water consumption for different areas in a city, the population within the areas vary from 100 - 20000 people. For the smaller areas the consumption is assumed to be what has been measured with electromagnetic flowmeters at the pump station connected to the area, while in the larger areas the consumption is assumed to be Inflow (from other areas)-Outflow (to other areas) -Reservoir. All measurements are done with electromagnetic flowmeters (Brands include ABB, Siemens & Endress Hauser)

Part of the thesis work revolves around looking at different factors that can be a source of error in the data. I've contacted producers and scanned the web looking for references and so far I've understood that installation effects can be a major source of error at least when it comes to the meters. However the data comes from an SCADA system so I suppose there are other factors that can contribute to errors in the data.( faulty PLC? Data transmission? Data handling?)

And this is were I can't find anymore information on the web, and also the reason I'm posting here. What could other sources of error be? While working with the data I've found some "errors" that show up every now and then and I can not find an explanation for why they occur.

I'll post a link with two examples (don't know if it is possible to attach pictures here)

Explanation to the figures. Flow (l/s) on the y-axis and hour on the x-axis. One can se a pattern in the data, one peak in the morning and then usually one in the afternoon.

In the first figure there is an example of a reoccurring "error", values are repeated for some time then they are followed by a peak (in some cases the peak value - sum of repeated values). In the other figure values are repeated but there isn't the same type of peak following the repeated values.

I've been trying to find examples of types of errors that can arise from flow measurement but I haven't found any.

Does anyone have any ideas why the data looks this way? Also what could other sources of error be except the flowmeter itself?
What sort of data acquisition/historian/HMI/SCADA/DCS is that? I'll bet it's a DCS, the graph is so bad.

The X axis is supposed to be time. What kind of 'hour' is 3700, 3750, 3800, 3850, 3900? My day only has 24 hours in it.

The time resolution on the lower graph is horrible. One can see the straight line interpolation between captured data points. I'd make it out to be about 30 points per time division, whatever that means, for example between 2950 to 3000.

In fact, it looks like the derivative (dy/dt) of slowly recorded totalizer data. Is it?

What size (diameter) magtube is in use to support a 300 L/sec scale? Mags are generally very repeatable to less than 1% FS, so the numbers are likely good, if not frequent enough. It is interesting that except for the single spike, the data is less than 10% of span.

The spike on the upper graph is a single point spike with little else to tell you what happened between data points. That's the nature of low time-resolved data. That was excusable when memory was expensive, but in an era when terabyte hard drives sell for well under $100 USD, it's deplorable, unless of course, data collection was an afterthought, a no-thought or a bureaucratic exercise.

There are legends about the cumulative toilet flush spikes during a Superbowl commercial, so maybe the spike is one of those 'events'.

What kind of signal comes from the flow meter to the HMI? A 4-20mA analog signal can be affected by a common mode ground fault and can have an error like the spike shown in the upper graph.

It could be an 'event' like a 25,000 HP motor starting. Suppose the motor has ground fault issues and the Mag is AC powered and its ground potential is affected by the motor ground fault and the flowmeter's analog output is not isolated and there's no loop isolator so the 4-20mA includes some common mode ground loop current. And a spike. Could be up or down.

However, digital protocols, like Modbus or Profibus generally do not provide bad data, instead they flag a communications error. An exception is Modbus from some gateways, like wireless gateways. If the gateway loses its field signal, the stale data from the field device can remain in the gateway's Modbus register and be polled over and over and over again. The data being stale, never changes. It's the same data, over and over and over.

The result is straight-lining, just like is shown in the lower graph when the flow straight-lines at 2 l/s for a full time unit. That is very suspicious data. What sort of flow is constant for 3000 hours? There are flow test stands that struggle to do that. (Yeah, I know it isn't 3,000 hours, but I have to vent about poorly presented data. 3,000 on a time axis? Really?) Any maintain a constant flow like that. Any chance that this data comes from a wireless gateway?

Things I have personally seen in magmeter flow measurement

1). If there were ever DP flow meters involved, the square root ghost in the HMI might be there to "haunt" the linear magmeter data. Incorrectly square rooted data will read higher than expected.

2). Until recently (last year or two) the error due to installation downstream of piping that creates flow anomalies was a known issue, but not quantified, at least at the user level.

Recently, Toshiba and Siemens came out with specs rating their mags to a maximum of 2% error for dual elbows out-of-plane, with enough credibility to get certified for water metering in Europe. While 2% error might make some batchers cringe, I've met numerous people who are delighted to live with a max 2% error in order to fit the mag tube into the piping without massive piping changes to get "5 up and 3 down" for a straight run.

3). There is rampant human error in configuration of both the transmitter but more so, the HMI. It's incredible how often the range of the meter does not match the range in the HMI. About as often as a policeman on the evening shift runs into domestic violence. And there's the cases of the square rooting of linear data in the HMI.

4). Metallic objects, like a pipe wrench or a box wrench or an open end wrench will affect the accuracy of the reading by altering the magnetic field pattern. Rebar, too. Who knows how it gets there?

5). Analog output drift from the really old converter/transmitters can be significant. They're out there and they're still running because there are vendors who offer generic converter/transmitters to mate up with old/obsolete magtubes. You've gotta wonder what the condition of the coil is on a 30 year old tube.

6). The engineer specs the factory shielded twisted multi-pair cable for the remote mounted converter/transmitter. The contractor orders it. It arrives but in a separate carton from the tube and converter. The installer doesn't even look for the cable, he runs THHN between the tube and the converter/transmitter. And the owner wonders where the offset due to noise comes from.

7). Volume error in low flow rate measurement situations where a horizontal tube accumulates mud and debris. I've seen 4 inches (100mm) of goo-mud in the bottom of a 10" (DN250) tube.

But I suspect none of these applies here, I think you're looking at consumption patterns; which isn't technical, it's sociological.