The procedure has eight steps, shown in Figure 1. They are ordered so that inexpensive checks that need no access to the pipe come first, and so that each step narrows the field for the next. A step is not skipped because its fault seems unlikely, since the premise of a persistent error is that the likely cause has already been tried.
Sequence of the proposed diagnostic procedure for a persistent flowmeter error
The investigator writes down what is believed to be wrong, by how much, in which direction, since when, and on what evidence. The evidence is nearly always a comparison, so the comparison is examined as critically as the meter. A mass balance across the works needs every stream that crosses the boundary, including backwash, sludge, sampling and overflow, and needs the change in storage over the balance period. Plotting the difference against time and against flow rate is the most informative single act of the whole procedure, because the pattern of the error points to its class, as Table 1 indicates.
Patterns of discrepancy and the classes of cause they suggest
| Pattern observed | Causes to consider first | Step that tests them |
|---|---|---|
| Constant offset, present at zero flow | Zero drift, earthing fault, low-flow cut-off disabled, loop current offset | Steps 3 and 5 |
| Constant percentage at all flows | Wrong calibration factor, range or pulse scaling; wrong pipe diameter entered | Steps 2 and 3 |
| Error changes with flow rate | Distorted velocity profile, partly full pipe, meter outside its range | Step 4 |
| Error appeared on a known date | Configuration change, replaced component, altered pipework or valve position | Step 2 |
| Intermittent or noisy | Entrained air, electrode coating, electrical interference, loose connection | Steps 4 and 5 |
| Totals disagree, instantaneous rates agree | Totaliser scaling, missed pulses, data gaps, time-base mismatch | Step 3 |
| Slow growth over months | Deposits reducing the bore, liner deformation, drift of a pressure transmitter | Steps 4 and 5 |
The calibration certificate, the nameplate and the parameters stored in the converter are compared line by line. The calibration factor in the converter must be the one on the certificate for that serial number, a point that is easily lost when electronics are exchanged between sensors. The nominal diameter, the full-scale value, the pulse weight, the damping time, the low-flow cut-off and the flow direction are recorded. The sizing of the meter is checked against the actual operating range, since a meter chosen for the ultimate design flow of the works may spend its life at the bottom of its range, where the relative error permitted by its class is larger. The maintenance log is read for anything that coincides with the onset of the discrepancy.
The reported figure passes through several conversions between the converter and the monthly return. The current output is scaled to engineering units at an input card; the range entered there must match the range in the converter. A known current is injected at the transmitter terminals and the value followed to the operator display and the historian. Where volume is totalised from pulses, the pulse weight and the maximum pulse frequency of the receiving input are checked, and where it is integrated from the rate signal, the integration interval, the treatment of missing samples and the handling of signals below the cut-off are examined. The converter's own totaliser is read at the start and end of a period and compared with the reported total for the same period. A disagreement here closes the investigation of the primary device before it has begun.
The installation is compared with what the manufacturer and the standard require (ISO, 2017). The straight lengths upstream and downstream are measured and the nearest fittings identified, with particular attention to partly open valves, bends in two planes and pumps close upstream, all of which produce asymmetric or swirling profiles (Mattingly and Yeh, 1991). The pipe must run full at the meter. A meter at a high point, on a descending leg or immediately before a free discharge may be partly empty or carry air at low flows, and an electromagnetic meter interprets the whole bore as liquid. Chemical injection points just upstream can create local conductivity variations and noise. The earthing arrangement is checked against the drawing, as are grounding rings on lined or non-metallic pipe, the routing of signal cable and the integrity of cable glands.
Most current electromagnetic meters can be checked in place with a verification tool, built in or connected externally, that tests coil resistance and inductance, electrode circuit impedance, insulation and the linearity and zero of the converter against values recorded at the factory. A pass shows that the magnetic circuit and electronics have not changed, which is strong evidence that the calibration factor still applies. It is not a flow calibration. It cannot see a change in the bore caused by deposits or a deformed liner, and it says nothing about the velocity profile. Where the meter can be isolated full of still water, the zero is read directly. If the works can spare the line, the sensor is removed and the bore and electrodes inspected and measured. For a differential-pressure meter the equivalent step is a check of the transmitter against a pressure standard and inspection of the tappings and impulse lines.
If the earlier steps find nothing, the meter is compared with a second measurement that shares none of its components. Three options are generally available at a treatment works, in increasing order of effort. A clamp-on transit-time ultrasonic meter (Lynnworth and Liu, 2006) can be fitted on a suitable straight length; its result depends on accurate pipe diameter, wall thickness and lining data, and its uncertainty is usually several times that of the meter under test, so it detects gross error only. A volumetric test uses a clear-water tank or service reservoir of surveyed dimensions as a measuring vessel, timing the change of level with all other connections isolated, in the manner of the volumetric-tank method used on calibration rigs (ISO, 1987). An insertion probe or a tracer dilution test gives a further independent value where a tapping exists. Whatever the method, the comparison is made at several flow rates that span the operating range.
An uncertainty budget is written for both the plant meter and the reference, with contributions for calibration, installation effects, secondary instrumentation, tank dimensions and level reading, and timing, combined in the usual way (JCGM, 2008; ISO, 2005). If the difference lies within the expanded uncertainty of the comparison, the test has not demonstrated an error, and the original evidence of Step 1 should be re-examined. If it lies outside, the size and flow dependence of the difference are compared with the patterns of Table 1 and the investigation returns to the step indicated.
The remedy follows from the cause: reconfiguration, repair of earthing, relocation of the meter or of a dosing point, fitting of a flow conditioner, cleaning, or replacement with a meter of suitable size. After the change, Step 6 is repeated to confirm the result. The file records the initial evidence, each check with its finding, including those that found nothing, the uncertainty budget and the final comparison. A cause-and-effect diagram of the kind long used in quality control is a convenient summary and a useful aid for the next investigation.