Noisy 4–20 mA Flow Signal on a Long Shielded Run With a VFD Nearby
A 4–20 mA signal from a loop-powered electromagnetic flowmeter in a remote pump house runs 210 m on shielded twisted pair to a PLC analogue input card. The card is single-ended, uses a 250 Ω sense resistor, and is specified at 60 dB CMRR at dc, falling to 45 dB at 50 Hz. The transmitter is galvanically isolated; its cable shield is bonded to the local earth bar at both ends. Earth resistance at the pump house is 1.4 Ω and at the control room 3.9 Ω, and a standing 1.2 V dc exists between the two earth bars. With the plant idle, the indicated flow is steady within ±0.2%. Once a 90 kW variable-frequency drive on a nearby pump starts, the reading wanders by ±2.5% and occasionally drives the controller to its low limit. The 4–20 mA current measured with a clamp meter at the PLC end is steady, but a 6.2 kHz ripple of about 3 mA peak appears on the shield. A colleague proposes simply raising the input filter time constant from 0.2 s to 2 s and lifting the shield at the PLC end.
1. Identify the dominant coupling path and show, with numbers, why the standing dc earth-potential difference cannot by itself explain the ±2.5% error.
Read the worked answer
Do the dc arithmetic first. The 1.2 V earth-potential difference arrives as common mode. At 60 dB, CMRR = 1000, so the input sees 1.2 V/1000 = 1.2 mV. Across the 250 Ω sense resistor that is 4.8 µA, or 0.03% of the 16 mA span. Even if the real card were ten times worse than its datasheet, the error would be 0.3%, not 2.5%. The ±2.5% corresponds to 0.4 mA, i.e. 100 mV across the sense resistor. So the standing offset is not the mechanism.
What does match is common-mode injection from the drive. The 6.2 kHz on the shield is the inverter carrier. That current exists only because the shield is earthed at both ends, forming a loop through two earth systems at different high-frequency potentials. The drive's PWM output charges the motor winding-to-frame capacitance, and the resulting common-mode current returns through the motor frame, the tray, the building earth and the cable shield.
The card's 45 dB figure is only guaranteed at 50 Hz. Single-ended inputs typically lose 20–40 dB of CMRR per decade above mains frequency, so at 6.2 kHz the rejection is far poorer, and the residual is rectified by the input amplifier and the ADC sampling process into a slowly moving dc offset - which is exactly what the operator sees. The loop current itself being steady confirms the disturbance is common mode, not an impressed series voltage in the pair.
2. Rank the corrective options and state which one is the real fix. Comment on the filter increase from 0.2 s to 2 s and on lifting the shield at the PLC end.
Read the worked answer
Ranked by benefit per unit of effort:
- Break the earth loop. Keep the shield bonded at the transmitter end only, and terminate the PLC end through a small capacitor or leave it open. This is free and usually decisive. It is safe only if the transmitter end is solidly earthed; both ends left floating is worse than both ends bonded.
- Take the single-ended input out of the problem. Fit an isolated signal conditioner, or move to a differential input card. Isolation breaks the galvanic path altogether; a differential input preserves CMRR at high frequency rather than only at dc.
- Fix the drive's own earthing. The motor cable must be shielded and glanded 360° at both ends - the one place where two-end bonding is correct, because it gives the common-mode current a controlled return path instead of routing it through the signal shield.
- Segregate the route. Separate trays, at least 300 mm from motor cables, crossings at 90°, metal conduit where cables must run together.
On the proposals: increasing the filter from 0.2 s to 2 s does not reduce the common-mode stress; it averages the demodulated offset and hides the fault, while adding enough phase lag to destabilise a flow loop. Lifting the shield at the PLC end is directionally right, but only as a single-point earth with the transmitter end still bonded - done on its own, without isolating the input or fixing the drive earthing, it just relocates the problem.
3. What measurements would you take, in what order, to confirm the diagnosis before buying an isolator?
Read the worked answer
Sequence the tests so each one eliminates a hypothesis:
- Drive off, drive on. Compare the indicated flow and the shield current with the VFD stopped and running at the same pump duty. If the ±2.5% appears only with the drive running, the drive is the source.
- Clamp the shield current with a wideband probe at both ends. Equal readings at both ends confirm a circulating loop current; a large difference means the current is being diverted through an unintended path such as the tray.
- Differential probe between the two earth bars. Measure the high-frequency voltage, not just the dc value. A few volts at 6.2 kHz is enough to drive the observed effect given the degraded CMRR.
- Substitute the transmitter. Replace it temporarily with a 250 Ω resistor across the loop terminals and leave the cable in place. If the noise persists, the transmitter is exonerated and the coupling is into the cable/earths; if it disappears, look at the transmitter's own earthing and its loop supply.
- Lift the shield at the PLC end and repeat step 1. A large improvement confirms the ground loop.
Only after 1–5 should an isolator or differential card be ordered. Doing the substitution test first is what prevents the common outcome of fitting an isolator and leaving the real coupling path intact.