How to Diagnose a Ground Loop in Panel Analog Signals
The symptoms are familiar to anyone who has commissioned a panel: an analog reading that sits offset from what the transmitter's local display says, drifts with no process explanation, picks up hum, or jumps whenever a drive or heater somewhere else runs. The instrument bench-tests perfectly; the input card calibrates perfectly; wired together in the building, they disagree. That pattern is the signature of a ground loop - the signal referenced to earth at more than one point, with the difference between those earths added to the measurement. This page is the diagnosis procedure: making the symptom repeatable, finding the second ground, and fixing the loop without ever compromising a safety earth.
Diagnose a Panel Ground Loop in one line: To diagnose a ground loop in analog signals, first make the symptom repeatable by correlating the error with what is running - a reading that shifts when a VFD, welder, or heater operates is coupling through a shared earth path. Then hunt the second reference: a healthy analog loop is referenced at exactly one point, so measure voltage between the panel's signal common and field ground - a significant reading between points that should be equipotential indicates current flowing through the reference - and inspect for the classic culprits: shields landed at both ends, signal commons bonded to ground at both the panel and a field device, and grounded-tip sensors feeding non-isolated inputs. Fix by restoring the single reference with isolators or corrected shield landing, and never by lifting a protective-earth conductor - safety grounding is untouchable and its configuration belongs to qualified personnel.
First Checks: Make the Symptom Repeatable
A ground loop diagnosis starts by turning a vague complaint into a reproducible experiment. Compare the transmitter's local indication against the value the control system displays, at the same moment; a consistent offset between them, with both devices individually healthy, is the core evidence. Then correlate: does the error change when a particular drive runs, when the welding crew is in the shop, when the trace heating cycles? Historical trends make this correlation cheap - an analog point whose recorded noise floor steps up during working hours, or whose offset tracks another machine's run status in a platform such as Merobix, has essentially dated its own coupling source. A ground loop's added voltage comes from current flowing through the earth path, so the error waxing and waning with big electrical loads is the fingerprint.
Separate the loop from ordinary induced noise while you are at it: routing a signal cable in a tray with power conductors also produces hum and jumps, but rerouting or better cable fixes that, whereas a ground loop follows the referencing, not the route. A quick discriminating test is to observe whether the error survives when the signal cable is temporarily moved away from suspect power runs - a true ground loop does.
Find the Second Ground
The theory is one sentence long: an analog signal must be referenced to ground at exactly one point, and a ground loop is what exists when there are two. The hunt is therefore for the second reference. With the loop's documentation in hand, walk the circuit's earth relationships: where is the signal common deliberately grounded - typically once, at the panel - and where else could it be touching earth? The classic culprits, in rough order of frequency: a cable shield landed at both ends, turning the shield into a conductor joining two different earths; a field device whose signal common is internally bonded to its chassis, adding a second reference at the far end; a grounded-tip thermocouple feeding a non-isolated input card; and a maintenance repair that bonded a terminal to the enclosure somewhere along the run.
Measurement confirms what inspection suggests. A voltage reading between the panel's signal reference and the field device's ground - taken by someone qualified for the environment, since panels contain more than analog terminals - reveals the potential difference driving the loop; measuring current in a shield or common that should carry none reveals the loop in the act. On a 4-20 mA circuit, comparing the current at the transmitter end against the current at the input card with a clamp-on milliamp meter is decisive: a healthy series loop carries identical current everywhere, so a discrepancy means current is escaping through an unintended path. Where the readings and the drawings disagree, trust the readings - panels accumulate undocumented bonds over years of work.
Fixes That Work, and the Fix That Is Forbidden
The legitimate fixes all restore the single reference. Land each cable shield at one end only - site standards commonly choose the panel end - and insulate the far end so it cannot brush the enclosure. Where a field device inherently grounds its signal side, or where two systems must exchange a signal while each keeps its own reference, insert a loop isolator or use an isolated input channel: galvanic isolation breaks the loop while the signal passes through, which is precisely what those devices exist for. Grounded-tip thermocouples feeding non-isolated cards get isolated transmitters. And where the diagnosis found an accidental bond - a pinched conductor, a screw through a terminal strip - the fix is the repair of that defect, documented so the single-point scheme on the drawings matches the metal again.
The forbidden fix needs saying explicitly, because it is tempting and it works, briefly: lifting the protective-earth conductor of the offending equipment breaks the loop by removing the safety grounding that keeps that equipment's chassis from becoming lethal during a fault. Protective earths are never sacrificed to clean up a measurement - the loop is always broken somewhere in the signal referencing instead, and any question about which conductors are safety-related versus signal-related is answered by qualified personnel against the drawings before anything is disconnected. Verification then closes the job: offset gone against the local indication, noise floor back to normal in the trend, and the error no longer answering when the drive across the building starts.
Frequently Asked Questions
Why does my analog reading change when a VFD somewhere else runs?
Because the drive's operation pushes current and high-frequency noise through the facility's earthing system, and if your signal is referenced to earth at two points, the varying potential between those points adds directly to the measurement. The reading is faithfully reporting your signal plus the ground-path voltage. Breaking the loop - one reference point, shields landed at one end, isolation where a second ground is unavoidable - removes the coupling; the drive is the trigger, not the defect.
Can I fix a ground loop by disconnecting the ground wire?
Only if the wire in question is a signal reference or a shield landing, never if it is a protective earth. Lifting safety grounding does break the loop, which is exactly why it is a notorious and forbidden shortcut: it leaves a chassis unprotected during an electrical fault. The distinction between signal grounds and safety grounds is drawn by qualified personnel from the drawings, and every legitimate fix - single-ended shields, isolators, isolated inputs - breaks the loop on the signal side while the safety earths stay exactly as designed.
Should cable shields be grounded at one end or both?
For analog instrumentation signals, the standard practice is one end only - commonly the panel end - because a shield earthed at both ends joins two grounds and invites circulating current alongside the signal it was meant to protect. Both-end bonding appears in other disciplines for high-frequency reasons, which is why the answer on a mixed site is the site's own grounding standard. What is never right is the accidental version: a shield landed at the panel and also brushing earth in a field junction box.
Automation services
Need help turning this into a working system?
Merobix integrates SCADA, programs Allen-Bradley and Siemens PLCs, and designs and fabricates industrial control panels.
Meeting requests are reviewed before confirmation.