The Two Systems That Were Fighting Through One Sensor

Why this matters

The first repair on this call was correct. The tech read the symptom right, measured the right thing, drew the right conclusion from what was in front of him, and installed the right part. It came back in nine days, and it would have come back after any number of correct repairs, because the fault was not in a component at all. It was in the fact that two independent control systems had been wired to one sensing element by a retrofit crew who saved a wire pull.

A shared element is invisible on a graphic, invisible in an alarm log, and invisible on the print that was drawn before the retrofit. It shows up only as an intermittent that correlates with something no reasonable person would connect to a temperature reading.

Isolating one panel does not make this circuit dead

Read this before opening either enclosure. A control circuit reached from two separate sources is not de-energized by locking out one of them. Establish where every source feeding those terminals originates, isolate all of them, lock and tag each, and prove dead at the terminals themselves under 29 CFR 1910.333(b)(2), with the live-dead-live sequence at NFPA 70E-2021, 120.5. A tech who locks the panel he is standing in front of and reaches in has done nothing to a circuit that is backfed from a transformer in another room.

Where a reading genuinely has to be taken with the circuit live, that is the narrow troubleshooting allowance at 29 CFR 1910.333(a)(1), and it comes with meter and leads rated for the circuit's measurement category and available fault current plus the shock and arc-flash protection the assessment calls for.

And a technique rule that is also a diagnostic rule: an ohmmeter is only valid on a circuit with no active source in it. An ohmmeter applied to a circuit that still has a source connected reports arithmetic performed on somebody else's current. That single fact is what hid this fault through the first visit.

The complaint, and the first repair, which was not wrong

A space with a heating system controlled from a wall-mounted 10 kilohm thermistor element. Setpoint 68.0 F. The controller displays 68.2 F and reports itself satisfied. The occupants say the space is too warm and have been opening windows.

The first tech does the right thing: he puts a reference thermometer at the wall element and reads 72.0 F. The controller is reading 3.8 F low, and it is controlling perfectly to a number that is wrong. He replaces the element, reassembles, and watches the display come up at 72.0 F against his reference. The complaint stops. He writes it up as a drifted element, which is exactly what the evidence in front of him described.

Nine days later the same complaint comes back.

What the recurrence pattern pointed at, before anything was measured

The second tech has something the first one did not: two data points and a calendar. He asks when the space is warm, and the answer is that it is warm on warm days and fine on cool ones - which for a heating complaint is backwards, and that inversion is the whole lead.

A heating system that overheats a space more on warm days is not responding to weather. Something else that responds to weather is affecting it. The building's cooling equipment is locked out below an outdoor temperature threshold, so on cool days that system is unpowered, and the first visit happened on a cool morning.

That single correlation reframes the search. The question stops being "what is wrong with this element" and becomes "what does this element share with a system that cycles on outdoor temperature."

The second pair on the element

At the wall element, with both sources isolated, locked and proved dead, there are two cable pairs landed on two terminals instead of one. The second pair runs to a controller in a separate enclosure on a separate control transformer, installed during a later retrofit to give the cooling side a space reading without pulling new cable.

Both controllers measure the same element, and both do it the way most controllers do: by putting a known voltage across a known internal resistor in series with the element and reading the voltage at the junction.

The arithmetic of two inputs on one element

Each controller computes the element's resistance on the assumption that its own internal resistor is the only one connected. Wire a second one in and that assumption is broken for both of them at once.

Take the heating controller as controller A, with a 5.00 volt reference through a 10 kilohm internal resistor, and the retrofit monitoring controller as controller B, a lighter input with a 100 kilohm internal resistor to the same 5.00 volts.

With B unpowered, A alone sees the element. At a true space temperature of 72.0 F the element measures 11,306 ohms. That figure comes from the element's published curve using a beta of 3892 K referenced to 10,000 ohms at 25 C, and the beta form is an approximation valid across the ordinary space-temperature span rather than a universal law, which is why you take the value off the manufacturer's table rather than deriving it at the extremes. A's junction voltage is 5.00 times 11,306 divided by 21,306, which is 2.653 volts, and A back-computes 11,306 ohms and displays 72.0 F. Correct.

Power up B and the two internal resistors are now in parallel: 10 kilohms with 100 kilohms is 9,091 ohms. The junction voltage becomes 5.00 times 11,306 divided by 20,397, which is 2.772 volts. A still divides by its own 10 kilohm assumption and back-computes 10,000 times 2.772 divided by 2.228, which is 12,437 ohms - 10.0 percent above the element's true value.

Ten percent of resistance on this element is 3.8 F, because a thermistor of this type changes roughly 4 percent per degree Celsius near room temperature. That sensitivity is the reason this fault is subtle rather than obvious. Run the identical wiring error on a 1000 ohm platinum element, which changes about 0.385 percent per degree Celsius, and the same 10 percent resistance error is roughly 26 C, which lands off scale and trips a sensor fault immediately. Same mistake, same physics, and one element family hides it while the other announces it.

So A reads 68.2 F while the space is at 72.0 F. Setpoint is 68.0 F, A believes it is 0.2 F above target, and it stops heating with the space 4.0 F warm. The occupants open windows and the heating system, doing exactly what it was told, keeps the space there.

The step test, predicted before it was taken

The arithmetic makes a prediction that is cheap to check and hard to explain any other way: A's displayed temperature should step by about 3.8 F, downward, the moment B is powered, with nothing at all happening in the space.

With A energized and B isolated, A displayed 72.0 F against a reference thermometer reading 72.0 F. Energizing B, A stepped to 68.3 F within one scan, and the reference thermometer did not move. Isolating B again returned A to 72.0 F.

The measured step was 3.7 F against a predicted 3.8 F, which is inside the resolution of the display and the tolerance on the element's published curve. A repeatable, instantaneous step in one system's reading, triggered by the power state of an unrelated system, is not a component failure and cannot be made into one.

Why the monitoring system had been announcing this for months

Run the same arithmetic from B's side and the asymmetry is stark. B reads the same 2.772 volt junction and divides by its own 100 kilohm assumption, back-computing about 124,000 ohms, which on a 10 kilohm curve is roughly 10 degrees below zero F.

B had been displaying an impossible indoor temperature continuously since the retrofit. Nobody had looked, because B was installed as a monitoring point and monitoring points that nobody has a reason to open are not read.

That is worth taking away separately from the wiring: the lighter input is the one that goes grossly wrong, and the heavier input is the one that goes subtly wrong. The system that screams is not the system that suffers, so the visible fault and the reported complaint land on different equipment, owned by different people, and never get put in the same sentence.

The correct ways to share one measurement

Do not parallel two inputs on a resistance element. There is no version of this that works, because each input's own excitation is part of the other's measurement.

Share a current loop in series, not in parallel. A two-wire transmitter feeding a series loop can have more than one receiver in it, because the same current passes through every element in a series circuit. That is the standard, correct way to give two systems one measurement, and the only thing to check is the loop's resistance budget: each added receiver's sense resistance comes out of the headroom between the loop supply voltage and the transmitter's minimum operating voltage.

Or pass the value between systems as data. Where both controllers sit on a bus or can exchange a point, one owns the element and publishes the value. This costs nothing in wire and introduces one new failure mode you must then handle: a stale value held after the publishing device dies, which is invisible without an age check on the receiving side.

Or, most often the right answer, install a second element. A sensing element is a low-cost part and a second one removes the coupling and the shared-failure problem in one move.

What stayed wrong after the wiring was fixed

The retrofit crew's second pair came off, a second element went in for the cooling side, and both systems read correctly. That closed the ticket and it did not close the design question, which is worth raising with the owner because it will come back in another form.

One element serving two systems is a common-mode failure: any fault in that element, its coupling to the space, or its position on the wall is delivered identically to both systems at the same instant, and the second system's disagreement can never be used to catch the first system's error. Two elements in the same space are not redundancy in the formal sense, but they do give you a cross-check that costs nothing to read, and a persistent disagreement between them is one of the cheapest early indicators of a drifting element you will ever get.

The related but separate case is two controllers with different targets acting on the same process through different equipment. That is a contest over an output rather than a coupling through an input, it produces different symptoms, and a sibling card owns it.

References

• 29 CFR 1910.333(a)(1) - live parts de-energized before work, and the narrow conditions permitting energized troubleshooting
• 29 CFR 1910.333(b)(2) - lockout and tagging for work on electric circuits and equipment, including circuits with more than one source
• NFPA 70E-2021, 120.5 - process for establishing and verifying an electrically safe work condition
• Manufacturer documentation for the element's published resistance-temperature table and beta value, and for each controller's input circuit and excitation
• See related: What a Sensor Actually Reports; What Happens When Two Controls Want Different Things; What a Control Signal Type Implies About the Fault