How to Tell a Command Problem From a Feedback Problem

Why this matters

A loop that will not hold its target gives you one symptom and three suspects: the controller's decision, the path out to the machine, and the path back from the process. Guess wrong and you replace a controller because a sensor lied, or a sensor because an output driver quit. The way out is not more staring at the screen. It is one cut in the right place and two substitutions, after which two of the three suspects are cleared on evidence.

This is the procedure for making that cut, and the part worth reading twice is not the tests. It is what the second test quietly does not cover, because that exclusion is where a large share of real faults live.

Isolate before the first cut

You are about to open an enclosure that in most equipment carries line-voltage terminals a few inches from the control terminals, and then disconnect wiring in it. De-energize the line-voltage supply, lock and tag it, and prove dead before your hands go in; for a panel or a branch circuit that is 29 CFR 1910.333(b)(2), and the live-dead-live proving sequence is NFPA 70E-2021, 120.5. Any resistance measurement in this procedure is taken on a de-energized, isolated circuit, because an ohmmeter applied to a live circuit reports nonsense and can be destroyed doing it.

Two of the three tests below require the control circuit energized to be meaningful. 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. The control side being 24 volts does not make the enclosure 24 volts.

Where the final element you are about to drive stores energy - a spring-return actuator, a loaded linkage, an accumulator, a pressurized vessel - that energy is isolated and relieved under 29 CFR 1910.147 before any part of you is inside its swing path.

The cut, and the three parts it makes

Make the cut at the controller's own terminal strip. That single boundary partitions the loop into three testable parts: the controller (its decision logic, its scaling, its output driver), the command path (wiring out, the actuator or switching device, the final element, the physical action), and the feedback path (the sensing element, its coupling to the process, its wiring back, the controller's input).

Any other cut point is legitimate but clears less. Cutting at the actuator's own terminals tests only the actuator; cutting at a field junction box splits the wiring but leaves the controller and the element on the same side. The terminal strip gives the biggest reduction per disconnection, which is why it is the default.

Test M: what the controller is actually putting out

Do this before you disconnect anything, because it is non-invasive and it can end the visit.

Read the physical output at the controller's terminals with the loop running, and compare it to what the controller says it is commanding. A controller displaying 100 percent output while its terminals sit at the bottom of their signal range has a fault entirely inside itself, and no amount of field testing will find it. A controller whose terminals match its display has cleared its own output driver, and the fault is now field-side or decision-side.

Skip Test M and you can spend an hour proving a healthy actuator is healthy.

Test A: drive the field side with a known command

Disconnect the command wiring at the terminal strip and apply a known command to the field side from a hand source: a signal generator for an analog output, a hand switch or a temporary source of the correct control voltage for a binary one.

Sweep the range rather than testing one point. Command the low end, the high end and the midpoint, and at each one record two things: what the final element physically did, measured against a fixed reference mark, and what the process did.

Do not command a final element that serves a combustion path, a pressurized or refrigerant-bearing path, or a relief or protective function. Those are proved by simulating at the device that controls them, never by driving the element itself. And before you drive anything, ask what the commanded position does to the rest of the system while you hold it there: a chilled water valve held wide open with little airflow can freeze a coil, and a damper driven closed against a running fan can pull enough negative pressure to damage duct. Where either is possible, stop the affected equipment first.

If the element follows your hand across the sweep and the process responds, the command path is cleared end to end.

Test B: feed the controller a known measurement

Disconnect the field wiring on the input side and inject a known value in its place: a decade resistance box or precision resistor for a resistance element, a calibrated signal source for a current or voltage transmitter.

Inject at two widely separated values, not one. One point can be matched by a wrong scaling with a compensating offset; two cannot. Read the controller's displayed engineering value at each and compare it to what your injected value represents on the element's published curve.

If the display tracks both injections, the transmission and the controller's input scaling are cleared.

The four outcomes, and the asterisk on one of them

The asterisk is on the first row. Injecting at the controller's input terminals substitutes for everything from that terminal outward, which includes the sensing element and every inch of field wiring between the two. Test B proves the controller reads honestly what arrives; it says nothing whatever about whether what arrives resembles the process. A tech who reads the first row as "the feedback is good" has misread it. The correct reading is "the feedback path from the terminals inward is good, and the element and its field wiring have not been tested at all."

The fourth row is the one people force. Neither side responding usually means both sides share something that is down, and the reflex is to keep cutting further out. Check the shared items first: control transformer output under load, the common return, and any permissive or safety chain feeding the controller.

Worked example: the loop that thought it was satisfied

A discharge air temperature loop on a cooling coil. Setpoint 55.0 F. The screen shows actual 54.8 F and output at 8 percent, which reads as a perfectly satisfied loop, and the complaint is that the air is warm. A reference thermometer in the duct near the sensing element reads 62.0 F.

So the loop is not failing to control. It is controlling correctly to a number that is wrong by 7.2 F.

Test M. The controller commands 8 percent; its output terminals read 8 percent of their span. Its own driver is honest. Cleared.

Test A. Command wiring disconnected at the strip, signal source applied. At the low end the valve stem sits at the mark for closed; at midpoint it sits at the mark for half; at the high end it reaches the full-open mark. Held at full open, the duct temperature falls from 62.0 F to 54.2 F over 6 minutes. Command path cleared, and worth noting: the coil, the valve and the actuator are all capable of doing the job.

Test B. Injection carries Test A's process hazard reached one step upstream: a false low drives the cooling valve toward closed, a false high drives it wide open, and a chilled water valve held open with little airflow can freeze a coil. Stop the affected equipment first where that is possible, inject only inside a band you have confirmed is harmless, and never inject on a loop serving a combustion path or a protective function. The element is a 1000 ohm platinum resistance element. Field wiring disconnected at the input, a decade box substituted. Injecting the resistance corresponding to 45.0 F, the controller displays 45.0 F; injecting the resistance for 75.0 F, it displays 75.1 F. Two widely separated points, both tracking, so transmission and scaling are cleared.

That is the first row of the table, so by the asterisk the remaining suspects are the decision, a permissive, or the element and its field wiring. The decision is behaving correctly for the number it has, and no permissive is involved in a modulating valve holding at 8 percent. That leaves the element and its wiring, which is precisely what Test B excluded.

Working it out. Reconnect the field wiring and measure the loop resistance at the controller terminals, de-energized and isolated. It reads 1048.8 ohms. For a platinum element on the 0.00385 ohms per ohm per degree Celsius curve, the linear approximation R equals 1000 times one plus 0.00385 times degrees Celsius holds well across ordinary duct temperatures though not at high process temperature, so 1048.8 ohms is 12.67 C, which is 54.8 F. The controller is reading its input correctly.

But the element is sitting in 62.0 F air, which is 16.67 C, and at that temperature the element itself should measure 1064.2 ohms. The circuit measures 15.4 ohms lower than the element alone should.

A resistance lower than the element's own can only come from something in parallel with it. Solve for that parallel path: 1064.2 times R divided by 1064.2 plus R equals 1048.8 gives 15.4 R equals 1,116,133, so R is about 72,000 ohms. Roughly 72 kilohms of leakage, which is what a damp field junction looks like, not a failed element.

The direction is the tell, and it depends on the element family. A parallel leakage path always lowers net resistance. On a positive-coefficient element like platinum or nickel that reads as a temperature too low. On a negative-coefficient thermistor the identical fault reads as a temperature too high. Get the family wrong and you chase the error in the opposite direction.

What a shortcut would have concluded. A tech who read 54.8 F on the screen, put a thermometer in the duct, saw 62.0 F and replaced the element would have installed a new element into the same wet junction box and handed the customer the same 7.2 F error, with a part on the invoice and a repair history pointing at sensors. The leakage path is unaffected by which element hangs off it.

What would flip the conclusion. Had the measured circuit resistance come back at 1064 ohms, matching the element's expected value at the true duct temperature, the element and its wiring would be honest and the disagreement would have to live in the controller's reading of it - which Test B just cleared - meaning the two measurements were not taken at the same place or the same time. That is a signal to re-check where the reference thermometer actually sat, not to keep testing.

What changes where you can put the cut

The terminal-strip cut assumes you can disconnect wiring. Two cases where you cannot.

A bus-connected device. Where the element and the actuator talk to the controller over a digital bus, there is no analog signal to inject and no terminal to lift. The equivalents are the bus device's own local override for Test A and its raw counts against its engineering value for Test B. What is lost is the clean partition: a bus fault takes out command and feedback together and lands you in the fourth row.

A loop you are not allowed to stop. Where the process cannot tolerate the interruption, Test A is unavailable that day. Run Test M and Test B only, and be explicit in your notes that the command path is untested rather than cleared. An untested path recorded as tested is how the next tech starts from a false premise.

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
• 29 CFR 1910.147 - control of hazardous energy for mechanical isolation and stored energy
• NFPA 70E-2021, 120.5 - process for establishing and verifying an electrically safe work condition
• Manufacturer documentation for the sensing element's published resistance curve and the controller's input type and scaling
• See related: What a Sensor Actually Reports; How to Verify an Actuator Reached Its Commanded Position; The Difference Between a Status and a Command