The panel is rarely the problem. Here is how to find out what actually is.
An HMI that passes on the bench and fails on the factory floor is one of the most expensive bugs in industrial embedded work. It cannot be reproduced on demand, so it cannot be closed. Three weeks later it comes back as a support ticket from a different site, a different operator, and a different machine — and everyone assumes it is the same root cause.
Usually it is not. Touch is not a property of the panel. It is an end-to-end property of a five-stage chain, and the panel is only the second stage. Debugging an intermittent failure means finding which stage is lying to you, in the cheapest possible order.
Stage 1: Is the input physically reaching the sensor?
Resistive panels work by mechanical deflection. A three-millimetre polycarbonate window taped over the panel looks like free protection and quietly absorbs half the operator's touch force. A gasket that pre-loads the bezel does the same thing from the other side.
Capacitive panels need a conductive path to ground through the finger. Nitrile gloves, thick leather gloves, and winter gloves generally do not provide one. A cover glass bonded with an air gap instead of optically clear adhesive will usually still work, but the sensitivity margin shrinks, and the failure only appears when the operator is tired and tapping lightly at the end of a shift.
The test is boring, which is why nobody runs it: put the actual glove on, and tap the actual panel, through the actual window, at the actual mounting angle.
Stage 2: Is the sensor producing a clean signal?
Resistive film wears where the interface concentrates presses: the confirm button in the bottom-right corner, the numeric keypad, the alarm acknowledge. After a few hundred thousand cycles the conductive coating can craze, and you get a dead spot that behaves intermittently because pressure distribution changes from operator to operator.
Capacitive electrodes delaminate at the flex tail, especially after thermal cycling. The symptom is one row or column dropping out, which is often misdiagnosed as a software bug because it looks so systematic.
Stage 3: Is the controller inventing touches?
This is where most ghost-touch reports die. Controllers run finite state machines with thresholds: debounce windows, drift compensation, palm rejection, water rejection. A tuning that is perfect for a bare finger in a lab will hallucinate contacts when a drop of coolant bridges two adjacent electrodes, and will silently drop real contacts when a glove pushes the delta below the detection threshold.
Ask the supplier for raw delta values, not coordinates. If the firmware only exposes processed coordinates, you have no observability, and that is itself the finding.
Stage 4: Is the host mis-transforming the coordinates?
Calibration drift on resistive panels is expected, not exceptional. But a coordinate transform with a hard-coded rotation and an assumed origin produces "touch is offset by eight millimetres" reports that read like a hardware fault and are actually arithmetic. Check the transform before you check the glass.
Stage 5: Is the enclosure the cause?
Vibration creates false deflection on resistive panels with no operator input at all. Variable-frequency drives in the same cabinet inject noise that a capacitive sensor reads as contact. A sealed enclosure traps heat, softens the bonding adhesive, and shifts baseline capacitance. None of this shows up on a bench.
Instrument it before you theorise
Log the raw signal, not the outcome, and timestamp it against machine state:
Then correlate failures with spindle on, drive enabled, door open, ambient temperature, shift start. The moment a fault aligns with a state transition, the argument is over and the fix is obvious.
Cheapest variable first
When someone reports "the screen doesn't work," change one variable at a time in ascending order of cost: the operator, the glove, the cover window, the mounting torque, the controller firmware, the panel. Teams that replace panels first spend the most money and learn the least, because a new panel in the same enclosure with the same glove will fail in exactly the same way.
I have watched this loop run three times on one machine. Each panel swap bought about a week of quiet — long enough to close the ticket as resolved — and then the winter shift started and the gloves came back.
"Works on the bench" is not a result. It is the beginning of the investigation. The bench has a regulated temperature, a clean surface, a bare finger, and an engineer who knows where to press. The floor has none of those things, and the floor is the only environment that counts.
Top comments (0)