Selection guide
Designing a touch HMI screen around real operator conditions
Short answer
Choose the touch technology for the input method and environment first: resistive when any glove or stylus must work, capacitive only with verified glove and stylus support on that model, infrared only with a clear optical path. Then build each screen so a tap, a displayed value and a completed machine action are three separately verified events, with command buttons sized at 7.0 mm bare-finger minimum (ISO 9241-910) and 12 mm with gloves.
On this page
- Match the technology to the input that will actually touch it
- Size every target to the finger that is there
- Put commands on buttons, leave gestures to content
- Verify a tap, a value and a machine action separately
- Bind every object through the four layers
- Read the fault scope before touching the binding
- The decision, case by case
- Questions asked
A touch HMI is an operator station whose only input device is the screen: values, status, alarms and commands all travel through touch. Designing one that survives the plant floor is a selection job before it is a drawing job. Pick the touch technology for the input method and environment that actually exist — resistive when any glove or a stylus must work, capacitive only where glove and stylus support is verified for that exact model, infrared only where the optical path stays clear. Then build each screen so a tap, a displayed value and a completed machine action are three separately verified events, and size every target to the finger that is actually there, with ISO 9241-910 as the floor.
Match the technology to the input that will actually touch it #
Resistive, surface capacitive, projected capacitive (PCAP), infrared and surface acoustic wave (SAW) panels detect contact differently, and the detection method decides what can actuate them. A resistive screen is layered sheets closed by pressure through a flexible top layer, so anything that presses — leather, nitrile, a stylus tip, a knuckle — registers. Projected capacitive screens sense a disturbance in an electric field, so a glove or stylus registers only where the model is tuned for it; some capacitive panel PCs list an IP65 front, which resists dust and low-pressure water jets but not hose spray. An infrared frame watches an optical grid and registers whatever interrupts it, including any glove, at the cost of needing that grid clean and unobstructed. SAW panels send acoustic waves across the glass and lose reliability where liquids or grime sit on the surface.
| Input reality | Technology that commits | Verify before you draw a single screen |
|---|---|---|
| Heavy or unpredictable gloves, stylus entry, pressing through dust | Resistive — pressure closes the layers, any glove or stylus works | Touch force, accuracy near the edges, with the actual glove |
| Bare fingers or named thin gloves, multi-touch views wanted | Projected capacitive — glove and stylus support is tuned per model | The model's documented glove thickness and approved stylus type, then a test with your gloves |
| Any-object input on a large panel behind thick cover glass | Infrared — the grid registers whatever breaks it | That the sensing path stays clean, and behavior with a water film on the glass |
The working rule is simple: resistive is the default for gloved and stylus work; capacitive earns its place only after the model's documentation names your glove and stylus or you have tested them on the panel itself; infrared suits large displays where the bezel path is kept clean. For the messy middle case, read up on capacitive and resistive panels with oily gloves and pens before deciding. For glove-first operation on a larger station, compare resistive touch screen options for glove use before fixing the panel size. And if the station itself is hosed down, specify IP69K washdown-rated panel PCs and monitors rather than trusting an IP65 front.
Size every target to the finger that is there #
ISO 9241-910, the tactile-interaction part of the ISO 9241 ergonomics series, treats 7.0 mm as the minimum touchscreen target size for accurate bare-finger activation, and 9.6 mm centre-to-centre as the spacing that stops the neighbouring target taking the hit. A gloved fingertip lands as a larger, less localized patch, so for gloved operation commit to 12 mm minimum and 20 mm as the standard command button, with a 3 mm edge gap between adjacent targets — which puts their centres 15 mm apart.
Worked example, with illustrative inputs and a computed result: a 15-inch 4:3 panel running 1024×768. The diagonal is 15 in = 381 mm, so the active width is 381 × 4/5 = 304.8 mm, and one pixel measures 304.8 ÷ 1024 = 0.298 mm. From that pitch:
- 7.0 mm ÷ 0.298 mm/px = 23.5, so 24 px is the bare-finger floor
- 12 mm ÷ 0.298 = 40.3, so 41 px is the gloved floor
- 20 mm ÷ 0.298 = 67.1, so 68 px is the standard button
- 15 mm pitch ÷ 0.298 = 50.4, so 51 px is the minimum centre spacing between command targets
The decision for that panel: nothing a gloved operator presses goes below 41 px, 68 px is the standard command button, and command targets never sit closer than 51 px centre to centre. When a layout demands smaller targets, the layout loses — rework the screen, and handle the crowded case with the method for sizing buttons in a dense matrix.
Put commands on buttons, leave gestures to content #
Every operating command lives on a single-touch button with a text label, because a tap on a visible target is the one input that works on every technology above. Gestures need two things at once — multi-touch sensing on the terminal and gesture-aware objects in the application — so they belong to content, never as the only route to a command.
| Interaction | Belongs on | Design decision |
|---|---|---|
| Single-touch button | Navigation, alarm acknowledge, every machine command | Meets the size floor above and carries a label |
| Swipe or scroll | Lists, recipes, alarm history | No command reachable only by scrolling |
| Pinch and zoom | Process detail graphics | One touch returns to the overview |
| Role-based visibility | User accounts and permissions | Configuration never shares a login with operation |
The test that keeps you honest: delete gesture support, and if any command becomes unreachable, the design is wrong. Validate the layout from the operator's standing position rather than the engineering chair, because viewing angle and arm reach change what is comfortably pressable. For color, alarm and layout discipline, pair this with designing screens for clear operator decisions.
Verify a tap, a value and a machine action separately #
A tap landing, a value being current, and equipment actually moving are three events on three different paths, and a screen that conflates them lies to the operator. The pressed state on a button proves the application accepted the touch; it proves nothing about the machine. Show command acceptance from the write path, and machine state from a separate read tag bound to the returned condition — motor running feedback, valve limit, drive status — so accepted, inhibited and in-progress are three visibly different indications.
Give the pressed state within 0.1 s, the human-factors breakpoint where a response feels instantaneous, and an explicit in-progress indication by 1 s; an action that can stall needs a timeout message, not an animation that never ends. One safety note, in its place: emergency stop and any safety-rated stop stay on hardwired devices, and the screen only mirrors their state.
Bind every object through the four layers #
Between a screen object and machine behaviour sit four layers: the object's binding, the HMI tag with its data type and scaling, the driver or connection to the controller, and the controller variable plus the logic that consumes it. Design each command as one write to one tag, and read the confirmation back through a different tag, so the write and its proof can never alias each other. An internal HMI value is not a controller value, and a tag with the wrong data type will happily display a plausible number the plant does not have. For the full architecture, read how an HMI works in an industrial control system; the design rule here is that every object you draw can be traced through all four layers before commissioning, not after.
Read the fault scope before touching the binding #
When the display looks wrong, scope tells you the layer before you open anything. One bad field beside healthy neighbours is never a network fault, and a whole page of stale values is rarely a binding fault.
| What the operator reports | Layer to check first | Deciding check |
|---|---|---|
| Many unrelated values frozen or unavailable | Driver, connection, controller availability | Connection status against several independent tags |
| One object wrong while its neighbours update | Object binding or that single tag | The binding versus the tag's runtime value and controller source |
| Value moves but reads the wrong size or units | Data type, scaling, conversion | The raw controller value against every conversion before display |
| Button reacts but equipment does not move | Write path, then permissives and mode | The command tag watched online in the controller |
| Controller value changes, display stays old | Read path and update configuration | The configured update rate against the refresh actually observed |
The decision, case by case #
- Gloves, styluses or dirt on the input path: resistive.
- Bare fingers or verified thin gloves with multi-touch views: projected capacitive, verified on that model.
- Large panel with a bezel that is kept clean: infrared.
- Station that gets hosed down: a washdown-rated unit; an IP65 front is not enough.
- Targets: 7.0 mm bare-finger floor, 12 mm gloved floor, 20 mm standard, 3 mm edge gaps.
- Commands: labeled single-touch buttons only; gestures for content.
- Feedback: pressed state within 0.1 s, machine state from a separate read tag, in-progress by 1 s.
- Faults: page-wide means the link, a single field means the binding — check scope first.
Questions asked
Can the emergency stop live on the touchscreen?
No. Emergency stop is a safety function and belongs on a hardwired device that does not depend on the display, the application or the communication link. The screen can mirror e-stop status and show why a stop occurred, but it must never be the only way to initiate one.
What button size should I use for gloved operators?
Treat 7.0 mm as the bare-finger floor from ISO 9241-910, then go to 12 mm minimum for gloved fingers and 20 mm as the standard command button. On a 15-inch 1024×768 panel that works out to 41 px minimum and 68 px standard, with 15 mm centre pitch so neighbouring targets keep a 3 mm edge gap.
Why does a button light up when pressed but the motor does not start?
The pressed state only proves the touch landed on the object. The command still has to cross the write path to the controller, and the permissives, mode and interlocks there decide the outcome. Show the returned equipment state from a separate read tag, and trace the permissives before blaming the screen.
Is a capacitive panel with an IP65 front suitable for washdown?
No. An IP65 front resists dust and low-pressure water jets, not hose-directed spray. If the station is cleaned with hoses, specify a washdown-rated unit and keep the gasket and mounting conditions the rating assumes; a front-panel rating alone does not describe the installed assembly.
Do all touch HMIs support pinch and zoom?
No. Gestures need both multi-touch sensing on the terminal and gesture-aware objects in the application; one without the other gives you none. Keep every operating command on a labeled single-touch button so a usable screen never depends on a gesture working.