Selection guide
How to Choose HMI Software for a New Control System
Short answer
Count operator stations, controllers and connections first, keep sequencing and permissives in the PLC, and pick the smallest architecture that meets the full requirement. One station with no logging duties takes a dedicated panel; multiple stations needing simultaneous control or sustained history take a PC-based runtime.
On this page
The decision threshold is concrete: at one operator station with operator actions only, choose a dedicated HMI panel application; at two or more stations controlling the same process simultaneously, or history in the 2.8 GiB class of the worked example below, choose a PC-based runtime with one server and clients. You reach that threshold by counting the requirement first — operator stations, controllers and connections, operator actions, data duties and degraded operation — never by opening the editor and letting the brand decide. The method below turns that into a repeatable decision with a shortcut check, an architecture table and a worked sizing example.
Count the requirement before opening an editor #
Five counts decide the shortlist before any product is compared. Write them down; every later trade-off refers back to this list.
- Operator stations: how many, which need independent control versus shared viewing, and whether two operators must act on the same process at the same time.
- Controllers and connections: the controller families, the physical ports and protocols, and any third-party devices whose values must pass through the HMI.
- Operator actions: start and stop requests, mode selection, setpoint entry, alarm acknowledgment and maintenance diagnostics.
- Data duties: logging, historical trends, reports and transfers to other systems.
- Degraded operation: what the machine must keep doing when the display, the network path or the supervisory computer is unavailable.
The operator-action list is also where interface design starts: ANSI/ISA-101.01-2015 expects an HMI to be engineered through a defined lifecycle with a display hierarchy and consistent, restrained use of color, and designing HMI screens for clear operator decisions covers the practical side. On the data side, verifying HMI historical trend retention before committing shows how to pin down what a platform really stores and for how long.
Shortcuts that hide maintenance cost #
Each of these shortcuts buys something real, and each hides a test that still has to run.
| Shortcut | What it hides | Check before committing |
|---|---|---|
| Lowest software price | Extra development, integration or recovery work | Build and restore a representative application |
| A spare panel already on the shelf | Missing project, incompatible runtime, unavailable engineering tool | Download the project to that spare and test it |
| Brand familiarity | Mismatch between local operator control and plant-wide data duties | Compare the offer against the five counts |
| A long driver list | Unsupported data types, addressing or write behavior | Test the exact controller connection and transactions |
| A firmware update for performance | Compatibility changes without fixing an overloaded design | Reproduce the performance problem first |
Keep control and data duties separate #
Machine sequencing and command permissives stay in the controller. An HMI button requests an action; the controller decides whether the action is allowed. One rule sits above the whole selection: a screen control is never a safety function, and anything the risk assessment rates, in ISO 13849-1 terms a required performance level, runs in safety-rated hardware or controller channels, never in HMI software.
Two stations writing the same setpoint or issuing conflicting mode requests need command ownership: give one station write authority at a time, hand it over through a controller-managed flag, and show the current owner on every screen that displays the object. For data duties, decide whether the HMI only displays values or actually moves them between controllers; if it moves them, that gateway role goes into the test application, not into a hope.
Match the architecture to the workload #
Once the counts are written, the architecture usually picks itself. The same boundary is walked through in panel PC HMI versus PLC plus HMI for meter data display: where the machine interface ends and the data system begins.
| Decision | Dedicated HMI panel | PC-based HMI or SCADA |
|---|---|---|
| Local machine operation | Direct fit for screens, alarms, setpoint entry | Fits when computer-based duties ride along |
| Data collection | Limited to the model's documented storage and export | Historian, reporting and export are runtime features |
| Multiple stations | Distributed projects, duplicated connections | Client/server design with central arbitration |
| Replacement | Compatible panel plus a recoverable project | Compatible computer, runtime, OS and licensing |
| Maintenance workload | Projects, firmware compatibility, spares | Adds OS updates, accounts, storage and recovery images |
A browser-based client is an architecture, not a simplification: the application still executes on a server or gateway, and that host joins the recovery plan. When the requirement lands on the PC side, run the runtime on industrial panel PCs at the machine rather than office-grade computers.
Worked example: three stations, one controller, two years of trends #
Take a packaging line with one controller, three operator stations that all need simultaneous control, and 12 analog tags logged once per second with two years of retention. The example is illustrative: it assumes raw 32-bit floating-point storage with no compression and no database overhead, and both of those increase the result.
- 12 tags × 86,400 seconds per day = 1,036,800 samples per day
- 1,036,800 samples per day × 730 days = 756,864,000 samples in two years
- 756,864,000 samples × 4 bytes = 3,027,456,000 bytes ≈ 2.8 GiB (divided by 1,073,741,824 bytes per GiB)
Every term moves the total linearly: sampling every 5 seconds instead of every second cuts it to about 0.56 GiB, and cutting retention to 30 days leaves 31,104,000 samples × 4 bytes ≈ 0.12 GiB. The decision: 2.8 GiB of history plus three stations with simultaneous control rules out a single dedicated panel, so put the runtime on one touchscreen computer as the server and make the three stations its clients. With one station and 30-day trends, the same counts point back to a panel, provided its documented storage covers 0.12 GiB with margin.
Test the shortlist before the order #
Two tests separate a candidate from a brochure, and both happen before the purchase order.
- Build a representative application with your screens, your tags and your worst screen, not the vendor demo.
- Connect it to the exact controller over the exact port and protocol, and exercise the writes, because a driver list is not compatibility. This is also where which HMI software runs on which panel hardware becomes a hard gate: editors and runtimes stay paired within each manufacturer's ecosystem.
- Run the project against a live or simulated controller before download; testing HMI simulation against a live PLC is the cheap way to find writes that never land.
- Time a full recovery from a clean station, and write the minutes down.
The decision #
The committed outcome, case by case:
- One station, one controller, operator actions only: a dedicated HMI panel, with the editable project, engineering software access and one spare panel carrying a tested download.
- Two or more stations controlling the same process simultaneously, or history in the 2.8 GiB class of the example: a PC-based runtime, server plus clients, on computers built for the cabinet door.
- Data moving between controller families through the HMI: prove the gateway role in the test application in both directions; protocol translation is a feature to demonstrate, not assume.
- Any station whose failure stops production: a spare on site and a recovery procedure timed before startup.
Select this way and the editor stops being the decision; the recovery minutes do.
Questions asked
Is free configuration software a good enough reason to choose a platform?
It is a reason to shortlist, never a reason to commit. Free authoring tools still tie the project to that vendor's runtime and hardware, so the real cost appears years later in spare panels and in continued access to the engineering tool. Gate the choice on a representative application that exercises the exact controller connection, and on a restore test to a spare station.
Where do interlocks and safety functions belong when operators start machines from the screen?
Permissives and sequence logic belong in the controller; the screen button is only a request. Safety functions that the risk assessment rates, such as an emergency stop or a guard interlock, must run through safety-rated hardware or controller channels and never depend on HMI software being alive. Latch alarm states in the controller as well, so a failed display does not lose them.
How many stations justify moving from panels to a server-based runtime?
Commit at two. A single station with operator actions only is a dedicated panel. As soon as two stations must control the same process simultaneously, or history grows into the gigabyte class as in the worked example, a server-based runtime with clients is the smaller risk, because it solves command arbitration and shared data in one place instead of per panel.
What should the recovery test include before the purchase order?
Back up the project, wipe or swap the target station, reinstall the runtime, restore the project, download to the panel or client, and reconnect to the controller. Record the elapsed minutes and the exact steps; that figure is your honest recovery estimate for a failure in the middle of a shift, and it belongs in the maintenance plan next to the spare.
Does a browser-based HMI client reduce maintenance?
It moves the dependency, it does not remove it. The application still executes on a gateway or server whose failure takes every client with it, so that host joins the backup and recovery list. Verify where the application runs, where the data lives, and what operators see when the server is down, then treat it as one more system to support.