IMDTouchKnowledge BaseShop main store

Selection guide

Choosing Web-Based HMI and SCADA Architecture for Production Lines

HMI and touchscreens Published 8 min read

Short answer

Run one web HMI/SCADA server per production line, so a server outage costs one line its screens instead of the whole plant. Add a central server only after writing down the outage cost per hour and the redundancy that cost justifies, and keep office viewers on status-only roles enforced at the tag write.

On this page
  1. Should each production line get its own web HMI server?
  2. When is one centralized SCADA server the right choice?
  3. How much production does one server failure take down?
  4. Do web-based HMI projects need dev, staging, and production servers?
  5. How do I keep office viewers read-only?
  6. What hardware should run the line server?
  7. What does deploying the per-line design look like?
  8. Questions asked

Run one web HMI/SCADA server per production line. That single decision contains a server failure to one line, keeps the application beside the controllers it serves, and lets any browser act as a client. Centralize only when you can write down what losing the shared server costs per hour and are willing to pay for the redundancy that number demands. The sections below give the architecture choice, the failure arithmetic, the environment count, the permission split and the deployment order.

Should each production line get its own web HMI server? #

Yes: give each independent production line its own server, so one failed server removes operator visibility from that line only — a six-line plant loses one line, not six.

Web-based changes the client, not the control system. The server still hosts the screens, the communications driver and the tag database; the browsers only display what the server sends. That concentration is exactly why placement decides the failure domain: whatever shares a server shares an outage. An HMI server outage removes supervision and HMI-issued commands, not the controller program itself.

Approach Server placement One server failure removes Where it fits
Server per line Line cabinet, beside its controllers HMI visibility for that one line Independent lines; the default choice
Central plant server Shared server room or virtual host HMI visibility for every line it serves Lines sharing a process, or one combined operations view
Redundant pair Line or central, with tested failover Nothing, when failover works Lines that cannot run without HMI visibility

For a plant of independent lines, the first row is what you deploy. The second row is what you accept deliberately, and the third is what a line earns when its process cannot tolerate the first row.

When is one centralized SCADA server the right choice? #

Only when you can write down what losing the shared server costs per hour and provide the redundancy that consequence demands — shared material systems between lines, or one operations team that must watch all lines at once, are the two cases that justify it.

Two situations legitimately centralize. First, the lines are not truly independent: they share a process header, shared utilities or a common material path, so one server mirrors the physical arrangement. Second, one operations desk must see every line in real time to balance load between them. Outside those two, a central server buys you one machine to maintain and one outage for the whole plant.

If either case is yours, the central server is owed a redundancy it would not need per line: a standby with tested failover, exercised on a schedule. Load balancing is not a substitute — it spreads client load across healthy servers and does nothing when the application or the database behind them fails.

How much production does one server failure take down? #

With six independent lines and a 90-minute restore from a cold spare, a failed per-line server costs 1.5 line-hours of lost visibility; a failed central server costs 6 × 1.5 = 9 line-hours — six times the exposure from the same hardware fault.

Work a representative plant: six independent lines, one spare server on the shelf, and 90 minutes to image, restore and reconnect it — 1.5 hours per failure. These inputs are illustrative; your line count, restore time and target replace them.

  • Per-line architecture: 1 line × 1.5 h = 1.5 line-hours lost.
  • Central server: 6 lines × 1.5 h = 9 line-hours lost, six times the exposure (9 ÷ 1.5 = 6) from one identical hardware fault.

Now set that against an availability target. At 99.9% over 8,760 hours a year, the allowed HMI downtime is 8,760 × 0.001 = 8.76 hours. Restores at 1.5 hours each give 8.76 ÷ 1.5 = 5.84, so five failures (7.5 h) fit the budget and a sixth (9.0 h) breaks it. Both architectures can meet 99.9% on that math; the difference is breadth. A per-line failure lands 1.5 hours on one line. A central failure lands 9 line-hours at once and burns most of a year's downtime budget in a single afternoon.

Decision for six independent lines: one server per line, no standby unless a line cannot run without HMI visibility. If you centralize anyway, the hot standby is mandatory, so the 1.5 hours never lands on six lines simultaneously.

Do web-based HMI projects need dev, staging, and production servers? #

Not automatically: for one line, develop on the engineering workstation and run production on the line server; add a staging server only when your change-control procedure requires full-scale approval before a release can touch the running process.

The dev/stage/production stack comes from web systems that can never stop and change under formal approval. A line HMI does not start there. Start with two environments and add the third when one of three things is true: your change-control procedure demands full-scale approval, production cannot be interrupted to validate a release, or the application now serves multiple lines.

Environment Runs on What it buys
Development Engineering workstation Edits and offline testing away from production
Staging Separate test server, added when justified Final approval against a production-like system
Production The line server Live screens and live controller communications

Two housekeeping rules keep the stack honest: promote the same build through every environment and never edit directly on the line server, and pick a web application server the HMI/SCADA vendor supports for production on that operating system, checking OS licensing restrictions before deployment.

How do I keep office viewers read-only? #

Give every office account a status-only role and enforce it at the write, not the button: the server must reject a tag write from a viewer who knows the page address, which is the test that proves the split works.

Build two roles minimum: an operator role with control rights on its line, and a status-only role for everyone else. The server validates the authenticated session and the role on every write request; hiding a button cleans up the interface but protects nothing. Prove it: log in as a viewer, address a control tag directly with a write request, and record that the server refused it. An unauthenticated, reachable write endpoint is remote control of the process open to anyone on that network — never leave the write path without role enforcement.

Keep the networks separated as well. IEC 62443 groups assets with the same security requirements into zones and allows traffic between zones only through defined conduits, which is the documented shape for letting office browsers see line data without reaching control writes. ISA-95, published also as IEC 62264, places business systems above plant operations and treats the handoff between them as a defined interface — in practice, one controlled path for office access instead of a flat network. Where you already have screens worth mirroring rather than rebuilding, a read-only web client mirror of an HMI screen gives office users data with no write surface at all. For accounts, domain login configuration that centralizes users on operator terminals keeps one identity per person across all line servers instead of per-machine logins.

What hardware should run the line server? #

An industrial mini PC in each line cabinet runs the HMI server next to the controllers it serves, and any browser client — office PC, tablet, panel terminal — reaches it over the plant network.

The line server hosts the runtime, holds the driver sessions to that line's controllers, serves browsers and stores the line's history. Mounting it in the line cabinet keeps the driver hop short and keeps the machine with the equipment it supervises. The runtime it hosts is a separate decision — for one common controller combination, start from choosing a PC-based HMI runtime for a Modbus TCP PLC — and licensing is a vendor question, not an architecture one: what actually differs between paid and free SCADA software is support and licensing terms, not where you may place the server.

Clients at the machine can be anything with a supported browser. A panel PC or touchscreen monitor at the line is the operator station, and a failed client affects one station, not the line.

What does deploying the per-line design look like? #

Deploy line by line: list lines and controllers, stand up the server, configure the driver and prove controller connectivity before building screens, then bind objects and open the application from a clean client with no engineering software installed.

  1. Inventory each line: controllers and protocols, operator stations, office viewers, and what that line loses when its HMI server is down.
  2. Place one server per line and give each a stable name, so clients never depend on a single IP address; custom domain URLs for web-based PLC HMI systems are the pattern for that naming.
  3. Define both network paths: browser to server, and server driver to controller. Office traffic crosses on the one controlled path, in line with the zone-and-conduit separation above.
  4. Create the control and status-only roles, and map every command to a role and an explicitly permitted tag write.
  5. Configure the communications driver and confirm live controller data before building screens — a finished page cannot repair a dead driver path.
  6. Bind each display object to its intended tag and check direction, data type, quality and access rights; a correct tag behind a wrong binding still displays wrong.
  7. Deploy to the line server, open it from a clean client with no HMI or engineering package installed, and log the viewer-role write test from the permissions section.
  8. Document recovery: server image, application backup, controller connection settings, and the spare-client procedure.

Questions asked

Does a web-based HMI have to run on the internet or in the cloud?

No. Web-based means clients reach the application through a browser; the server stays in your plant on the machine network. Nothing in this architecture needs an internet address, and keeping the control-write path inside the plant network is the safer default under the same zone-and-conduit thinking that separates office from control traffic.

If a line server dies, does the line stop?

The controllers keep executing their program; what disappears is operator visibility and any command issued through the HMI until the server is restored. If your operating procedure does not allow a line to run without HMI visibility, give that line a standby server with tested failover, or document a local-operation fallback the operators are trained on.

How many browser clients can one line server handle?

There is no vendor-independent number, and you should not buy on one. A line with a handful of operator stations plus a few office viewers produces tens of concurrent sessions, not thousands. Before adding parallel servers, measure page response, CPU and memory at peak shift, and confirm the bottleneck is server capacity rather than the controller driver or the network.

What makes replacing a failed client fast with this architecture?

The client is only a browser. Any working PC on the plant network points at the line server's address and the operator is back in minutes, with no HMI package to install or license to move. Keep one spare client machine imaged and logged out for exactly this.

Keep reading

  • Selection guide

    How to Choose HMI Software for a New Control System

    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.

    HMI and touchscreens

  • Selection guide

    How to Select Lightweight HMI Software for Modbus I/O Control

    Select on three gates in order: a driver that speaks Modbus RTU (and Modbus TCP if the LAN path is required), a tag license at least 20 percent above your counted total, and a runtime that runs the exported project on a clean PC. A 56-device floor plan counts 154 tags, so a 150-tag edition already fails. Prove one floor plan, one popup, one live indicator and one write before you pay.

    HMI and touchscreens

  • Troubleshooting

    Fix the libQt5WebEngine.so.5 error when an HMI editor will not start

    The message means the dynamic loader cannot resolve one shared library the editor binary needs, so startup fails before any editor code runs. Run a dependency report on the exact failing binary, install your distribution's Qt 5 WebEngine runtime package that provides libQt5WebEngine.so.5, and relaunch only when the report shows no unresolved libraries. Do not symlink or rename another library to match the name.

    HMI and touchscreens