There is a DIY hardware project that has stuck with me. A designer spent a year building “The Industrial” — a CNC-machined aluminum case for a small form factor PC. He taught himself CAD, worked with overseas machine shops, sourced components, built a website, designed packaging and assembly instructions. Then he wrote a final update: the project was ending. Not because the design failed, but because of tolerances, quality control, logistics — the realities of producing hardware at a small scale. “No project goes exactly as planned,” he said.

If you’re buying an industrial PC instead of making one, the same gap exists between what’s printed in the datasheet and what actually happens inside your plant. The datasheet tells you the CPU, the ports, the temperature range. It doesn’t tell you whether the unit will still be running after three summers in a sealed cabinet next to a VFD. This guide is about the second list — what to check before you commit anEmbedded Industrial computerto a project that will outlive its purchase order.

1. What the temperature range is really saying

“Operating temp: −25°C to +75°C” looks like a promise. It’s actually a lab result — usually measured with the unit open-air, or with airflow over it, under conditions the manufacturer controls. Your cabinet isn’t a lab. Put a box that draws 15W into an unventilated steel enclosure at 40°C ambient, and the air around it can sit 15–20°C above the room. The part that fails first isn’t the silicon; it’s everything around it — capacitors, connectors, the thermal paste that was applied a little thin on a Tuesday shift.

This is why the word “fanless” matters less than the word “aluminum.” A fanless unit is only fanless in the sense of not having a fan — if the chassis can’t move heat, it’s just a slower oven. Look at how the box sheds heat: a die-cast aluminum body that doubles as a heatsink, not a plastic shell with vents painted on. For example, the fanless EG228 industrial PC uses an aluminum housing and is rated for −25°C to +75°C; that’s the kind of design decision that shows up in a five-year deployment, not in a one-week evaluation.

2. Dirty power is the real killer

Factory power is not lab power. Motor starts cause voltage dips, welding arcs cause noise, and someone’s phone charger on the same breaker causes transients. A unit with a wide input range absorbs all of this; a unit that expects a clean, tight 12V brick does not.

Check two numbers: the input range and whether reverse polarity is protected. The EG228 accepts DC 9–36V with reverse polarity protection — so a 24V plant line with sag is fine, and a wiring mistake doesn’t fry the board during installation. The X86 line (EC300, EC500) goes DC 12–36V for the same reason. A wide-range input is not a luxury feature; it’s the cheapest insurance against the most common field failure — power.

3. Count your ports, not your pixels

The second-biggest mismatch between datasheets and reality is I/O. Vendors publish port counts, but what you need is a count of your own field devices: meters on RS485, drives on CAN, sensors on discrete I/O, and a legacy PLC that still expects a serial connection.

Do this before you shortlist: list every device at the site, its protocol, and its physical interface. Then match the box. A mid-range ARM unit like the EG628 gives you 2×RS485, 1×RS232, 1×CAN, 2×Ethernet and 2×USB — enough for a typical machine-monitoring node. When you’re aggregating an entire floor, the EG828-EMS steps up: 4×Ethernet, 8×RS485, 2×CAN, plus 16×DI and 16×DO for direct control of I/O devices. And if you’re migrating old serial-based machinery, the X86 EC500’s 8 serial ports plus 5×RJ45 exist precisely for that job. Datasheets list ports; the selection question is whether your wiring, with margins, fits.

4. The watchdog is a staffing decision

In a control room, a hung PC costs one reboot. At an unmanned pumping station, it costs a drive out there at 3 a.m. — or worse, a fault that nobody notices until the morning shift. This is why anEmbedded Industrial computer’s watchdog is a real feature and not a checkbox.

Two levels matter: hardware and software. A hardware watchdog is independent of the OS — it resets the machine even when the kernel has locked up, not just when a process stops responding. The EG228 ships with a hardware plus software dual watchdog, so a hung process or a dead kernel both recover without a human. Before you buy, ask the vendor: “If this unit hangs, does it recover by itself, and how do I configure the time-out?” If they can’t answer in one sentence, assume you’ll be rebooting it by hand.

5. The software stack is half the reliability

Nobody’s project fails because the box is 5% slower than the spec. Projects fail because integration takes three months longer than planned. The software that ships on the device — and the protocols it speaks natively — is a bigger reliability factor than any benchmark.

Check three things. One: does it ship with a real OS and toolchain (Ubuntu, C/C++, Docker), or do you have to build your own image? Two: does it come with an edge engine and Node-RED pre-installed, so your first data pipeline runs the first afternoon instead of the first quarter? Three: does it speak the protocols your site already uses — IEC 61850, IEC 104, DLT645, DNP3.0, BacNet — and the PLC brands already installed (Siemens, Mitsubishi, Omron, Modbus)? The EG228 and EG628 both ship with Ubuntu 22.04, the WukongEdge engine and Node-RED out of the box, which is precisely the difference between a project that takes weeks and one that takes months.

6. The logistics lesson from small-run hardware

Here’s the part of that project that applies most directly to you. The project didn’t die on the drawing board — it died on logistics. The designer couldn’t store inventory, couldn’t move the remaining units, couldn’t justify a second production run. Hardware reliability isn’t only what happens to a device on your wall; it’s what happens when it fails and you need a replacement, or when firmware needs a fix, or when you order the next batch for phase two.

So add supply questions to your evaluation, and treat them as seriously as the spec: How long is the product guaranteed in production? Do they keep stock, or is every order made to order? Are firmware updates maintained, and is there a documented upgrade path? A box with a beautiful datasheet and a six-month lead time is not an industrial product — it’s a prototype you paid for.

7. When you actually need a screen

Sometimes the requirement isn’t a headless gateway; it’s an HMI that a line operator touches. Before you bolt a consumer monitor onto a gateway, consider a panel PC that was built as one unit. The USR-SH800 industrial panel PC combines an RK3568 quad-core processor with a touchscreen, and gives you 2×RS485, 2×RS232, 2×Ethernet and 2×USB with Ubuntu and WukongEdge built in — so data collection, local configuration and display run on one device instead of three. The rule of thumb: if the operator needs to see it, spec a panel PC; if nobody will ever look at it, save the money and go headless.

8. The three-question test before you sign

Whenever someone asks me how to pick between twoEmbedded Industrial computercandidates that look identical on paper, I tell them to ask three questions:

• Thermal:“Can you show me a thermal test with this unit sealed in a cabinet, not just the temperature range on the datasheet?”
• Power:“What does this unit do on a dirty 24V line with motor starts on it — and is reverse polarity protected?”
• Recovery & supply:“If it hangs at 3 a.m., does it recover on its own — and what’s your stock and RMA policy if it dies next year?”

Datasheets are honest about features and silent about behavior. The products that earn the “industrial” label — wide power input, aluminum thermal design, dual watchdog, pre-built software stack, guaranteed supply — are the ones whose answers to those three questions are more than a spec line.