A portable industrial PC from the 1990s is still in service today: the 386 control board on its ISA backplane was swapped for a modern half-length ISA single-board computer, the floppy bay now holds a CF card used as the system disk, and a stripped-down Linux without a graphical desktop runs on it. The case, keyboard, and CRT screen stayed where they were. Over three decades, what got replaced was the computing part, not the machine itself. Replacing an obsolete industrial PC on an automation line is the same job: the goal is to change the middle layer while leaving PLC programs and field wiring untouched.
ARBOR's AGV and AMR pages revolve around Jetson AGX Orin, GMSL cameras, and Hailo-8 and MemryX accelerators — the story is a closed loop from perception to action. Sintrones offers 14th-generation Core processors with discrete GPUs, up to eight PoE Ethernet ports, 5G and Wi-Fi, and a one-sided I/O layout. TSL Automation's nine-point checklist runs from 9–36 VDC input, -20 to 60 °C (–40 to 80 °C for cold storage), and IEC 60068-2-6/-2-27 vibration and shock ratings through EtherCAT segmentation, TPM 2.0, and a 5–7 year availability commitment. All three answer how to choose new hardware. None of them answers how to take the old machine out, and how to do it without touching the PLC.
That last question is where retrofit projects actually spend money. The emphasis differs as well: competitors compare compute power and the number of ports, while a replacement project first has to confirm whether interface types match, whether communication parameters can be copied across, and which operating system the legacy software needs. When shortlisting an arm industrial pc, those three questions rank ahead of clock speed.
Computing work on a machine splits into three layers. The bottom layer is the PLC and field instruments, running control logic, and it stays. The middle layer is the industrial PC, handling acquisition, forwarding, local display, and uplinks — this is the layer to replace. The top layer is the host system, reading values against a tag list. The test for a successful swap is simple: afterwards, no PLC program has to be downloaded again, no instrument address has to change, and no host tag list has to be rebuilt. One more point is easy to miss: if the host system reaches the machine by IP address, giving the new unit the address the old one used keeps the host configuration untouched.
First, wiring and terminals. The RS-485 bus, termination resistors, and shield grounding on the field side stay as they are, and the new machine must cover every serial port the old one was using. USR-EG628 provides two RS-485 ports, one RS-232, and one CAN. USR-EG528 offers four serial ports, two fixed RS-485 and two switchable between RS-485 and RS-232, so a change in wiring order does not force another hardware change.
Second, the four serial parameters plus station addresses. Baud rate, data bits, parity, and stop bits, together with Modbus slave addresses and register mapping, get copied item by item from the old machine to the new one. Most of the "no data after the swap" rework traces back to this step.
Third, protocol and tag list. Whether the host reads over Modbus TCP, MQTT, or a proprietary TCP socket, tag names and addresses stay identical and only the IP and port change. USR-EG528 ships with Ubuntu 24.04 and WukongEdge, with built-in local SCADA/HMI configuration and Node-RED, so existing tags can be entered directly instead of writing a new acquisition program.
Fourth, where the logic lives. If logic sits in the PLC, replacing the computer only affects acquisition and display, which carries the least risk. If logic runs in software on the old machine, it moves across with everything else. PUSR ARM models share the same Ubuntu, WukongEdge, and Node-RED layer, so applications and flows transfer between models without rewriting.
Run in parallel. Mount the new machine alongside the old one and read the same bus, listening only and sending no queries — an RS-485 line accepts receive-only nodes. Let it run for a few hours or a full shift, comparing values and refresh cycles point by point, and watching reconnect behaviour and alarms as well, not just normal readings.
Cut over. Photograph the terminal order and wire numbers before touching anything; then stop host writes, move the wiring, power up the new unit, check every point against the tag list, and restore writes. Keep the old machine powered and in place rather than removing it immediately.
Fall back. Archive the old machine's system image and configuration files before the swap, so the original wiring can be restored if something goes wrong. USR-EG528 has an external TF card slot with system and data storage kept separate, so a saved image can be flashed back. Later parameter tweaks go through PUSR cloud configuration and remote management instead of another site visit for one setting.
When the host software exists only as a Windows build and cannot be modified, the target stays x86: USR-EC300 ships with Windows (Linux optional), i5-7200U, 8 GB of memory, a 128 GB wide-temperature SSD, six serial ports of which two switch between RS-232/RS-485/RS-422 and four are fixed RS-232, dual gigabit Ethernet, DC 12–36 V, -20 to 60 °C, and a 255-level programmable watchdog. ARM models draw less power and take less space but cannot run Windows-only software, so they get ruled out during evaluation.
Component headroom matters too, since these units stay powered for years: USR-EG528 takes DC 9–36 V with reverse polarity protection, -25 to 75 °C, an aluminium housing, and both hardware and software watchdogs. USR-EG628 supports 4G/5G, Ethernet, and Wi-Fi as mutual backups with built-in routing, VPN, and firewall, so a change in the uplink method means changing the link only and leaving the host in place.
The risk in a replacement lies less in the performance of the new machine than in the parameters nobody wrote down. Freeze wiring, the four serial parameters, the tag list, and the location of the logic into a checklist, then work through parallel run, cutover, and fallback, and both the PLC and field wiring stay as they were, with downtime held inside a single shift. Read the other way round, that is also the selection order for an arm industrial pc: check whether the interfaces and protocols match first, and whether the compute is enough second.
