Troubleshooting
Why HMI Remote Access Fails While PLC Access Still Works
Short answer
Working local traffic between the HMI and the PLCs proves only the switch path, not the default-gateway path a translated remote connection needs. Verify the panel's gateway field, the forwarded ports 6080, 6002 and 6000, and address duplication, in that order; a DHCP laptop with no Internet in the same building points at the same gateway branch.
On this page
When the PLCs, the supervision PC and the cameras all answer from off-site but the HMI does not, the network is not the suspect — the panel's default-gateway path is. Traffic between devices on one subnet is switched by MAC address and never touches a router, so a panel with a blank or wrong gateway can read PLC data all day while every translated remote connection to it times out. A DHCP laptop in the same building with no Internet, wired or wireless, points at the same missing route. Take three readings in order: the panel's gateway field, the forwarded remote-access ports, and address ownership. One of them closes the case.
Local traffic and remote access take different paths #
The installation that frames this article: two buildings a few tens of metres apart, every device on a fixed address in 192.168.2.xx, one Internet router (a Livebox) in the first building, and two unmanaged switches joined by a direct copper link. From off-site, the PLCs answer through port redirects, the supervision PC answers through remote desktop, and the cameras answer through their clients — but a Magelis HMISTU panel in the second building refuses both of its remote clients, and a DHCP laptop in that building has no Internet on wire or Wi-Fi.
When the panel polls the PLC across the yard, both addresses fall on one network, the switches forward the frames, and the router is never involved. That is why everything communicates and remote access to the panel fails are not contradictory statements: the first never exercised the gateway.
An off-site client does exercise it. The client connects to the router's public address, the router translates the session to the panel, and the panel's reply is addressed to the client — off-subnet — so it must be handed to the panel's default gateway. A blank gateway field deletes that hand-off: the inbound packet can arrive, the reply cannot leave, the session times out. The same split with a different transport appears when HMI Modbus TCP works locally but not remotely over cellular.
The panel's three services and their ports, as documented for this installation: Webgate, the browser client, on TCP 6080; the Vijeo Design'air client on TCP 6002; and the data channel on TCP 6000. The PLCs are reached on forwarded TCP 502, the registered Modbus TCP port. Working 502 redirects prove the router translates; they say nothing about the panel's own gateway field. So the first reading is the panel's gateway field, not the switch configuration.
The readings that isolate it, in order #
- Read the panel's network screen: address, subnet mask and default gateway, written down. A blank gateway after static configuration is a documented failure mode on industrial equipment — one documented case needed the three static fields entered twice before they held — so re-enter them and confirm they survive a restart.
- Ping the real gateway from the panel's building, using the gateway address the DHCP laptop actually received, never an assumed router address. No answer is not proof of a dead path: a router can filter ICMP while passing every other protocol.
- Compare the laptop's lease in both buildings: same laptop, recording address, mask, gateway, DNS servers and DHCP server each time. A lease whose gateway or DNS changes between buildings moves the fault to the DHCP or segment branch before any panel setting is touched.
- Verify each forward on the router. TCP 502 to the PLCs already works, so the translation pattern is proven; TCP 6080, 6002 and 6000 must each point at the panel's exact address. One wrong final octet forwards to a device that answers nothing.
- Test from outside by IP address, not hostname, until the raw connection works; name resolution only adds a second fault domain on top of the first.
- Check who owns each address: sweep the range, then read the address-to-MAC mapping address by address. A ping sweep alone can miss a duplicate because both holders answer the same destination — the MAC behind the address is the evidence.
- Isolate branch by branch in an agreed window. Sweeps and cable pulls on a live control network can interrupt controller traffic, so schedule them and never pull the link the running control conversation depends on. Connect the switches one at a time and confirm one owner per address before adding the next.
- Only then spend time on VLAN or hidden-router theories. Here the physical inventory — one router, two unmanaged switches, one copper link — excluded that branch before it cost an evening.
Tracing the return path for a panel at 192.168.2.50 with a blank gateway #
The inputs: the panel sits at 192.168.2.50 with mask 255.255.255.0 and a blank gateway field — a typical case inside the installation's documented 192.168.2.xx fixed range, whose notation gives the range but not the mask, so read yours rather than inferring it. The PLC it polls is 192.168.2.20 in this example. The router's LAN address is 192.168.2.1, forwarding TCP 6080 to 192.168.2.50, and the client connects from a public address.
The rule: every IP stack makes one comparison before sending — destination AND mask against own address AND mask. Equal means same subnet, so the frame is delivered by MAC through the switches. Different means the packet is handed to the default gateway. A blank gateway removes the second branch.
The arithmetic:
- 192.168.2.50 AND 255.255.255.0 = 192.168.2.0 — the panel's network.
- 192.168.2.20 AND 255.255.255.0 = 192.168.2.0 — same network, so panel-to-PLC traffic is switched delivery. The gateway is never consulted and the panel keeps reading its tags.
- The client's public address lies outside 192.168.0.0/16, the private block this network lives in, so its masked network cannot equal 192.168.2.0 — different subnet, gateway required, gateway blank, reply dropped inside the panel.
- 192.168.2.1 AND 255.255.255.0 = 192.168.2.0 — the router is on the panel's own subnet, so writing it into the gateway field is valid.
- Mask check with the real mask: had the screen shown 255.255.255.240, then 192.168.2.50 AND 255.255.255.240 = 192.168.2.48 while 192.168.2.1 AND 255.255.255.240 = 192.168.2.0 — the gateway itself lands on a foreign network and even its ARP fails. Then the mask is the fault, not the gateway value.
The result: local traffic is unaffected, every translated inbound session on 6080, 6002 and 6000 times out, and a DHCP laptop holding an equally dead gateway shows no Internet — two symptoms, one missing route.
The decision: a blank gateway, or one that differs from the gateway in the laptop's lease, is the fault. Enter the verified router address, restart the panel and retest 6080 from outside. If the field already reads correctly, move to address ownership without touching the router.
What each reading tells you #
| Reading from the panel's building | Meaning | Next check |
|---|---|---|
| Panel and PLC answer locally; the gateway does not answer | The local path works and the gateway path is unproven | Who owns the gateway address, its MAC entry, the copper link, router filtering |
| The gateway answers; a test to an external IP fails | Outbound routing or policy is broken | Router WAN state and access rules |
| The external IP works; the hostname fails | Name resolution is the fault | DNS servers in the lease and their reachability |
| The laptop browses; the panel stays unreachable | The laptop branch is closed | Panel gateway, panel services, the three forwards |
| One address shows two MAC addresses over time | Duplicate address confirmed | Renumber; one owner per address |
The duplicate-address and flood branch #
A second installation with the same signature — devices on one switch unreachable until a different switch was disconnected — closed on two findings: duplicate addresses handed out from a wrong address plan, and an embedded system flooding the network with retry frames because its peer never answered. The method was sequential: switches connected one at a time, a scripted ping sweep across the whole range, every address listed per switch, then renumbering. Duplicate addressing in industrial plants has two documented cures, both administrative: give every address a single owner, or separate the duplicated ranges behind routing.
The classic warning is an endpoint reporting that another device is trying to use its address; panels and controllers often stay silent, which is why the MAC evidence outranks the ping. The same conflict seen from the client side is why a PC fails to reach an HMI when Wi-Fi owns its address. Every step in the list above is built so one test proves one cause, the discipline behind industrial troubleshooting that improves when each test proves a cause.
Panel-side settings, and where this case stopped #
Two panel settings can mimic a network fault. Some panel lines expose an IP routing toggle — Red Lion documents that IP routing must be disabled for direct PC-to-panel access on isolated networks, and a panel that routes when it should not will answer on paths you never planned. And the services behind TCP 6080 and 6002 must be enabled and running on the panel itself, because a forward to a stopped service times out exactly like a routing fault.
Once access works, do not leave three forwards open to the public Internet; remote HMI access with dynamic DNS and a VPN is the hardened design, and the direct forwards can then close.
In the case that motivated this article, the investigation stopped before the panel's gateway reading and the two-building lease comparison were taken — the two readings that separate every branch above. With the switch and hidden-router branches excluded by physical inventory, a fault with this signature lives in the panel's gateway field, the three forwards, or a duplicate address. Take them in that order.
Questions asked
Why does the laptop in the same building lose Internet if the panel is the faulty device?
The two faults share a branch, not a cause. The laptop's lease can carry a dead gateway or DNS from the same segment, or the laptop itself can block the path with a second active interface, a manual proxy or a stale static route. Test with every other interface disabled and compare the lease fields building by building. If the laptop then browses while the panel stays unreachable, close the laptop branch and stay on the panel gateway, the three ports and duplicates.
Which ports have to reach the panel for remote access to work?
Webgate on TCP 6080, the Vijeo Design'air client on TCP 6002, and the data channel on TCP 6000, each forwarded on the router to the panel's exact address. The PLCs in this installation answer on forwarded TCP 502, the registered Modbus TCP port, so the router's forwarding pattern is proven; replicate it for the three panel ports before suspecting anything deeper.
Can a duplicate IP address really leave local traffic working while remote access to one device dies?
Yes. Inbound connections arrive through the router, whose address-resolution cache can bind the panel's address to the other holder's MAC, so forwarded traffic lands on the wrong device while local peers that resolved the real MAC keep communicating normally. An address that alternates between two MAC addresses over time, or becomes reachable the moment the other device powers down, confirms the duplicate. Renumber so each address has one owner.
The gateway answers ping but the panel still times out from outside. What is next?
Confirm the services behind TCP 6080 and 6002 are enabled and running on the panel itself, confirm each forward targets the panel's exact address rather than a neighbour's, and retest from a phone on cellular data to rule out inside-the-router translation behaviour. If the session still fails with a verified gateway and verified forwards, the duplicate-address check is the remaining branch.
Should these port forwards stay open once remote access works?
No; prefer a VPN. Three forwarded ports put a web-facing operator panel on the public Internet, where ageing panel firmware is an exposed surface. Tunnel remote access through a VPN and close the direct forwards after testing; the panel then stays reachable through dynamic address changes without being publicly addressed.