What we can do.
Control and supervisory layer of the plant.
- Legacy PLC migrationPLC-5, SLC 500, S7-300 and platforms no longer sold.
- DCS upgradesCut over inside the window, rollback written first.
- SCADA and HMILicensing and screen hierarchy settled before purchase.
- Operator consolesConsoles and central control rooms, designed and built.
- Full handoverSource code, licences, as-built drawings, backups.
- Any brandAllen-Bradley, Siemens, Schneider, Mitsubishi.
Delivered by our team.
Recent work by our engineering team, by industry.
- AluminiumCentralised SCADA and temperature monitoring across a reduction line
- PowerPLC-5 migration on a captive power plant
- AluminiumRodding-shop hydraulic and control upgrade
- SemiconductorRO plant control and site support
- AluminiumMachine control console upgrade
Where it applies Aluminium smelterPower, balance of plantWater & wastewaterSemiconductor & E&E
From the field.
The other layers.
Start with the plant, not the proposal.
Send a photo of the panel, the drive or the problem. An engineer answers.
Technical notes for engineers
When a controller is actually end of life
Obsolescence is a procurement word, and it hides four different situations that need four different answers. Before anyone quotes you a processor, the platform needs grading.
| Stage | What it means | What it justifies |
|---|---|---|
| Active | Still sold, still supported | Nothing. Do not migrate. |
| Active mature | Sold, but the successor is announced | Budget planning. Start collecting the documentation you will need in three years. |
| End of life | No longer sold. Repair and support continue for a stated window | A migration plan with a date, and a spares bridge to cover the gap. |
| Discontinued | No repair, no support, grey market only | Migration now. Every month is borrowed. |
Families we see most often at this point in Malaysian plants: Allen-Bradley SLC 500 and PLC-5, early ControlLogix firmware that will not talk to anything modern, Siemens S5 and early S7-300, Mitsubishi A-series, Koyo and other small platforms that arrived with a packaged machine and were never on anyone's asset register.
A small platform often has no hot swap. Reseating a module on a live rack to diagnose a fault can raise a permanent I/O error by itself, and then you are troubleshooting the repair rather than the failure. Ask what the platform does on module removal before anyone opens a door. It is cheaper to ask than to buy a processor.
Like-for-like, or re-engineer
This is the decision that sets the cost and the risk, and it is usually made by accident.
Like-for-like reproduces the old logic on new hardware. It is faster, it is testable against known behaviour, and it is the right answer when the process is stable and the shutdown window is short. The cost is that you carry across twenty years of patches, dead branches and workarounds nobody can explain.
Re-engineer writes the control philosophy fresh from what the plant actually does now. It costs more up front and it needs process people in the room. It is the right answer when the original design has been overtaken by modifications, or when the migration is the excuse to fix something that has annoyed operations for a decade.
Automatic conversion tools sit uncomfortably between the two. They exist for most cross-vendor paths and they do produce running code. What they produce is code shaped like the source platform, with remapped tags and instruction emulation that no maintenance technician will ever enjoy reading. Used as a reference for what the old logic did, they save real time. Shipped as the deliverable, they hand you an unmaintainable system on new hardware, which is the worst of both outcomes.
The architecture decisions that cost money later
Redundancy is not a second processor in the same chassis
A redundant ControlLogix pair does not permit local I/O in the redundant chassis. So "add a second CPU for redundancy" is never the real bill of materials. The I/O has to move to a remote chassis on a network, which means network modules, a second remote chassis, cabling, and a rework of every I/O reference in the program. If a quotation shows you a processor and a redundancy module and nothing else, it has not been engineered, it has been priced.
Network segregation is decided now or never
A migration is the one moment the control network can be laid out properly. Once the plant is running on the new system, nobody will take another shutdown to separate an OT network from the office LAN. Decide the segmentation, the remote-access path and the patching responsibility during design, and write it into the handover.
Spares strategy is part of the migration, not a follow-up
There is always a gap between the decision to migrate and the cutover. That gap has to be covered by spares for the old platform, bought while they are still findable. Plan it deliberately: which modules, how many, and whether refurbished stock is acceptable for a system with a known retirement date.
How we run a migration
- Survey. Rack inventory, firmware revisions, I/O count and type, network topology, which drawings are current and which are fiction. Most plants discover here that the as-built and the panel disagree.
- Recover the logic. Upload and archive everything before anything is touched, including the HMI application, drive parameter sets and any recipe data. This is also the last chance to recover a program from a processor that may not power up again.
- Control philosophy. Written down, in plain language, and agreed by operations before a line of code is written. Interlocks, permissives, failure states, manual paths.
- Build and FAT. Panel built and the new system tested against simulated I/O, with operations present. Faults found here cost hours. Faults found on site cost a shutdown.
- Cutover. Sequenced against the shutdown window, with a written rollback point and a defined abort criterion.
- Handover. Source code, licences, as-built drawings, parameter backups and a documented restore procedure, in your hands, not ours.
"The plant can still run in manual" is an assumption that has to be tested, not stated. If the manual path runs through a PLC output, a processor in STOP kills manual too. Check the wiring, not the intent, before a migration plan leans on manual operation as the fallback.
The licensing questions to settle first
Every major SCADA platform licenses on a different axis, and the axis is what makes two quotations for the same plant differ wildly.
| Axis | What to check | Why it bites |
|---|---|---|
| Tag or I/O count | Whether the count is physical I/O, server tags, or every derived point | Alarm, calculated and historian tags can multiply a count several times over |
| Client count | How many simultaneous viewers, and whether web viewers are counted differently | A view-only web client is often free or cheap where a full client is not |
| Station vs server | Whether the product you are buying can serve other machines at all | Some station-class products are standalone by design. A second console is then a second licence, not a client |
| Redundancy | Whether the standby needs its own full licence | Doubles a line item that was presented as a single number |
| Historian | Retention, tag count, and whether the client tools are separate | Reporting is routinely quoted after the fact |
On one widely used platform, the station-class product cannot host networked clients at all. So a request for "a second screen in the control room" is not a client licence, it is a second station licence plus a second machine. On the same platform, read-only web clients are free. The difference between those two answers is the whole budget, and it is decided before anyone draws a graphic.
Some platforms publish no list price at all and quote only through a channel. That does not mean you cannot control the number. It means the specification has to pin the tag count, the client count and the redundancy model precisely enough that two quotations are comparable.
Screens that work at three in the morning
An operator screen has one job: answer the question the operator is holding, in one look, without hunting. Almost every bad HMI fails the same way, by putting everything on the top level so that nothing stands out.
The structure that survives a night shift is layered:
- Level 1 answers one question. Is the plant normal. Sparse, high contrast, no decoration, and an abnormal state is the only thing that draws the eye.
- Level 2 is the area or unit. Enough to localise a problem and see the interlock that is holding.
- Level 3 is the loop, the device, the trend, the parameter set. This is where detail belongs, and it is one click away, not on the overview.
Two rules that matter more than any style guide. Colour means state, never decoration, which implies a grey plant with colour reserved for abnormality. And alarms are rationalised before they are configured. An alarm nobody actions is worse than no alarm, because it trains the shift to clear the list without reading it.
Architecture, redundancy and the boring parts that keep it alive
Where the data actually lives
Decide early whether the historian is on site, at a group level, or both, and who is responsible for its backup. A historian nobody backs up is a reporting obligation waiting to fail an audit.
Redundancy that has been tested
Redundant servers that have never been failed over are not redundant, they are expensive. Failover gets tested at FAT, at commissioning, and then on a schedule, with the result written down.
Remote access, decided deliberately
Every plant ends up with remote access. The choice is whether it is designed, with a defined path, named accounts, logging and a change window, or whether it accumulates as a series of individual exceptions nobody has a list of.
The handover pack
Licences in the plant's name. Source projects, not just runtime. Server build documentation. A restore procedure that has been performed at least once, by someone other than the person who wrote it.
Questions we get asked
How long does a PLC migration take?
The engineering is usually months and the cutover is usually days. A single machine on a small platform can be surveyed, built and cut over inside a quarter. A plant-wide DCS migration is planned a year ahead because it is scheduled against annual shutdowns, and the constraint is almost never our capacity, it is the window the process will give you.
Can you migrate between brands, for example Allen-Bradley to Siemens?
Yes, and it is common when a plant is standardising. Expect the instruction set, the tag structure and the HMI to be genuinely rewritten rather than translated. Budget for that honestly rather than trusting a conversion tool to absorb it.
Do we have to replace the field devices and wiring too?
Usually not. Most migrations reuse the field wiring and terminate into new I/O, which is why marshalling and terminal strategy matter more than people expect. Devices get replaced where they are themselves obsolete or where the signal type is changing.
What happens if the cutover fails?
There is a written rollback point and an abort criterion agreed before the window opens. The old racks stay in place and re-terminable until the new system has run through a defined proving period. Any plan that removes the old system on day one is a plan without a reverse gear.
Do you hand over the source code?
Yes. Source code, licences, as-built drawings and backups belong to the plant. A control system you cannot open is a control system you do not own.
Can you work on an existing SCADA rather than replacing it?
Usually yes, and usually that is the right call. Extending, rationalising alarms, rebuilding a screen hierarchy or adding a historian to a healthy system costs a fraction of a replacement. Replacement is justified when the platform is out of support, the licensing model has become the constraint, or the server operating system can no longer be patched.
Do we need a second operator console?
Answer the licensing question before the hardware one. On some platforms a second console is a client licence, on others it is a second full station. On several, a read-only web viewer would have satisfied the actual requirement at a fraction of the cost.
How many tags do we actually need to license?
Count physical I/O first, then ask the vendor specifically whether derived, alarm and historian tags are counted against the same limit. That single question changes the band on most platforms.
Can the SCADA be reached from outside the plant?
It can, and it should be designed rather than improvised. Outbound-only data paths, a defined change window for anything going the other way, named accounts and full logging. Anything safety-critical must keep working with the link down.