Engineering notes on building, testing and integrating IEC 61850 and IEC 60870-5-104 systems. Written for people who have to make the thing work on site, not for keyword density.
Both options work in a lab. The difference shows up at type approval, on VxWorks, and the first time a third-party master does something unexpected. A practical comparison of commercial IEC 61850 stacks and open-source alternatives.
When an open-source IEC 61850 stack no longer fits — licensing, non-Linux targets, Edition 2 depth, conformance evidence — this is how to migrate without rewriting the application, and what to test before and after.
Most edition arguments are settled by the device's own SCL file. How to read version, revision and the Services block, what actually changes behaviour between Edition 1 and Edition 2, and how to run a mixed-edition station.
The SCL template-and-reference graph explained: DataTypeTemplates, LNodeType, DOType, DAType, LDevice and DOI/DAI resolution, plus the seven gotchas that break naive parsers — with a minimal walker you can adapt.
What an IEC 60870-5-104 conformance test exercises — startup, k/w windows, t0/t1/t2/t3 timers, general interrogation, command select/execute — and the five failures that show up on real links rather than in the lab.
How GOOSE retransmission actually works, why the first thing to break is usually the network rather than the stack, and how to test a GOOSE implementation so the lab result survives the substation.
How to map IEC 61850 data to Modbus registers so the point table survives a firmware update: model-driven vs list-driven mapping, carrying quality and timestamps into a protocol that has neither, and handling the control path safely.
A working MMS client is five steps: parse the model, associate, browse, enable reporting, handle indications. Where each step actually goes wrong, with an illustrative C skeleton and the reconnect logic nobody writes until production.
How to map real device data into IEC 61850 logical nodes: start from the CDC rather than the register list, choose LN classes deliberately, use GGIO sparingly, and treat the ICD as a versioned interface rather than a build artefact.
What the five control models mean in practice, how origin and test flags change behaviour, how command termination should be reported, and the four control defects that show up at site acceptance.
How report control blocks behave under load, what buffering really guarantees, how to size datasets and RCB counts, and the five reporting defects that appear when a link comes back after an outage.
How the file transfer service works, why large disturbance files fail at the master, how COMTRADE file pairs must be handled, and the operational rules that keep a 200 MB transfer from blocking your reports.
Why a correct value with a wrong timestamp is worse than no value, how SNTP and PTP differ in what they can promise, and how quality bits should propagate from the source to the control room.
A practical comparison of IEC 61850 and IEC 60870-5-104 for product designers: what each protocol is good at, why the choice is usually made by the customer rather than the engineer, and when you end up implementing both.
A short, practical explanation of the IEC 60870-5 family: 101 for serial telecontrol, 104 for the same application layer over TCP, and 103 for protection equipment — plus what changes when you move a product between them.
A conformance certificate says a defined test suite passed on a defined configuration. It does not say your product will interoperate on a specific site. Here is how to read one, and what to ask for instead of a PDF.
From treating the ICD as documentation to ignoring buffered report overflow: the seven integration mistakes we see repeatedly, what each one looks like in a capture, and the cheap check that catches it.
Licence scope, source delivery, support time zones, escalation when a customer site is down, and the four contract terms that matter more than price when the vendor is on the other side of the world.
Throughput tests pass in an hour and fail in a month. What to measure during a long run — memory, handles, sequence counters, timing drift — and how to build a soak test that finds the defect instead of just running for a long time.
Why IEC 61850 has three protocol stacks instead of one, what each carries, why they use different transports, and which one your device actually needs before you buy anything.
How sampled values are published and consumed, what a receiving device must do about loss, reordering and time alignment, and why SV is mostly a platform problem rather than a protocol problem.
How parallel redundancy protocols work, what duplicate handling means for a protocol stack, and why redundancy is mostly a network design question that the software must not undermine.
Three protocols that answer overlapping questions: what each does best, which markets expect which, and why the deciding factor is usually the customer rather than the technology.
What to capture, verify and sign off during site acceptance so that the protocol layer is not the thing discovered during the first incident: a checklist you can hand to a commissioning engineer.
What actually changes when a substation goes digital: merging units and the process bus, the station bus, where IEC 61850 fits, and which parts are a software problem versus a network design problem.
A one-to-two-week integration is a published figure, and this is what it is spent on: model design, map building, wire-level verification, and the parts that always take longer than expected.
How protection testing changes when CT/VT and interlocking arrive over Ethernet: injecting sampled values, testing GOOSE intertrip, isolating test traffic, and what to capture.
How to budget a gateway: point counts per southbound leg, event rates rather than point totals, the cost of protocol translation, and the measurements that tell you it is already too late.
The real differences between network-based and copper-based interlocking: fault behaviour, testing, isolation, cyber surface, and the questions a utility will ask before approving GOOSE in a protection path.
A field guide to 104 problems: links that drop under load, commands that hang, general interrogation storms, stale data and duplicate events — with the first thing to check for each.
Send it with the device, the platform and the constraint you are stuck on. We answer within one business day.