Engineering Work

Industrial Signal Mapping & Machine Semantics

Technology Stack

  • Python
  • PLC
  • Mitsubishi
  • ClickHouse
  • YAML
  • HMI
  • Electrical Schematics
  • RAGFlow

Problem

The machine exposed thousands of active PLC and register signals, but most lacked usable names or direct links to physical nodes. PLC source logic was unavailable, so semantics had to be derived from evidence rather than guessed.

What I built

  • Consolidated collector catalogs, YAML registries, HMI references and electrical schematics into a working signal inventory.
  • Joined candidate registers with historical ClickHouse telemetry and checked whether signals had stable, meaningful histories.
  • Ran read-only smoke tests and field validation for disputed mappings, then recorded confirmed links in semantic registries.

Approach

For each machine node, I reduced a broad address list into a smaller set of reference signals, compared transitions against real operations and documented confidence and unresolved ambiguity. Technical documentation was also indexed for assisted search through RAGFlow.

Scale

The consolidated collector catalog contained approximately 7,271 active signals spanning multiple functional groups and machine nodes.

Result

  • Turned low-level addresses into a maintainable map of machine components and process meaning.
  • Created a stronger semantic foundation for monitoring, feature selection and later ML investigations without claiming access to PLC source code.