Skip to content
realvirtual.io realvirtual v6.5.4

AI Diagnosis

AI Diagnosis helps an operator investigate an alarm using the documentation delivered for that machine. It retrieves relevant passages from manuals and service knowledge before writing an answer. The answer can present a likely cause, a remedy and links to the supporting PDF pages.

The diagnosis can include project context such as the fault code and current machine context. It is an advisory aid: verify the cited source and follow the machine’s approved safety and service procedures before taking action.

How the diagnosis is grounded

Section titled “How the diagnosis is grounded”
  • Machine documents provide the searchable manuals, datasheets and service information.
  • Fault context identifies the active error or PLC signal associated with the request.
  • Retrieved passages ground the answer in relevant source text instead of relying on a general response.
  • Source links open the supporting document at the cited page when that reference is available.
  • Operator comments are stored for service history but are not sent to the language model.

In a CONNECT-based deployment, PDFs are split and the search index (vectors) is stored on the CONNECT host. Building the index and processing every question sends text to the configured embedding endpoint: by default the EU-hosted Requesty cloud gateway, or a local OpenAI-compatible endpoint you configure. Answering sends the retrieved passages, your question and, unless you switch off MachineContext, the selected part’s context (component type, configured values, signal values, active alarms and node notes) to the configured language model: an EU-hosted service or a local OpenAI-compatible endpoint. The API key stays on the gateway. See AI diagnosis in CONNECT.

Asking about the part in front of you

Section titled “Asking about the part in front of you”

Select a part before asking a question to include its information in the request:

  • Documents: the search prioritises manuals linked to the part and its parents to find the relevant document and chapter.
  • Current state: the request includes the component type, signal values and active alarms, so the answer can address the current state.
  • Earlier findings: any note left on the part is also included.

The dialog shows a context chip naming the included part and an expandable line showing exactly what was sent. Removing the chip re-runs the question without that part’s context. Use this when the selected part turns out not to be at fault.

Each answer also lists affected parts: the parts of the model whose documentation matches the cited sources. Clicking one selects and focuses it in 3D, so a cited paragraph leads you to the hardware.

Your installation may have live machine state switched off for privacy. The documentation search then works exactly as described; only the current-state part of the question is left out, and the context line shows that.

Quick Start

Section titled “Quick Start”
  1. Open an alarm or diagnostic entry that offers Ask AI.
  2. Start the diagnosis and wait for the document search and response to complete.
  3. Read the proposed cause and remedy together with the current fault context.
  4. Open the listed source and verify the cited PDF page before following a recommendation.
  5. Add an operator comment when the deployment provides shared history and the note will help future service work.

See also

Section titled “See also”