What This Category of Data Entails
Data for this category comes from product digital models, GD&T specifications, process route documents, coordinate measuring machine data, on-site trial feedback, and metrology system records. Daily documents include fixture requirement documents, positioning datum schemes, fixture design drawings, gauge operation guides, acceptance reports, calibration records, and change histories. Updates are carried out per individual project, triggered by product changes or scheduled calibrations. Documents contain fields such as GD&T tolerance parameters, positioning and clamping schemes, measurement thresholds, and calibration cycles. Common engineering units are used, including millimeters, newtons, and calibration cycle days.
Constraints Imposed on Knowledge Base Retrieval and Recall
Data for this category updates per individual project or product change, and includes drawings, numeric parameters, and calibration records. This requires knowledge base retrieval to precisely match document versions linked to the target project, to avoid calling expired calibration records or pre-change schemes. Documents include precise numeric values such as GD&T tolerances and coordinate measuring machine measurement thresholds. Retrieval logic must support linking these numeric values to drawings. A single retrieval must cover multiple document types: positioning datum, interference verification, GRR acceptance, and others. Non-current project fixture data must be excluded to prevent incorrect recall related to clamping deviation or gauge misjudgment.
Configuration Settings
These configurations are based on FastGPT v4.17.0, as detailed in the table below:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match documents with precise numeric values such as GD&T tolerances and coordinate measuring machine thresholds, and avoid recalling irrelevant general fixture schemes |
limit | Top 6 results | Cover multiple document types including positioning datum, interference verification, and GRR acceptance, while controlling context length to meet business requirements |
searchMode | embedding | Use semantic/vector retrieval. Exact-field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Reorder recall results by relevance to the current review task, prioritizing critical documents such as positioning datum and change histories |
collectionIds | Bind exclusive collections per project | Isolate fixture and gauge data across different projects, preventing incorrect recall of cross-project documents |
datasetSearchUsingExtensionQuery | Disabled | Prevent automatic query expansion from introducing non-project fixture documents, ensuring precise retrieval scope |
Three Common Misconfigurations
- Retrieval results include non-current project fixture documents. Cause: Failed to bind exclusive project collections via
collectionIds, leading to retrieval covering the entire knowledge base. - Embedded drawing images in the knowledge base fail to load properly. Cause: Failed to configure multi-modal document storage path synchronization, or image resources were not deployed alongside the knowledge base.
- Extra complementary content is generated even after the question optimization switch is turned off. Cause: Failed to synchronously disable
datasetSearchUsingExtensionQuery, so the system still automatically expands queries and supplements content outside the retrieval scope.
How to Verify Proper Configuration
- Retrieve the fixture requirement document for a specified project, confirm that returned results only include documents for the current project, with no cross-project data.
- Upload a fixture design drawing, run a retrieval, confirm that the image loads normally and is linked to the corresponding retrieval result.
- Manually disable question optimization and
datasetSearchUsingExtensionQuery, initiate a retrieval, confirm that returned results only include content from the knowledge base, with no extra AI-completed information. - Clear the test knowledge base, initiate a retrieval request, confirm that the workflow skips the retrieval step and proceeds directly to the preset branch.
The values provided here are common starting points, and should be evaluated against the reader's own samples.