What the Data for This Category Looks Like
Daily documents for this category include tooling requirement sheets, positioning datum schemes, fixture drawings, gauge operation manuals, acceptance reports, calibration records, and change histories. Data sources include product digital models and GD&T, process routes, coordinate measuring machine data, on-site trial feedback, and metrology systems. Documents contain fields such as positioning datum parameters, assembly dimensional tolerances, calibration cycles, and change version numbers. Units include millimeters and radians; calibration cycles follow project milestones. Data is updated according to project progress, triggered by each product change, and during regular calibration updates. The technical parameter accuracy requirements for individual documents are high.
Constraints Imposed on Knowledge Base Retrieval and Recall
The data characteristics of this category impose multiple constraints on retrieval and recall. Mixed document types (drawings, reports, calibration records) require retrieval to support filtering by document type, to avoid interference from irrelevant results. The presence of high-precision technical parameters such as GD&T tolerances and coordinate measuring machine coordinate values requires retrieval to balance semantic matching and exact field matching, to prevent low-precision results from causing clamping deviations. Frequent updates tied to projects and changes require the knowledge base to support incremental synchronization and version retrospective retrieval, to ensure that recalled results are the latest valid versions. The timeliness of on-site feedback and metrology data requires prioritizing recall of documents updated after the latest calibration and change.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category’s documents include high-precision tolerance parameters. A threshold in this range filters low-match irrelevant results and avoids clamping scheme deviations |
limit | Top 6 entries | Fixture process workflows rely on multiple associated documents such as fixture drawings, calibration records, and change histories. 6 entries cover core associated data without exceeding context length limits |
searchMode | mixedRecall | This category includes both semantic positioning scheme descriptions and exact numerical GD&T parameters. Hybrid retrieval balances semantic and exact matching |
usingReRank | Enabled | In mixed document scenarios, re-ranking improves the ranking priority of high-precision parameter matching results |
datasetSearchUsingExtensionQuery | Enabled | This category’s documents contain specialized terminology such as GD&T and coordinate measuring machine testing. Extended queries cover term variants and synonyms |
collectionIds | Collection IDs grouped by document type | This category’s documents are divided into drawing, report, and calibration categories. Filtering by collection narrows the retrieval scope and improves recall efficiency |
Three Common Configuration Mistakes
- Issue: After uploading files to the knowledge base in an offline environment, the system remains in a loading state for a long time, and normalcy resumes only after connecting to the internet. Cause: FastGPT v4.17.0 defaults to relying on external resources to pull index dependencies, and offline environments cannot complete initialization verification.
- Issue: When using semantic models such as bge-m3, bge-large to retrieve, target documents cannot be hit, but full-text retrieval returns results normally. Cause:
datasetSearchUsingExtensionQueryis not enabled, specialized terminology is not matched via extension, and thesimilaritythreshold is set too high, filtering exact matches with low semantic similarity. - Issue: Retrieval results include fixture documents not from the current project, and cannot accurately match the positioning datum scheme for the current process. Cause:
collectionIdsis not configured to filter the document collection for the specified project, so the recall scope covers the full knowledge base.
How to Verify Correct Configuration
- Review the document type proportion of retrieval results, confirm alignment with the document requirements for the current process, and adjust the
similaritythreshold until matching meets business accuracy requirements. - Test the knowledge base upload and retrieval workflow in an offline environment, confirm there is no prolonged loading state, and verify that offline dependency configurations are correct.
- Enter search terms containing specialized terminology such as GD&T and coordinate measuring machine, confirm that recalled results include documents with corresponding parameters, and verify that
datasetSearchUsingExtensionQueryis working. - Compare the number of results from full-text retrieval and semantic retrieval, confirm that the
limitconfiguration’s recall count fits within context length limits.
The values provided here are common starting points and should be measured against the reader's own samples.