What the Data for This Category Looks Like
Core data for precast concrete components comes from BIM detailed design, rebar and concrete batch ledgers, mold table and curing kiln operation records, laboratory test reports, and product QR code traceability systems. Daily generated documents include component detailed drawings, rebar cutting lists, mold inspection forms, pouring and vibration records, curing cycle records, and demolding and factory inspection reports. Data updates in real time based on individual components, each batch of materials, and each curing kiln batch. Individual documents contain fields with clear units such as component number, rebar specification, curing temperature, and strength test values, and are bound one-to-one with on-site lifting positions.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall?
Precast concrete component documents are mostly engineering documents with precise fields, and data updates in real time based on individual components, each batch of materials, and each curing kiln batch. This requires that knowledge base recall content must match clear fields such as component number, rebar specification, and curing parameters. The one-to-one binding between individual components and on-site lifting positions requires that retrieval results must not mix data across components. The full-process data associated with QR code traceability requires recalling multiple types of documents across mold inspections, pouring records, curing reports and other categories, so the retrieval scope must cover associated documents across all production links. At the same time, incorrect data will lead to component scrapping or construction delay, so the confidence of recall results must meet business requirements.
Configuration Settings
This configuration is based on FastGPT v4.17.2.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precise field requirements of engineering documents, avoids recalling irrelevant data from non-associated components |
limit | Top 8 entries | Covers multiple types of associated documents for a single batch of components, while controlling context length to avoid exceeding model limits |
searchMode | fullTextRecall | The query examples emphasize identifiers or specialist terms; test this mode with representative data, not as a product default. |
usingReRank | Enabled | Performs secondary sorting on recall results by business fields such as curing strength and component position, to improve result relevance |
datasetSearchUsingExtensionQuery | Enabled | Supplementally retrieves extended data such as batches and curing kilns associated with components, covering associated documents across all production links |
collectionIds | Configured by factory production grouping | Limits the retrieval scope to exclusive documents such as molds, curing, and inspections from the current factory, avoiding data mixing across factories |
Three Common Configuration Mistakes
- Importing WeChat Official Account links after configuring
datasetSearchUsingExtensionQueryfails. The cause is that the Official Account article has not been converted into structured content that conforms to engineering document formats, so the system cannot identify core fields such as component numbers and specifications. - Retrieval results only meet some specified conditions. The cause is that the
similaritythreshold is not set, or the set value is too low, which recalls data from other non-associated component batches, and fails to prioritize matching all required conditions. - A 418 error occurs after configuring the knowledge base. The cause is that
collectionIdsis configured with a non-existent dataset ID, or the production document collection of the current factory is not correctly associated.
How to Verify Correct Configuration
- Upload a single component detailed drawing to the knowledge base, retrieve the exclusive number of that component, and confirm that the recall results include associated documents such as rebar cutting lists and curing records.
- Adjust the
similaritythreshold to 0.8, retrieve a query that includes more than three specified business conditions, and confirm that the recall results meet all conditions. - View the association status of the knowledge base dataset, and confirm that
collectionIdsonly includes production-related document collections from the current factory, with no cross-factory data. - Test importing structured engineering documents such as mold inspection forms, and confirm that retrieval results can accurately match the component specification fields in the documents.
The values provided here are common starting points and should be evaluated against your own specific samples.