Data Profile for This Category
Data for precast concrete components originates from BIM detailed design, rebar and concrete batch ledgers, mold table and curing kiln operation records, laboratory test reports, and outgoing QR codes. Daily documentation includes component detailed drawings, rebar cutting lists, mold inspection records, pouring and vibration logs, curing cycle records, and demolding and factory inspection reports.
Data is updated per single component, per raw material batch, and per curing kiln batch. Fields include component ID, raw material batch number, curing duration, strength test values, and similar metrics. Unlike fluid ready-mixed concrete delivered by truck, this data must correspond to specific on-site lifting positions.
Constraints for Source Citation and Traceability
The requirement that each single component maps to an on-site lifting position means traceability must be bound to a unique component ID. It cannot rely solely on raw material batch associations.
Multi-source data is scattered across design, production, and testing systems. Cross-data source matching of components, raw materials, and test records is required.
The high-frequency update rhythm for single components, raw material batches, and curing kiln batches demands indexing strategies adapted to real-time or near-real-time synchronization.
Diverse document types (drawings, ledgers, logs) need tailored text extraction and chunking rules. This ensures accurate matching to records from the correct production link during citation.
Severe errors such as embedded part misalignment and insufficient concrete strength require traceability results to clearly link to specific production links and responsible person records.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Inspection records for precast concrete components have long single-segment text, and need to cover complete inspection items and batch information |
responseData | Enabled | Traceability information such as component ID and raw material batch number must be displayed synchronously to meet cross-check requirements during audits |
detail | Show first 300 characters of source document | Core information of inspection records is concentrated at the start, allowing quick location of corresponding production links |
| Single recall limit | Top 6 entries | Inspection and process records for a single component are scattered across 3-5 documents. Too many recalls will increase audit burden |
| Variable reference mode | {{}} format | Dynamic knowledge base parameters need to be passed via API, compatible with the fixed features of FastGPT v4.17.0 and later versions |
Three Common Misconfigurations
- Phenomenon: Returned citation list is empty or only includes irrelevant documents. Cause: Unique component ID is not bound as a retrieval keyword, and only general question retrieval is used, which cannot match exclusive records for the target component.
- Phenomenon: Knowledge base search node in the workflow cannot read dynamically passed knowledge base variables. Cause:
{{}}format is not used for variable reference, or the FastGPT version does not support this format. - Phenomenon: Citation content is truncated, making complete inspection items unviewable. Cause:
quoteMaxTokenis set too small, failing to cover the full text length of a single inspection record.
How to Verify Correct Configuration
- Submit a retrieval request that includes the target component ID, and check whether returned results include inspection and process record documents for the corresponding component.
- View the citation detail module, and confirm that key fields and content fragments of the source document are displayed.
- Test passing dynamic knowledge base variables via API, and confirm that the knowledge base search node in the workflow can correctly read variable values.
- Adjust the
quoteMaxTokenparameter, and verify whether the returned citation text length meets current audit requirements.
The values provided are common starting points and should be measured against the reader's own samples.