What Data for This Category Looks Like
Data sources include CAD/CAM software, order BOM systems, sheet metal inventory ledgers, laser cutting machines, bending equipment, and MES systems. Daily generated documents include unfolded drawings, CNC cutting programs, bending process cards, nesting diagrams, first-piece records, and process transfer cards. Unfolded drawings mark bending coefficients, hole position coordinates, and external contours. CNC cutting programs contain cutting path parameters. Bending process cards mark bending sequences and angles. Data updates trigger per order, daily, or each time equipment undergoes a changeover.
Constraints on Knowledge Base Retrieval and Recall
Documents for sheet metal cutting and bending are mostly structured or semi-structured engineering content. They contain many professional terms and high-precision parameters. Retrieval must accurately match process details. Frequent update frequency requires the knowledge base to support incremental synchronization and rapid updates. The multi-variety, small-batch business model disperses data across different orders and changeover batches. Retrieval result quantity must be limited to avoid context redundancy. Units for engineering fields differ from general semantic retrieval matching logic. Retrieval queries must be optimized for this scenario.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Sheet metal process parameters have high precision requirements. Filter low-match irrelevant documents to avoid introducing incorrect parameters during audits. |
limit | Top 6 entries | In multi-variety, small-batch scenarios, too many single retrieval results increase context redundancy and reduce audit efficiency. |
searchMode | embedding | Adapts to professional terms and structured parameters of sheet metal processes, and matches core process content in documents. |
usingReRank | Enabled | FastGPT v4.17.0 supports full functionality calls for this configuration item. It performs secondary ranking on initial recall results, filters interference items, and improves the display priority of core process documents. |
datasetSearchUsingExtensionQuery | Enabled | Generates extended retrieval queries for professional terms related to sheet metal cutting and bending, covering high-frequency search fields such as bending coefficients and unfolding compensation. |
collectionIds | Bind the ID of the exclusive sheet metal cutting and bending process library | Isolate knowledge base data from different categories to avoid cross-category documents interfering with retrieval results. |
Three Common Mistakes
- Symptom: No matching results appear in knowledge base search tests, and the interface displays "No relevant results". Cause:
datasetSearchUsingExtensionQueryis not enabled, and retrieval queries are not optimized for professional terms related to sheet metal cutting and bending. As a result, semantic matching fails to cover core process fields. - Symptom: Some files show training exceptions, and duplicate references remain after re-uploading. Cause:
collectionIdsis not specified to bind the exclusive process knowledge base, causing mixing with documents from other categories. The system cannot automatically identify and deduplicate. - Symptom: Audit lag occurs due to overly long context, and the number of returned results exceeds expectations. Cause: The
limitvalue is set too high, and result filtering is not performed withusingReRank, introducing a large number of non-core scrap management documents.
How to Confirm Correct Configuration
- Upload one standard sheet metal unfolded drawing document, initiate a retrieval query containing "bending coefficient compensation", and verify that the returned results include the document's metadata.
- Check the knowledge base training logs to confirm that all uploaded cutting programs and process cards show successful training with no abnormal markers.
- Adjust the
similaritythreshold to verify that low-match irrelevant documents are filtered out, meeting preset precision requirements. - Test combined multi-query retrieval to confirm that extended queries generated by
datasetSearchUsingExtensionQuerycover professional engineering terms.
The values provided here are common starting points, and should be evaluated against the reader's own samples.