What This Category’s Data Looks Like
Business data for sheet metal cutting and bending scenarios comes from CAD/CAM software, order BOM, sheet material inventory systems, laser cutting machine and bending machine equipment logs, and MES production management systems. Daily documents include unfolded drawings, CNC cutting programs, bending process cards, nesting diagrams, first-piece records, and process flow cards. Fields cover sheet material grade, thickness, bending coefficient, cutting size, process number, and similar items. Common units are millimeters, kilograms, and square meters. Data updates trigger with orders, daily production, or equipment model changes, with no fixed cycle. Each order uses exclusive process documents.
Constraints on Knowledge Base Retrieval and Recall
The multi-variety, small-batch business characteristics of this category require knowledge base retrieval to accurately match exclusive process parameters for individual orders, preventing irrelevant document recall. Frequent on-demand updates require retrieval systems to quickly respond to newly uploaded process documents without full reindexing. Documents contain large numbers of numerical process parameters, so retrieval must prioritize exact field matching to avoid recalling incorrect bending coefficients or cutting sizes via fuzzy matching. The number of process documents linked to a single order is limited, so excessive recall results increase retrieval time and context processing pressure.
Configuration Settings
This configuration is based on the knowledge base retrieval function of FastGPT v4.17.0.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Sheet metal process parameters have high accuracy requirements. This range filters low-matching irrelevant documents, preventing incorrect bending coefficients or cutting sizes from being recalled |
limit | Top 3–5 results | The number of process documents associated with a single order is limited. Limiting the number of recalled results reduces vector matching time and avoids redundant data interfering with context processing |
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 filtering on initially recalled results to further match current order BOM and sheet material inventory information, improving result accuracy |
datasetSearchUsingExtensionQuery | Enabled | Uses sheet material grade and thickness from order BOM as extended query terms to supplement retrieval of related documents such as nesting diagrams and cutting programs |
collectionIds | Exclusive sheet metal process collection ID | Limits the retrieval scope, excluding process documents from other categories and reducing interference from irrelevant data |
Three Common Mistakes
- Symptom: Knowledge base retrieval takes 4–5 seconds, exceeding business response requirements. Cause: The retrieval scope is not limited via
collectionIds, resulting in recall of all process library documents, increasing vector matching and result processing time. - Symptom: Multiple consecutive bending parameter rows in an imported XLSX bending process card are split into multiple independent chunks, making it impossible to fully match process parameters during retrieval. Cause: No applicable chunking rule is set for structured tables, and the default character-based chunking strategy splits consecutive parameter rows, destroying parameter integrity.
- Symptom: Bending coefficients that do not match the current order's sheet thickness appear in retrieval results. Cause: The
similaritythreshold is set too low, recalling low-matching documents that do not meet similarity standards, leading to incorrect process parameters.
How to Confirm Proper Configuration
- Review FastGPT retrieval logs to confirm that the number of documents recalled per single retrieval matches the preset value of
limit. - Import a test XLSX bending process card, and check that the chunked content fully retains consecutive multi-line parameter information.
- Enter a query containing specific sheet thickness and bending angle, and confirm that all recalled documents match these parameters, with no irrelevant process documents from other categories.
- Test retrieval response times across multiple attempts, confirming compliance with expected business time standards.
Note: The values provided here are common starting points, and should be measured against the reader’s own samples.