What the data for this category looks like
Business data for metal cutting machining comes from product drawings and 3D models, CAPP/CAM systems, MES, tool management systems, and measurement and inspection tool data. Daily documents include process routes, operation cards, CNC programs, first-piece inspection records, patrol inspection forms, and measuring tool calibration records. Data updates are performed with work orders, tool changes, or random inspections. Each document contains fields such as cutting parameters, tolerance zones, tool life, and calibration cycles, with units covering professional measurement standards including m/min, mm/r, μm, and others.
What constraints these characteristics impose on the "knowledge base retrieval and recall" link
The multi-source heterogeneous data format of metal cutting machining requires retrieval to adapt to different document types such as CNC programs and operation cards, to avoid parsing exceptions. Frequently updated business data requires real-time synchronization with the knowledge base to prevent recall of expired process parameters. The precise requirements for professional cutting parameters and unit fields require retrieval matching accuracy higher than that of general scenarios, to avoid recalling irrelevant content and ensure that measurement units are fully retained.
How to set the configuration
This configuration is based on FastGPT v4.17.0
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
similarity | 0.82–0.88 | Matches the precise parameter requirements of metal cutting processes, avoids recalling irrelevant operation content with low similarity |
limit | Top 3–6 entries | Single process documents contain large amounts of information, control the number of recalled entries to fit the context window |
usingReRank | Enabled | Optimizes ranking results for heterogeneous documents, prioritizes recalling core process specification content |
datasetSearchUsingExtensionQuery | Enabled | Supplements industry synonyms for machining processes, such as "feed rate" and "feed speed", to improve recall accuracy |
collectionIds | Knowledge base IDs bound to the workshop | Limits the retrieval scope, avoids recalling redundant documents across workshops |
Three common mistakes
- Phenomenon: Imported process documents display garbled text, with no such issue in local deployments. Cause: Text parsing rules for CNC program G-code and PDF-exported drawings are not configured, resulting in format errors during cloud parsing.
- Phenomenon: A large number of non-target documents are recalled when searching for process-related questions. Cause: The retrieval knowledge base scope is not limited via
collectionIds, or a reasonablesimilaritythreshold is not set, leading to recall of low-matching content. - Phenomenon: Unable to set a single retrieval result count of around 300. Cause: The platform's default
limitgear only provides fixed interval options, adjust the value via custom parameter configuration.
How to confirm the configuration is complete
- Upload one typical operation card document, verify that there is no garbled text after parsing, and that fields such as cutting parameters and tolerances are fully retained.
- Enter a question containing professional terminology, such as "Cutting speed range for turning 45 steel outer circles", check that all recalled content comes from the target workshop's knowledge base documents.
- Adjust the
similaritythreshold, verify that irrelevant content with low similarity is no longer recalled. - Check the number of results returned by a single retrieval, confirm that it matches the preset
limitvalue range.
The values provided here are common starting points, and should be evaluated against the reader's own samples.