Data Profile for This Category
Data sources for this category include product drawings and 3D models, CAPP/CAM systems, machine tool collection data, tool management systems, MES systems, and inspection tool records. Daily documents cover process routes, operation cards, tool lists, CNC programs, first-piece inspection records, patrol inspection forms, and measuring tool calibration records. Data updates are triggered by work orders, single-piece production, spot checks, or tool changes. Fields include cutting speed, feed rate, tool life, dimensional tolerances, and more. Common units are revolutions per minute, millimeters, and hours.
Constraints Imposed on Knowledge Base Retrieval and Recall
Data in this category comes from multiple sources and updates frequently. This requires knowledge base retrieval to support joint recall across multiple datasets, while adapting to event-triggered incremental synchronization mechanisms. Documents include multiple content types such as CNC program code, process parameter text, and inspection records. Adjust chunking and indexing strategies for different document formats. Fields have clear attached units. Perform unit matching during retrieval to avoid parameter confusion. The update frequency at the single-piece or work order level requires recall results to align with the specific process requirements of the current work order, rather than using generalized matching.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category has high parameter accuracy requirements. Use a threshold to filter low-match results and avoid mixing unrelated process parameters |
limit | Top 8 entries | The volume of process-related data for a single work order is limited. Excessive recall will exceed the context window |
searchMode | embedding | Match process intent and parameter associations. Keyword-only matching cannot cover professional scenarios |
usingReRank | Enabled | Filter results that are semantically similar but do not match process requirements, improving recall accuracy |
datasetSearchUsingExtensionQuery | Enabled | Supplement and expand process-related professional terms to adapt to cutting machining scenario needs |
collectionIds | Configured by workshop/supplier grouping | Precisely recall process data for the corresponding supplier or workshop, avoiding cross-scenario confusion |
Common Configuration Mistakes
- Retrieval results include a large number of process parameters unrelated to the current work order. The cause is failure to configure the
similaritythreshold or setting the threshold too low, which does not filter low-match unrelated content. - Knowledge base updates lag, causing retrieval results to not match actual production. The cause is failure to configure synchronization rules triggered by tool changes or work orders, and only using fixed-cycle synchronization.
- CNC program retrieval results appear truncated or garbled. The cause is failure to adjust the chunking strategy for CNC program code formats, using generic chunking rules directly which leads to incomplete code fragments.
How to Verify Correct Configuration
- Manually enter the process parameter keywords for the current work order, and check whether the
similarityscores of the recall results fall within the preset 0.75–0.85 range. - Trigger an event synchronization for a tool change, check the dataset update log in FastGPT v4.17.0, and confirm that the corresponding tool data has completed synchronization.
- Switch the dataset IDs of different suppliers, verify that the recall results only include process documents for the corresponding supplier, with no cross-supplier data mixed in.
- Enter a query that includes units, such as "operation card with 1500 revolutions per minute speed", and confirm that the recall results match the unit parameters.
The values provided are common starting points and should be measured against the reader's own samples.