This page is compatible with FastGPT v4.17.0
What This Category of Data Looks Like
Technical documents for metal cutting machining include process routes, operation cards, tool lists, CNC programs, first-piece inspection records, patrol inspection forms, measuring tool calibration records, and more. Data sources cover product drawings and 3D models, CAPP/CAM systems, machine tool collection data, tool management systems, MES systems, and measuring and inspection tool data. Document updates are triggered by work orders, single-piece machining, spot checks, or tool changes. Each document includes fields such as process number, tool number, dimensional tolerances, and feed parameters. Units include machining-specific measurement standards such as millimeters, revolutions per minute, and millimeters per tooth.
What Constraints Do These Characteristics Impose on the Citation Sources and Traceability Link
Multiple heterogeneous data sources require citation traceability to link unique identifiers from each system. This prevents confusion between identically named fields across different documents. The high-frequency update rhythm requires the traceability link to pull the latest tool wear and measuring tool calibration data in real time. This ensures review references use currently valid versions. The existence of long-text documents such as CNC programs and operation cards requires the traceability link to adapt to large-token text splicing. It must also avoid truncating critical dimensional parameters. The business characteristic with severe error consequences requires traceability to accurately bind document versions and processing work orders. This prevents the use of expired or incorrect data from causing workpiece scrap.
How to Configure the Settings
| Configuration Item | Recommended Setting | Rationale for This Setting |
|---|---|---|
quoteMaxToken | 8000–12000 token | The length of individual chunks of CNC programs and operation cards for metal cutting machining typically falls within this range. This avoids truncating core review content such as critical dimensional tolerances and feed parameters. |
responseData | Full source data | Calibration dates for measuring tool calibration records, tool lifespan, machine tool operating parameters, and other non-text fields need to be traced. This ensures complete traceability information. |
detail | Enabled | Baseline selection instructions and clamping scheme details for operation cards must be displayed. This meets the detailed verification requirements for cross-review. |
| Single search limit | Top 20 entries | The number of technical documents associated with a single work order typically ranges from 15 to 25. This covers core review references such as processes, tools, and inspections. |
| Maximum context | 15000 token | This adapts to the total token count after splicing multiple operation cards and CNC programs. It meets scenario requirements for long-text cross-review. |
| Reordering rule | Sort by source document update time | This prioritizes retrieval and citation of the latest version of documents. Examples include patrol inspection forms after tool changes and the latest version of process routes. This ensures valid review references. |
Three Common Mistakes
- Phenomenon: The model context token count displayed on the interface does not match the actual source document token count passed to the AI. Cause:
quoteMaxTokenis not configured correctly. This causes some long document chunks to be truncated early, with only partial content passed through. - Phenomenon:
quoteMaxTokenis set to 1500, but document chunks in the knowledge base that exceed this length are still cited. Cause: The document chunk length is not set to match the range specified forquoteMaxToken. This causes individual chunks that exceed the limit to be passed in their entirety. - Phenomenon: Knowledge base citation data processed by HTTP requests cannot be recognized by the AI. Cause:
responseDatais not configured to the standard format. Unique identifier fields for source documents are missing.
How to Confirm the Configuration Is Correct
- View the preview interface after knowledge base chunking. Confirm that the token count of individual document chunks does not exceed the set value of
quoteMaxToken. - Initiate a test work order review request. Compare the context token count displayed on the FastGPT interface with the passed token count recorded in the logs. Confirm that the values match.
- Check the citation traceability display module. Confirm that each citation entry includes identifiers such as the source document's update time and unique number.
- Simulate a work order review after a tool change. Confirm that the latest tool management system data is prioritized for retrieval and displayed in the citation list.
The values given are common starting points and should be measured against the reader's own samples.