This document is written based on FastGPT v4.17.0.
What data for this category looks like
Data sources include product drawings and 3D models, CAPP/CAM systems, data collected from machine tools, tool management systems, MES systems, and inspection and measuring 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. The system triggers data updates per work order, each workpiece, sampling batch, or tool change. Document fields include operation number, tool model, cutting speed, feed rate, calibration error value, and more. Most formats are structured tables or parameterized process text.
What constraints do these characteristics impose during deployment and upgrade
To support multi-source data access, configure cross-system data pull adaptation rules during deployment to avoid format incompatibilities. Match data refresh frequency thresholds to data synchronization triggered by high-frequency updates, to prevent incremental data accumulation. When processing process documents that include structured parameters and CNC code in chunks, retain complete key fields to avoid losing process benchmark information during retrieval. During upgrades, synchronize and migrate business data associated with multiple systems to prevent the knowledge base content from being disconnected from actual production. The accuracy of retrieval results directly affects machining quality. Configure recall rules during the deployment phase based on the process logic of metal cutting machining.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 20–30 | Match the concurrency requirements of multi-source data synchronization in metal cutting machining scenarios, and adapt to the concurrency volume of workshop-level data collection |
MONGODB_URI | Bound to the dedicated connection address of the workshop process database | Isolate production process data from other business data to ensure data access security |
REDIS_URL | redis://:custom 32-bit random password@target server IP:6379/0 | Cache frequently accessed process parameters and retrieval results to reduce database read and write pressure |
STORAGE_S3_ENDPOINT | Access endpoint of enterprise internal object storage | Store large-volume process documents such as CNC programs and 3D models to ensure large file reading speed |
FILE_TOKEN_KEY | Custom 32-bit random string | Encrypt access tokens for process documents to prevent unauthorized access to production data |
Three common errors
- Phenomenon: After upgrading to a new version, the original process knowledge base content is not displayed synchronously. Cause: The database migration step was not executed, causing the process data from the old version to not be mapped to the database structure of the new version.
- Phenomenon: After redeploying the reranking model, each retrieval returns results marked
falsefor reranking. Cause: The access key or endpoint address of the reranking model was not configured, causing the system to fail to call the reranking model to complete result sorting. - Phenomenon: Request timeout occurs when multiple stations concurrently query process data. Cause: The
DB_MAX_LINKconfiguration value is too low, causing the database connection pool to be exhausted and unable to handle concurrent process data query requests.
How to confirm configuration completeness
Perform a database connection test to verify that the configured MONGODB_URI can normally connect to and read process document data. Upload a CNC program or operation card document to trigger vector generation and retrieval, and confirm that retrieval results include key process parameters from the document. Check system logs to confirm that the cache service configured via REDIS_URL normally receives and stores frequently accessed process data. Simulate concurrent queries from multiple stations to verify that the connection pool configured via DB_MAX_LINK can support concurrent requests without connection timeout errors.
The values provided are common starting points and should be measured against the reader's own samples.