What the data for this category looks like
Core data for this category comes from ERP material master data, melting furnaces, thermometers, spectrometers, MES heat records, mold management systems, customer drawings, and technical specifications. Data updates occur per heat, per batch, or daily. Covered documents include casting process cards, batching lists, heat records, mold histories, first article inspection records, material test reports, nonconforming product review forms, and similar documents. Document fields include alloy batching ratios, melting temperatures, pouring parameters, defect types, heat numbers, mold cavity numbers, and other relevant fields. No extra statistical values beyond preset ones are included.
Constraints imposed during deployment and upgrade
Multi-source heterogeneous data sources require connecting standard interfaces of different systems during deployment. This avoids data synchronization gaps. Frequently updated production data requires configuring near-real-time synchronization tasks. This prevents expired heat records from being used in bid documents. Document fields have high professional specificity. Preset extraction rules for corresponding fields must be configured during deployment. This ensures generated bid documents contain accurate casting process parameters. Additionally, data errors can cause serious consequences. A data verification link must be configured during deployment. This intercepts abnormal alloy batching ratios or temperature values in advance.
How to configure settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 10–20 concurrent connections | Adapts to the parallel document reading needs of multiple heats in foundries, avoiding database connection exhaustion |
MONGODB_URI | Connection address pointing to the heat records and mold history collections synced from the foundry MES | Stores bulk casting process documents and real-time production data |
REDIS_URL | Local cache instance address, cache duration 300–600 seconds | Reduces loading latency for frequently accessed customer drawings and material test reports |
STORAGE_S3_ENDPOINT | Endpoint address of the enterprise internal object storage | Stores attached files such as casting process cards and nonconforming product review forms |
FE_DOMAIN | Access domain name deployed on the enterprise intranet | Restricts front-end access scope, ensuring the security of bid-related data |
ROOT_KEY | Complex random string | Highest privilege key for the management backend, only accessible to deployment and operations personnel |
Three common mistakes
- Issue: OneAPI channel list is empty, and models cannot be selected. Cause:
MONGODB_URIis not correctly configured to point to the dataset containing channel configurations, or the database connection has timed out. - Issue: Images cannot be pulled during Docker offline deployment, or installation packages fail to load. Cause: The offline package compatible with FastGPT v4.17.0 was not prepared in advance, or
STORAGE_S3_ENDPOINTis not configured to point to the local image storage source. - Issue: Model testing prompts an error, but short-context references run normally. Cause:
DB_MAX_LINKis set too low. Test requests trigger database connection limits, while short-context requests do not exceed the connection threshold.
How to confirm proper configuration
- Access the configured
FE_DOMAIN, log in to the FastGPT management backend, and check the database connection panel to confirm the connection status ofMONGODB_URIandREDIS_URLis normal. - Upload a standard casting process card document, test extracting fields such as alloy batching ratios and heat numbers, and confirm the extraction results match the field content in the source document.
- Initiate a bid document generation test, and check whether the generated content includes accurate casting-related information such as melting temperatures and pouring parameters.
- Check the backend running logs to confirm there are no error records for database connection timeouts or file storage failures.
The values given are common starting points and should be measured against the reader's own samples.