What This Category of Data Looks Like
Daily documents for this category include 3D part unfolding drawings, CNC cutting programs, bending process cards, nesting diagrams, first-piece records and process flow cards. Data originates from CAD/CAM software, order BOM systems, sheet metal inventory ledgers, laser cutter and bending machine device logs, and workshop MES systems. Data updates trigger with order creation, daily inventory checks, or production changeovers. Document fields include sheet thickness, bending coefficients, cutting dimensions, bending angles, and more. Units are mostly millimeters and degrees. Some process cards include CNC code snippets.
What Constraints These Characteristics Impose During Deployment and Upgrade
Multi-type heterogeneous production documents require parsing logic adapted to different formats. During deployment, configure parsing rules compatible with CAD drawings, CNC code, and text-based process cards. Frequently updated production data requires the database to support high-concurrency writes. Adjust connection pool parameters to match the workshop’s concurrent access volume. Cross-system association requirements for multi-source data require the database to support cross-table queries, while adapting to data format differences across systems. Additionally, the multi-variety, small-batch production model requires the knowledge base to support dynamic addition of document categories without fixed binding to product templates, avoiding frequent adjustments to deployment configurations.
Configuration Settings
The following configurations are adapted for FastGPT v4.17.0, as detailed below:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
MONGODB_URI | mongodb://fastgpt:securepass@mongodb.internal:27017/fastgpt_prod?authSource=admin | Adapts to storage and retrieval of multi-source heterogeneous production documents, supports high-frequency small-batch data writes |
REDIS_URL | redis://:redispass@redis.internal:6379/1 | Caches frequently retrieved bending coefficient and nesting rule data, reducing load on the core database |
STORAGE_S3_ENDPOINT | http://minio.internal:9000 | Stores large files such as CNC cutting programs and unfolding drawings, adapting to newly added production documents each day |
DB_MAX_LINK | 200–300 connections | Matches concurrent access volume from workshop equipment and orders, preventing service interruptions caused by exhausted database connections |
FE_DOMAIN | https://sheet-metal-process-review.your-domain.com | Configures the frontend access domain name, ensuring cross-domain requests for document upload and retrieval complete normally |
FILE_TOKEN_KEY | Randomly generated 32-digit hexadecimal string | Secures access permissions for uploaded process cards, nesting diagrams, and other files, preventing unauthorized access |
Three Common Misconfigurations
- Symptom: No output from database connection calls in workflows, with tool call timeout displayed in the interface. Cause:
MONGODB_URIorPG_URLis not configured as an address accessible from the workshop intranet, preventing FastGPT from connecting to MES or sheet metal inventory databases. - Symptom: No response after uploading a bending process card, with no retrieval index generated. Cause: Read/write permissions for the storage bucket corresponding to
STORAGE_S3_ENDPOINTare not configured, preventing the parsing service from reading uploaded files. - Symptom: Locally deployed large language models fail to respond to document queries. Cause: The model invocation secret key for
ROOT_KEYis not set correctly, or theREDIS_URLcache configuration does not adapt to the large language model’s context cache requirements, preventing the model from accessing uploaded document data.
How to Verify Successful Configuration
- Access the document upload page at
FE_DOMAIN, upload a bending process card, and confirm that the file is saved normally and displayed in the knowledge base’s category list. - In the FastGPT test chat interface, enter "Query current sheet metal inventory", confirm that the corresponding database query result is returned, with no connection errors or timeout prompts.
- Trigger a batch upload of documents during a production changeover, check the database connection count monitoring metrics, and confirm that the connection count does not exceed the
DB_MAX_LINKconfiguration limit. - Invoke the locally deployed large language model, enter "Summarize the bending sequence in this cutting program", confirm that the model generates a valid response based on the uploaded document, with no empty responses or error prompts.
The values provided are common starting points and should be measured against the reader's own samples.