What the data for this use case looks like
PCB manufacturing inspection and process record data sources include customer Gerber files, drill files, BOMs, CAM engineering files, equipment operating parameters, bath test results, AOI inspection reports, electrical test reports, and sectioning reports. Data is updated per production batch and shift. Some equipment parameters and bath data are collected continuously. Daily documents include engineering production instructions, drilling lamination parameter sheets, chemical analysis records, and more. Fields include drilling rotation speed, lamination temperature, defect coordinates, on-resistance, and more. Units include revolutions per minute, degrees Celsius, ohms, and more. All fields are tightly bound to production batches.
What constraints do these characteristics impose during deployment and upgrade
Data sources are scattered and update frequency is high. Deployment must support multi-data source connection and real-time synchronization configuration. This prevents data delays from affecting inspection efficiency. Document types and fields are diverse. Pre-configure parsing rules and metadata mapping for different format files. This ensures field completeness of inspection records. PCB manufacturing inspection errors can lead to full batch board scrap or high-value claims. The upgrade link must configure data backup and rollback mechanisms. This protects production data from loss. Internal access permissions must be strictly isolated. This prevents unauthorized external access to sensitive production data.
How to set configurations
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 80–120 connections | Adapts to PCB manufacturing data connection demands that fluctuate with production batches, prevents database connection exhaustion |
MONGODB_URI | mongodb://localhost:27017/pcb_prod_records | Stores batch-level inspection data such as AOI and electrical test results, local deployment ensures production data security |
REDIS_URL | redis://localhost:6379/0 | Caches continuously collected bath and real-time equipment parameters, reduces main database access pressure |
FE_DOMAIN | http://pcb-factory-intranet:8080 | Binds to the factory intranet domain name, only accessible to internal engineering and quality personnel |
STORAGE_S3_ENDPOINT | http://local-storage:9000 | Connects to local object storage, stores large-volume Gerber files and sectioning reports, ensures intranet access speed |
Three common configuration mistakes
- Symptom: The interface and knowledge base content display in English after login. Cause: The default language environment was not configured correctly during deployment, or the language pack path associated with
FE_DOMAINdoes not point to Chinese resources. - Symptom: Download failures or 404 errors occur when searching and trying to download AOI reports or sectioning reports. Cause: The storage endpoint configured in
STORAGE_S3_ENDPOINTdoes not match the actual file upload storage path, leading to incorrect file path mapping. - Symptom: Calls to the locally deployed large model return a model unavailable prompt. Cause: The API address of the local large model was not added to the system's allowed call list, or the corresponding access key was not configured.
How to confirm configuration is complete
- Access the intranet address corresponding to
FE_DOMAIN, confirm that the interface language is Chinese, and that knowledge base categories and fields match the preset configuration. - Upload a test Gerber file and AOI inspection report, confirm that the system can correctly parse the file and generate corresponding metadata fields.
- Initiate a local large model call, enter a preset inspection audit question, confirm that results that conform to business logic are returned normally.
- View the database connection number monitoring panel, confirm that the current connection number does not exceed the
DB_MAX_LINKconfiguration value, and there are no connection exhaustion alerts.
The values provided are common starting points and should be measured against the reader's own samples.