What this type of data looks like
Typically, inspection and process record data for crystalline silicon cell manufacturing comes from silicon wafer incoming material ledgers, production MES systems, real-time operating parameters of diffusion furnaces, coating equipment, and printing presses, test results from EL and IV testing equipment, and laboratory sampling data. Documents produced daily include process cards, diffusion and coating process parameters, printing and sintering process records, EL detection images, IV performance grading data, and batch yield reports. Data update frequencies cover per-batch materials, per-product inspection, and daily process summary statistics. Data volume grows in line with production scale.
What constraints these characteristics impose on the deployment and upgrade phase
Multiple heterogeneous data sources require the deployment phase to support integration of different formats of device logs, structured parameter files, and image-based detection data. High-frequency update rhythms require the system to support near-real-time data synchronization and retrieval, to avoid delays that disrupt process adjustments. Diverse document types require adaptation to multi-modal file storage and parsing, covering structured parameters, static images, and batch reports. Fine-grained wafer-level data requires storage and retrieval engines to support high concurrent processing capacity. Meanwhile, the micro-structure sensitivity of production processes requires data verification links to have high-precision field matching capabilities, to prevent incorrect parameters from entering production workflows.
How to configure the system
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
MONGODB_URI | Enterprise internal MongoDB cluster connection address dedicated to MES and inspection data | Used to store unstructured EL detection images, IV grading data, and yield reports. Dedicated clusters isolate resource usage between business systems and the AI platform |
STORAGE_S3_ENDPOINT | Endpoint address of the enterprise self-built object storage service | Used to store high-resolution EL detection images and batch process parameter files. Self-built storage meets data compliance and low-latency access requirements |
REDIS_URL | Enterprise internal cache cluster connection address, set connection timeout to 30 seconds, maximum connections to 200 | Caches frequently accessed process parameters and device status data, shortens retrieval response times, and reduces database load |
DB_MAX_LINK | Set to 200–300 | Meets concurrent requirements for processing multiple batches of wafer-level detection data and real-time device upload requests, avoids connection exhaustion |
FILE_TOKEN_KEY | Enterprise internal unified file access key | Performs permission control on sensitive process documents and detection data, prevents unauthorized access |
ROOT_KEY | Set to a 16-character random string | Used to configure the platform's highest privilege access, ensures operational security during deployment |
Three common configuration errors
- The symptom is a directory not found error during docker build. The log displays
failed to solve: failed to comput(truncated full error). The cause is failure to mount local process document and device log directories to the specified paths in the build image, which prevents reading dependent configuration files and sample data during the build. - The symptom is empty results returned when an HTTP interface processes txt files. The cause is incorrect configuration of
FILE_TOKEN_KEY, which prevents reading process parameter txt files uploaded by the frontend, and prevents completion of subsequent processing and response. - The symptom is excessively high latency for batch retrieval requests. The cause is failure to select a lightweight large language model adapted for non-local deployment, which causes inference speed to fail to meet real-time requirements when processing large volumes of wafer-level detection data.
How to confirm successful configuration
- Access the FastGPT v4.17.0 admin backend, check if the
MONGODB_URIandSTORAGE_S3_ENDPOINTconfigurations match the enterprise internal storage service addresses. Initiate a test connection to verify connectivity. - Upload a simulated EL detection image and IV grading data file, verify normal storage and parsing of the file by the system, and confirm normal opening and viewing of the generated file access link.
- Initiate a batch process parameter retrieval request, check the cache hit rate of
REDIS_URL, and confirm that the response time meets preset business requirements. - Review system operation logs, confirm absence of
DB_MAX_LINKrelated connection exhaustion errors, and confirm normal concurrent request processing with no backlog. - Verify normal configuration of root access permissions. Use
ROOT_KEYto complete the platform's highest privilege operations, with no permission exception errors.
The values provided are common starting points and should be measured against the reader's own samples.