What Data for This Category Looks Like
Photovoltaic cell manufacturing data typically originates from process, equipment, quality control, and production teams at crystalline silicon cell enterprises. Daily documents include process cards, diffusion and coating parameter tables, printing and sintering records, EL detection images, IV grading data, and yield reports. Data sources include silicon wafer incoming material records, MES systems, production equipment such as diffusion furnaces, EL and IV testing devices, and laboratory data. Updates occur per material batch, per-piece test, and daily summary. Documents contain core fields including batch identifiers, process parameters, and test values.
Constraints Imposed by These Characteristics on Document Parsing and Chunking
The data characteristics of this category create multi-dimensional constraints for document parsing and chunking. First, document types include structured parameter tables, semi-structured test records, and unstructured image reports. Parsing rules must be configured to support multiple document formats. Second, update frequencies span fine-grained per-piece test data to daily summary bulk data. Chunking must align with data granularity to prevent mixing information across batches or processes within a single chunk. Third, documents contain associated fields such as batch identifiers, process parameters, and test values. Chunking must preserve field associations to enable full traceability of single-batch production and testing workflows during later audits. Fourth, image documents like EL detection images require simultaneous extraction of metadata and annotation details. This prevents loss of key judgment evidence.
Configuration Settings
The following configurations are based on parameter rules for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category of documents contains fine-grained process parameters and test data. Chunk length must balance information completeness and retrieval accuracy, and avoid including cross-process associated data in a single chunk |
chunkSplitter | Split by paragraph and semantic meaning | Documents cover structured parameter tables, semi-structured test records, and unstructured image reports. Rules must retain semantic relevance of content, and avoid splitting that breaks associations between parameters and batches |
customPdfParse | Enabled | PDF documents for this category include embedded parameter tables and EL detection image annotations. Custom parsing supports extraction of structured data and image metadata |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Parsing large EL detection image summary files and bulk production logs takes significant time. Sufficient processing time must be reserved to prevent mid-process interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Enterprises need to upload complete batch test reports and device logs. Support for large file uploads is required |
chunkSettingMode | Manual configuration | Document formats for this category vary significantly. Manual configuration allows flexible adjustment of chunking rules to adapt to different types of process and test documents |
Three Common Mistakes
- Symptom: Connection failure error occurs when calling an external parsing service. Local address testing still fails to establish connectivity. Cause: Docker container port mapping is not configured correctly. This prevents FastGPT from accessing locally deployed parsing services.
- Symptom: Excel format files cannot be recognized by the knowledge base after upload. CSV file uploads only extract the first two columns of data. Cause: Custom file parsing configuration is not enabled, or CSV parsing column count adaptation rules are not adjusted.
- Symptom: Knowledge base retrieval results do not retain the reply portion of the original question-and-answer pair, and extra generated content appears. Cause: Rules for directly returning knowledge base original text after retrieval are not configured, or chunking splits the complete reply content of a question-and-answer pair.
How to Verify Proper Configuration
- Upload a typical process card PDF file. Check if parsed text contains complete process parameters and batch identifiers. Verify that chunked content retains semantic relevance.
- Upload an EL detection image and corresponding IV data table. Confirm that the parsing service extracts both image metadata and test values from the table simultaneously.
- Adjust the value of
chunkSize, then upload documents of different lengths. Verify that chunked content lengths match the preset range. - Test uploading bulk log files under 1000 MB. Confirm that files upload normally and complete parsing.
The values provided are common starting points and should be evaluated against the reader's own samples.