What Data for This Category Looks Like
Data for this category comes from battery cell grading, glass adhesive film backplane batch ledgers, MES systems, string welding and lamination equipment logs, and EL/IV tester output files. Daily documents include layout drawings, material batch records, lamination process records, EL defect images, IV test reports, insulation voltage resistance records, and packing lists. Data updates per individual module or material batch. Documents contain multi-column structured tables, numerical parameters, image annotations, and text records. Fields include batch number, current gear, lamination process parameters, defect coordinates, power values, and other items.
What Constraints These Characteristics Impose on Document Parsing and Chunking
Multi-column structured ledgers and frequently updated data per single module or batch require document parsing and chunking to preserve data association integrity, and avoid splitting cross-column batch information. For documents mixing EL defect images and IV test reports, separate text and image parsing logic must be used to prevent incorrect splitting of image annotation content. Heterogeneous format logs from different devices require parsing rules to adapt to multi-source data structures. Chunk granularity must also match the single-link query needs of audit scenarios. For example, lamination process records must fully extract temperature, time, and pressure parameters to facilitate subsequent audit calls.
How to Set Configurations
This content is written for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the single-link parameter length of photovoltaic module documents, avoids splitting single batch material records or single process step content |
chunkSplitter | Split by paragraph and table boundaries | Preserves column association relationships for multi-column tables, meets structured retrieval requirements for photovoltaic documents |
customPdfParse | Enabled | Adapts to parsing mixed-format PDFs such as EL images and IV test reports, extracts image annotations and supporting text parameters |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Addresses parsing time for large EL image PDFs or multi-batch ledgers, avoids parsing timeout failures |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports batch uploads of multi-batch material ledgers and high-definition EL image files |
chunkSettingMode | Manual configuration | For heterogeneous formats of photovoltaic documents, customize chunk rules to adapt to different document types |
Three Common Configuration Mistakes
- When uploading multi-column Excel material batch records, only the first two columns of data are identified. The cause is that parsing rules adapted to multi-column structured tables are not configured, and the default parsing logic only extracts the first two columns of content.
- When enabling PDF enhanced parsing, EL images and IV reports cannot be parsed normally. The cause is that the
mineru APIkey and interface address are not configured correctly, or interface permissions are not enabled. - Parsed boolean-type process parameters become empty values after being processed by the conditional judgment component. The cause is that the parameter type identifier is not retained during parsing, resulting in loss of parameter format during transfer, and the conditional judgment component cannot correctly identify them.
How to Verify Correct Configuration
- Upload a multi-column material batch Excel document, check if the parsed knowledge base content retains associated data for all columns.
- Upload a PDF document containing EL images, check if the parsing result includes both image annotations and supporting process text parameters.
- Upload a single-batch lamination process record, check if the chunked content fully covers all process parameters for that batch, with no split breaks.
- Test the transfer of boolean-type process parameters, check if the conditional judgment component can correctly identify the parameters and execute the corresponding logic.
The values provided here are common starting points, and should be evaluated against the reader's own samples.