What this category of data looks like
Typical entities are injection molding, stamping, die-casting mold enterprises and mold centers of manufacturing enterprises. Daily documents include mold design task sheets, DFM reports, part drawings, processing technology cards, trial mold reports, mold modification notices, mold history records, etc. Data sources include customer product models, mold flow or forming simulation files, CAD/CAM output files, machining and measurement data, and records related to trial mold samples. Document updates are triggered by project milestones. New records and reports are generated after each trial mold or mold modification. Some documents contain CAD-embedded text, scanned drawings, and structured Excel data. Fields include mold number, cavity count, material grade, trial mold defect type, etc. Units include millimeters, megapascals, hours, and other professional engineering units.
What constraints do these characteristics impose on the "document parsing and chunking" link
The document characteristics of mold design and manufacturing create multiple constraints for the parsing and chunking process. First, scanned trial mold reports and part drawing PDFs require OCR support to extract text. Without OCR, professional descriptions in the drawings cannot be retrieved. Second, PDFs exported from CAD may contain embedded text and layered drawing content. The association between text and drawings must be retained. Third, structured data such as Excel technology cards and trial mold records have hidden columns and multi-field links. Field integrity must be preserved during chunking. Fourth, documents are updated by project and closely connected. Chunking must retain cross-paragraph context to avoid splitting professional analysis paragraphs. Finally, large assembly drawing PDFs have large file sizes. The system must adapt to the longer parsing time required for these files.
How to set the configurations
The following configurations are based on the functional logic of FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Mold documents contain many professional terms and associated paragraphs. 800–1200 characters preserves the context of a single analysis paragraph or complete report chapter, avoiding splitting professional content |
chunkSplitter | Split by paragraph + title | Mold documents have structured chapters such as DFM reports and trial mold records. Splitting by title retains chapter relevance, making it easier to match complete analysis logic during retrieval |
customPdfParse | Enable OCR mode | Scanned trial mold reports and part drawing PDFs require OCR to extract scanned text. This ensures professional descriptions in drawings can be retrieved |
PARSE_FILE_TIMEOUT_SECONDS | 1200 seconds | Parsing large mold assembly drawing PDFs and multi-sheet Excel technology cards takes a long time. 1200 seconds prevents timeout errors during standard large file parsing |
chunkSettingMode | Manual configuration | Chunking rules must be adjusted for different document types, including CAD-exported PDFs, scanned files, and Excel files. Manual configuration adapts to the diverse formats of mold documents |
UPLOAD_FILE_MAX_SIZE | 2000 MB | Some large mold assembly drawing PDFs have large file sizes. 2000 MB supports uploads for most large documents |
Three common errors
- Empty search results after uploading a scanned PDF: The cause is that the OCR mode of
customPdfParseis not enabled, so text content in the scanned image is not extracted. - Fewer than expected results returned for Excel knowledge base searches: The cause is that associated fields for structured data are not preserved during chunking, or Excel structured parsing mode is not enabled. This causes some eligible rows to not be split correctly.
- Timeout error during PDF parsing: The cause is that
PARSE_FILE_TIMEOUT_SECONDSis set to the default short duration, which does not match the parsing time required for large mold drawings.
How to verify correct configuration
- Upload a scanned trial mold report PDF, and confirm the parsed text includes trial mold defect descriptions to verify OCR functionality.
- Upload an Excel technology card with multiple columns, search for entries in a specified field, and confirm returned results cover all eligible rows.
- Upload a mold assembly drawing PDF larger than 100 MB, and confirm the parsing process completes within 1200 seconds without timeout errors.
- Split a DFM report PDF, and confirm chunked content retains complete paragraphs such as "demolding analysis" and "material selection" with no forced splitting of professional content.
The values given are common starting points and should be measured against the reader's own samples.