What this category of data looks like
Data sources include internal equipment operation and maintenance manuals, equipment ledger Excel files, paper scan copies of maintenance records, employee job training documents, and spare parts inventory lists. Update cadence: Equipment ledgers are updated alongside spare parts purchases and major equipment overhauls. Maintenance records are updated after each individual fault handling is completed. Training documents are updated quarterly. Document structures include: PDF manuals with parameter tables that include units, long-form operation steps, and illustrated fault troubleshooting workflows; Excel ledgers with fixed fields such as equipment number, operating hours, rated pressure, and replacement cycle; scanned maintenance records are mostly printed or handwritten tables, with blurry printed and handwritten markings. Field units include hours, megapascals, pieces, days, and other common engineering units.
What constraints these characteristics impose on document parsing and chunking
Parameter tables with units and long-form operation and maintenance manuals require retaining table structure during parsing, and avoiding splitting cross-page parameter descriptions. Structured Excel ledgers with fixed fields and large numbers of entries require grouping chunks by equipment number to ensure complete context for operation and maintenance data of the same equipment. Image content in scanned maintenance records requires enabling OCR parsing mode to avoid losing handwritten fault descriptions and processing results. Large operation and maintenance manuals with large file sizes require adjusting upload and parsing thresholds to avoid task interruptions mid-process.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the paragraph length of construction machinery manuals, retains complete context for parameter tables |
chunkSplitter | By paragraph + fixed character boundaries | Avoids splitting fault descriptions and subsequent processing steps in maintenance records |
customPdfParse | Enable OCR + table retention mode | Parses scanned maintenance sheets and equipment manuals with images |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Meets upload requirements for complete operation and maintenance manuals of large complete sets of equipment |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Avoids parsing timeouts when processing PDF documents over 100 MB |
chunkSettingMode | Manual configuration | Adjusts chunk grouping rules for structured equipment ledgers |
Reference only; verify against the target FastGPT release and deployment documentation.
Three common mistakes
- Phenomenon: After uploading a scanned maintenance sheet, the parsing result only contains a small amount of plain text, and tables and handwritten content are lost. Cause: The OCR mode of
customPdfParseis not enabled, only the plain text layer of the PDF is parsed, and text in images is not recognized. - Phenomenon: In the chunked equipment ledger data, multiple records of the same device are split into different chunks. Cause:
chunkSplitteronly splits by character boundaries, and does not associate context grouping by equipment number. - Phenomenon: After uploading a large operation and maintenance manual, the parsing task shows a timeout failure. Cause: The value set for
PARSE_FILE_TIMEOUT_SECONDSis lower than the actual duration required for document parsing.
How to confirm the configuration is correctly set
- Upload a scanned equipment maintenance sheet, and verify that the parsing result includes handwritten fault descriptions, parameter values, and processing steps from the image.
- Upload an equipment operation and maintenance manual with multiple tables, and verify that chunked content retains complete table structure, with no cross-chunk parameter association breaks.
- Upload a complete set of equipment operation and maintenance manuals over 100 MB, and verify that the parsing task completes within 600 seconds without timeout errors.
- Upload a structured equipment ledger Excel file, and verify that chunking results are grouped by equipment number, with multiple records for the same device not split across different chunks.
All provided configuration values are common starting points, and should be evaluated against the user's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.