What data looks like for this use case
Core data for stamping and sheet metal forming scenarios comes from steel coil labels and quality certificates, press and feeder operation data, mold management systems, MES systems, and coordinate measuring machine or fixture inspection data. Daily documents include stamping work instructions, material utilization rate tables, mold inspection checklists, first and last piece records, dimensional inspection reports, and mold change records. Fields include steel coil grade, mold number, mold change time, first piece dimensional deviation values, and material utilization rate values. Updates are synchronized per coil of material, per production batch, and each time a mold is changed.
What constraints these characteristics impose on document parsing and chunking
Data for this use case comes from scattered sources and has diverse formats. It includes PDF work instructions with combined text and images, and multi-sheet Excel parameter tables. Update frequency is high, and the total volume of documents parsed in a single task may fluctuate with production batches. Documents contain many process parameters with specific units. When splitting content, the corresponding relationship between fields and units must be retained to avoid losing process details. Some documents include visual mold schematic diagrams. Parsing must distinguish between text and image content to avoid mistakenly converting image content into OCR text and losing visual information. Multi-sheet Excel documents require accurate extraction of worksheet names. Without this, parameter information for different molds cannot be matched.
Configuration settings
Configurations below are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of single process step descriptions in stamping process documents, and avoids losing contextual connections between adjacent process steps after splitting |
chunkSplitter | Split by paragraph and heading | Aligns with the chapter structure of stamping documents, such as process nodes in work instructions and inspection items in mold checklists, and retains the integrity of process information |
customPdfParse | Enable image-text separation mode | Addresses work instructions with combined text and images, retains original image content such as mold schematic diagrams, and avoids overwriting original visual information with default OCR conversion |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Accommodates parsing time for multi-sheet Excel material utilization rate tables and batch-imported steel coil quality certificates, and avoids timeout errors for large document parsing |
UPLOAD_FILE_MAX_SIZE | 500 MB | Meets the total file size requirements for single-import multi-batch mold change records and multiple dimensional inspection reports, and supports bulk data processing |
chunkSettingMode | Manual configuration | Adjusts chunking rules for different document types, such as PDF work instructions and Excel parameter tables, to adapt to multi-format data in stamping scenarios |
Three common errors
- Phenomenon: After importing a PDF document, images such as mold schematic diagrams are converted to OCR text, and no original images are displayed. Cause: The image-text separation mode of
customPdfParseis not enabled. Default OCR conversion overwrites original image content. - Phenomenon: After parsing an Excel material utilization rate table, only the content of the first worksheet is read, and the names and data of other sheets are not extracted. Cause: Chunking rules adapted for multi-sheet documents are not configured. Default settings skip metadata extraction for non-first worksheets.
- Phenomenon: When importing Word or PPT format process documents, the parsing result has missing content or formatting errors. Cause: Parsing mode is not adjusted based on document type. Enhanced parsing function is only enabled for PDF formats.
How to confirm configuration is correct
- A single PDF stamping work instruction including a mold schematic diagram is uploaded. The parsing result is verified to retain the original image, and the image is not replaced by OCR text.
- An Excel material utilization rate table with 3 or more sheets is uploaded. The parsing result is checked to confirm it includes the names and corresponding parameter content of all sheets.
- A Word document of mold change records is uploaded. Core fields such as mold change time and mold number are verified to be fully extracted during parsing.
- A compressed package of steel coil quality certificates with a total size of approximately 450 MB is uploaded. The upload and parsing process is checked to confirm it does not trigger file size limit or timeout errors.
The values provided are common starting points and should be measured against the reader's own samples.