What the Data for This Category Looks Like
Daily documents for this category include rolling plans, heating schedules, rolling specifications, plate shape and thickness records, surface quality reports, and quality certificates. Data sources include slab or billet batches, heating furnaces, mill automatic control equipment, online thickness and width measuring devices, surface inspection systems, and MES systems. Updates occur in real time per coil or per bar. Document fields cover heating temperature, reduction rate, plate shape deviation, thickness tolerance, batch number, and other items. Units include degrees Celsius, millimeters, tons, and others.
Constraints Imposed on Document Parsing and Chunking
The real-time coil- or bar-level updates and multi-source heterogeneous document structure of this category create multiple constraints for document parsing and chunking. First, independent data for each coil or bar requires parsing results to retain batch number associations, to prevent incorrect splicing of cross-batch content. Second, documents contain complex parameter tables, such as multi-column parameter tables for rolling specifications. Full table structure must be preserved to support field comparison for cross-review. Third, layout differences across document types — such as linear layout for plan documents and structured sections for quality certificates — require parsing tools to adapt to multiple formats, to avoid missing key fields. Fourth, a single document may cover data for multiple coils. Chunking must split along coil-level boundaries to prevent mixing cross-coil content.
Configuration Settings
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the length of single-section content in rolling documents, preventing truncation of cross-batch information |
chunkSplitter | Split by document structure | Retains coil-level and batch-level content boundaries, avoiding damage to data associations from forced character-based chunking |
customPdfParse | Enable digital signature compatibility mode | Adapts to quality certificate documents with digital signatures, resolving issues where signature information causes content recognition failures |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Adapts to the parsing duration of large-volume rolling plan documents, preventing interruptions to the parsing process due to timeout |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports uploading batch rolling record documents, meeting the requirement for one-time submission of multi-coil data |
chunkSettingMode | Manual configuration | Adjusts chunking rules for the special document structure of this category, adapting to complex tables and batch information |
Three Common Misconfigurations
- Phenomenon: After uploading a quality certificate PDF with a digital signature, the parsing result is empty or has missing fields. Cause: The digital signature compatibility mode for
customPdfParseis not enabled, causing signature information to block the parsing process. - Phenomenon: Calling the document parsing tool returns a file read error, and the log shows a file read timeout. Cause:
UPLOAD_FILE_MAX_SIZEis not adjusted to adapt to large-volume rolling plan documents, orPARSE_FILE_TIMEOUT_SECONDSis set too short. - Phenomenon: Chunked content includes cross-coil rolling data, making cross-review by batch impossible. Cause:
chunkSplitteruses a fixed character chunking mode, and does not retain coil-level boundaries according to document structure.
How to Verify Correct Configuration
- Upload a single-coil surface quality report PDF, and check that the parsed table fields are fully preserved, with no missing or misaligned content.
- Upload a quality certificate PDF with a digital signature, and confirm that the parsing result includes the main body content, with no empty fields or recognition failure prompts.
- Check the parsing log to confirm that no timeout errors related to
PARSE_FILE_TIMEOUT_SECONDSappear. - Select a rolling plan document that includes data for multiple coils, and check that the chunking result splits along coil-level boundaries, with each chunk corresponding to a single coil.
The values provided are common starting points and should be measured against the reader's own samples.