What data for this category looks like
Data for this category comes from rubber product enterprises including tire and sealant manufacturers, their formula systems, raw material batch ledgers, internal mixers, Mooney and rheometer devices, vulcanizers, and laboratory testing equipment. Daily documents include formula sheets, weighing records, mixing curves, rubber quick test reports, vulcanization process cards, mold records, and finished product performance reports. Most documents use structures such as multi-column parameter tables, time-series curves, and fixed-field forms.
Data updates align with production milestones: per batch of rubber, per raw material batch, per mold production cycle. Fields include rubber type code, feeding ratio, mixing temperature, vulcanization duration, and more. Some documents include test images and parameter curves.
Constraints on document parsing and chunking
Multi-column parameter tables and time-series curve documents require accurate identification of column boundaries and time-series nodes during parsing, to avoid splitting rows or curve segments that span critical parameters.
High-frequency single-batch documents need to adapt to short-cycle parsing tasks to prevent timeouts. Significant field differences across document types require retaining contextual associations within the same document during chunking, to prevent loss of cross-chunk formula ratios and vulcanization parameters.
Mixing curve documents with attached test images must parse both text parameters and image content, ensuring complete binding between images and their corresponding parameters.
Since production errors have severe consequences, parsing and chunking must retain the integrity of key fields, to avoid omitting core data that affects subsequent ledger summarization.
Configuration settings
These settings are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the core parameter capacity of single chunks for documents in this category, avoiding splitting across critical fields such as formula ratios and vulcanization temperatures |
chunkSplitter | By paragraph and table boundary | Matches the structure of multi-column table and time-series curve documents, retaining contextual associations within a single chunk |
customPdfParse | Enabled | Addresses complex layouts of mixing curves and multi-column parameter tables, improving recognition accuracy for text and images |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Adapts to the parsing duration of high-frequency single-batch production documents, preventing task timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 500 MB | Adapts to the volume limit for packaged uploads of multiple single-batch documents, meeting bulk ledger upload requirements |
chunkSettingMode | Manual configuration | Enables flexible adjustment of chunking rules for different document types such as formula sheets and mixing curves |
Three common misconfigurations
- Parsing results do not include accompanying images for mixing curves. The symptom is that returned chunked content has no image fields, and logs show no image extraction records. The cause is that the
customPdfParseconfiguration is not enabled; the default parsing logic does not support mixing curve documents with attached images. - Column data from multi-column formula tables is incorrectly merged. The symptom is that chunked fields have cross-column confusion, such as feeding ratio and Mooney viscosity values being grouped into the same field. The cause is that the
chunkSplitteris not configured to split by paragraph and table boundary; the default chunking logic does not recognize multi-column structures. - Single-batch document parsing triggers a timeout. The symptom is that the task returns a 408 status code, and parsing progress stalls halfway before interrupting. The cause is that the
PARSE_FILE_TIMEOUT_SECONDSconfiguration value is lower than the actual required parsing duration, failing to account for the time required to parse images in mixing curves.
How to confirm correct configuration
- Upload a test document that includes mixing curve images and multi-column parameter tables, and check if the returned chunked content includes both text parameters and image descriptions.
- Review the parsing task run logs to confirm no timeout errors are triggered, and the task completes normally.
- Extract chunked content, verify that multi-column table fields match the original document, with no cross-column merging or loss of core fields.
- Adjust
chunkSizeto different ranges, and test that single chunks do not split complete single-batch parameter groups, such as the full content of a single vulcanization process card.
The values provided are common starting points and should be evaluated against the reader's own samples.