What data for this category looks like
Data for this category mainly 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 sheets, mold inspection logs, first and last piece records, dimensional inspection reports, and mold change records. Data updates are triggered per coil of material, per batch of material, or each mold change. Document fields include coil number, batch number, mold ID, inspection values, equipment operation parameters, mold change time, and others. Dimensional fields commonly use millimeters as the unit, and time fields use hours or minutes as the unit.
This page applies to FastGPT v4.17.0.
What constraints do these characteristics impose on the "document parsing and chunking" link
The multi-source heterogeneous documents and high-frequency update characteristics of this category impose multiple constraints on the document parsing and chunking link. First, structural differences across document types are significant. Stamping work instructions are mostly step-by-step text, while dimensional inspection reports are structured tabular data. Differentiated parsing logic is required to adapt to these differences. Second, high-frequency updates triggered per coil, batch, or mold change require the parsing link to have fast response capabilities to avoid timeouts that delay production workflows. Third, documents contain equipment parameters, inspection values and batch identifiers. Chunking must retain field associations to avoid breaking data logic after splitting, which affects subsequent audit matching.
How to set the configuration
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category's documents include structured tables and step-by-step text. This range balances text integrity and chunk granularity, avoiding breaking associations between fields such as batch and dimension |
chunkSplitter | Split by paragraph and table boundaries | Category documents mix text and tabular data. Splitting by paragraph and table boundaries preserves data structure integrity and prevents incorrect splitting of table content |
customPdfParse | Enable scanned PDF OCR enhancement | Workshops commonly use scanned inspection reports and work instructions. Enabling this allows accurate extraction of text and tabular content from images |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Some documents in this category contain multiple batches of data, with long parsing times. This range covers the parsing needs of most documents and avoids timeout failures |
UPLOAD_FILE_MAX_SIZE | 500 MB | Collections of mold change records and inspection reports uploaded in batches by workshops usually do not exceed this threshold, preventing upload failures |
chunkSettingMode | Manual configuration | Category documents have large structural differences. Manual configuration allows adjusting chunking rules for different document types to adapt to business needs |
Three common mistakes
- Phenomenon: After uploading an Excel-format material utilization sheet, field misalignment or missing data appears in the vectorized index. Cause: No chunking rules adapted for structured tables were configured, causing Excel cell data to be incorrectly split.
- Phenomenon: After uploading a scanned dimensional inspection report, the parsing result is empty or has garbled text. Cause: The OCR enhancement configuration of
customPdfParsewas not enabled, so printed content in scanned images cannot be recognized. - Phenomenon: After uploading batch mold change records, the parsing task times out and fails. Cause:
PARSE_FILE_TIMEOUT_SECONDSwas not adjusted to the value range adapted to the category's documents, causing long parsing tasks to be forcibly terminated.
How to confirm the configuration is correct
- Upload a single stamping work instruction, check that the parsed text retains step-by-step logic with no obvious content truncation.
- Upload a scanned dimensional inspection report, confirm that the parsed result includes all table and text content with no garbled text or missing data.
- Upload batch documents, check that parsing tasks complete within the preset duration with no timeout errors.
- View the chunked text fragments, confirm that batch identifiers and corresponding inspection data are included in the same chunk with no field misalignment during splitting.
The values given are common starting points and should be measured against the reader's own samples.