What Data for This Category Looks Like
This category serves manufacturers of cabinet chassis, ventilation equipment, electrical cabinets, and non-standard equipment. Daily documents include expansion drawings, CNC cutting programs, bending process cards, nesting drawings, first-piece records, and process flow cards. Data sources cover CAD/CAM software, order BOMs, sheet metal inventory systems, laser cutters, bending machines, and MES systems. Update frequency is per order, daily, or each time a setup change occurs. Documents include exclusive business fields such as bending coefficients, hole position coordinates, sheet thickness, and process sequence. Units are mostly millimeters, angles, batch numbers, and similar.
Constraints on Document Parsing and Chunking
The multi-variety, small-batch business model creates differences in single-batch document structures. Fixed chunking must be avoided to prevent disruption of process association logic. Document types include CAD-exported PDFs, scanned process cards, Excel flow cards, and more, so multi-format parsing support is required. High-frequency updated order data demands an efficient parsing process to avoid delays that impact quality traceability. Some documents include image-format expansion drawings, so OCR support for extracting text information is necessary. Consequences of errors include sheet metal waste and assembly errors, so field associations must be retained to support precise retrieval.
Configuration Settings
The following configurations are compatible with FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Expansion drawings and process cards for sheet metal cutting and bending contain continuous process parameters and dimensional information. This range preserves process-related context |
chunkSplitter | Split by paragraph + fixed character count | Adapts to multiple document types, avoiding splitting associated content such as bending sequence and nesting logic |
customPdfParse | Enable OCR mode | Adapts to scanned expansion drawings and process card PDFs, extracting dimensional and process text from images |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Large nesting drawings and CAD-exported PDFs take longer to parse, avoiding timeout errors |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports batch upload of multiple process cards and expansion drawings for a single order batch |
chunkSettingMode | Manual configuration | Adapts to multi-variety, small-batch scenarios, allowing adjustment of chunking parameters for different document types |
Three Common Misconfigurations
- After uploading a scanned bending process card PDF, the parsing result contains no valid text content. This occurs because OCR mode for
customPdfParseis not enabled, so text and dimensional information in images cannot be recognized. - When parsing a large CAD-exported nesting drawing, an interface "parsing timeout" error is triggered. This occurs because
PARSE_FILE_TIMEOUT_SECONDSis set to the default low value, and does not adapt to the parsing time required for large drawings. - When searching an Excel order BOM, the number of returned entries with "usage location is overseas" is less than the actual recorded volume. This occurs because correct field matching rules are not configured, and custom fields such as "usage location" are not associated, leading to failed filtering logic.
How to Confirm Configurations Are Properly Set
- Upload a scanned bending process card PDF, and check if the parsing result contains text information such as bending angle and sheet thickness.
- Upload a nesting drawing PDF larger than 100 MB, and confirm that no parsing timeout error is triggered during the process.
- Upload an Excel order BOM containing the "usage location" field, search for content with "usage location is overseas", and confirm that the number of returned entries matches the actual recorded data.
- View the chunking preview interface, and confirm that the process sequence of expansion drawings is not incorrectly split.
The values provided are common starting points, and should be evaluated against the reader's own samples.