What the data for this category looks like
Core data for this category comes from CAD/CAM software, order BOMs, sheet metal inventory systems, laser cutter and bending machine device logs, and MES production management systems. Daily documents include 3D part unfolding drawings, CNC cutting programs, bending process cards, nesting diagrams, first-piece records, and process transfer cards. Data updates are adjusted in real time alongside order generation, daily inventory changes, and production changeovers. Document fields include sheet metal thickness, bending coefficient, hole position dimensions, cutting specifications, process numbers, and more. Common units are millimeters, square meters, and pieces.
What constraints do these characteristics impose on the document parsing and chunking link
Different document types have significant structural differences. Unfolding drawings exported from CAD contain vector annotations and dimensional parameters. CNC programs use a pure code format. Process cards include tables and text descriptions. Standard chunking logic cannot adapt to all document types. The multi-variety, small-batch production mode results in inconsistent document structures for individual orders. General chunking rules cannot cover all scenarios. Frequently updated data requires parsing processes with real-time capabilities. Delayed parsing will reduce cross-review efficiency in production links. Loss of key parameters will directly cause subsequent assembly errors or sheet metal waste. Therefore, chunking must retain the logical relationships between parameters. It cannot separate process steps and dimensional data.
Configuration guidelines
This document is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Sheet metal documents contain key parameters such as bending coefficients and sheet dimensions. Chunks that are too long will lose associated information. Chunks that are too short will disrupt process logic |
chunkSplitter | Split by paragraph + table | Sheet metal documents have many process card tables and CNC program paragraphs. Splitting by fixed characters will disrupt table content |
customPdfParse | Enable vector text extraction | PDF drawings exported from CAD contain vector annotations. Enabling this setting preserves the original format of dimensional fields and avoids garbled parsed text |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Large nesting diagrams or CNC program files take longer to parse. Timeouts will cause parsing failures |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapt to upload requirements for large nesting diagram files. Avoid parsing failure due to oversized files |
chunkSettingMode | Manual configuration | Document differences vary by order. Manually adjusting chunking parameters can adapt to the document structure of single orders |
Three common mistakes
- Phenomenon: After uploading an Excel sheet with multiple columns of cutting lists, automatic chunking results are messy. It cannot treat each row of data as an independent segment. Reason: The default character chunking mode is not disabled. The rule for splitting table content by row is not configured. This causes Excel tables to be randomly split by characters.
- Phenomenon: After configuring the sheet metal inventory database connection, queries return empty results. It cannot associate sheet parameters in cutting documents. Reason: No field mapping for the inventory table is not added in the
customPdfParseconfiguration. Parsing does not extract associated fields such as sheet thickness and inventory quantity. This prevents matching database data during retrieval. - Phenomenon: After uploading a complete bending process card PDF, question answering accuracy is low. It cannot accurately answer questions related to bending sequence or compensation coefficients. Reason: Chunking does not retain associated information between process steps and parameters. It splits complete process logic into scattered fragments. This prevents matching the complete process chain during retrieval.
How to confirm the configuration is correct
- Upload a typical bending process card PDF. Review the parsed chunking results. Confirm that each process table and its corresponding text description are treated as associated chunks.
- Upload a multi-column cutting list Excel. Verify that chunking generates independent segments for each row of data. There is no scattered character splitting.
- Upload a nesting diagram file larger than 500MB. Confirm that the parsing process does not trigger a timeout error. The file is successfully imported into the knowledge base.
- After associating the sheet metal inventory database. Submit a query containing a specified sheet thickness. Verify that returned results include correct inventory parameters and associated cutting information.
The values provided are common starting points and should be measured against the reader's own samples.