All configurations below are based on FastGPT v4.17.0.
What data for this category looks like
Daily documents for general mechanical equipment assembly include assembly process cards, BOMs, torque records, gap adjustment records, commissioning reports, factory inspection sheets, and packing lists. Data sources include PLM/ERP, part inspection data, assembly tool collection systems, test benches, MES, and customer technical agreements. Document update frequency is once per equipment unit or order. Some fields use professional units, such as torque values (N·m), gap values (mm), and coaxiality tolerances (μm). Document structures vary across different orders.
What constraints do these characteristics impose on document parsing and chunking
Most documents for this category are structured content customized per order. Process cards and inspection sheets have different formats across orders. Chunking must retain the original hierarchical and logical structure of documents. Generic chunking may split process steps or inspection items incorrectly, so this must be avoided. Documents also contain large amounts of numerical data with professional units. Parsing must fully retain fields and units. Otherwise, matching errors will occur in subsequent retrieval and abnormal report drafting. Additionally, single documents may have multiple pages. Assembly documents for large non-standard equipment are lengthy. Parsing and chunking must adapt to different document lengths to avoid timeouts or incomplete parsing.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category’s documents include long text such as process descriptions and inspection data. 800–1200 characters preserves the complete logic of a single process step or inspection item, avoiding split breaks |
chunkSplitter | markdown | Most documents for this category are exported from PLM in formats with hierarchical headings. Using markdown title-based chunking retains the original document structure |
customPdfParse | Enabled | PDF documents for this category often contain embedded tables such as torque records and gap adjustment records. Enabling this option preserves structured table data |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large assembly process cards or multi-page commissioning reports takes significant time. 600 seconds covers parsing needs for most standard documents |
chunkSettingMode | custom | Document formats vary widely across orders for this category. Custom chunking rules adapt to all document types |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Assembled document collections for large non-standard equipment may include multiple pages of drawings and supporting records. 1000 MB covers standard upload requirements |
Three common mistakes
- Issue: When calling the file collection creation API, embedded table content is lost in PDF parsing results, with no clear error prompt. Cause: The
customPdfParseparameter is not set to enabled in the API request. The default parsing mode cannot retain structured table information. - Issue: Recent PDF parsing tasks frequently fail, with timeout status listed in logs. Cause:
PARSE_FILE_TIMEOUT_SECONDSis not adjusted to a value suitable for large documents. The original short timeout setting cannot complete full parsing of multi-page commissioning reports. - Issue: After uploading assembly documents from multiple orders, search results mix content from different orders, making accurate matching to target equipment impossible. Cause: No metadata with order identifiers is added to uploaded files. Chunking does not retain document source information. This leads to parsing results that cannot be distinguished.
How to confirm configurations are correct
- Upload a PDF document containing embedded torque tables. Check if the parsed text retains table row and column structures and professional units such as N·m.
- Call the parsing interface. Check if the returned parameters include the
customPdfParseconfiguration flag to confirm the parameter is active. - Upload two assembly process cards from different orders. Search for exclusive keywords for each order. Confirm that returned results only include documents from the target order.
- Upload a large assembly manual with more than 500 pages. Check if the parsing task completes within 600 seconds with no timeout errors.
The values provided here are common starting points. They should be evaluated against applicable samples.