What the data for this category looks like
Data for general mechanical equipment assembly scenarios comes primarily from PLM, ERP, MES systems, assembly tool collection terminals, test benches, and customer technical agreements. Daily documents include assembly process cards, multi-level BOMs, torque records, clearance adjustment records, commissioning reports, and factory inspection sheets. Fields include torque values with units (Nm), clearance parameters (mm), concentricity indicators, and assembly node data updated per order or individual equipment.
What constraints do these characteristics impose on document parsing and chunking?
Multi-level BOMs and process parameter documents with units require retaining hierarchical associations and unit information during parsing and chunking, to avoid losing parameter context after splitting. Documents updated per individual equipment or order need to be linked to unique equipment or order identifiers during chunking, to ensure corresponding batch data can be located during traceability. Documents include multiple structures such as step-by-step process cards and tabular inspection records. Parsing logic for different layouts must be adapted to prevent splitting that breaks the association between process steps and their corresponding inspection results.
How to configure the settings
This document is compatible with FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the text length of process parameters with units and multi-level BOMs, avoiding loss of parameter associations after splitting |
chunkSplitter | Split by paragraph + heading | Adapts to the step structure of assembly process cards and hierarchical headings of BOMs, retaining the logical structure of documents |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Addresses parsing time for multi-level BOMs and multi-page inspection sheets, avoiding timeout during large file parsing |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports batch upload of full batch assembly documents for a single piece of equipment |
customPdfParse | Enabled | Retains tables and hierarchical structures in PDF documents, ensuring complete parsing of BOMs and inspection records |
chunkSettingMode | Manual configuration | Customizes chunking rules for the document structure of this category, adapting to different types of assembly documents |
Three Common Configuration Mistakes
- Phenomenon: Calling the parsing interface returns split-related error logs, or text fragments are missing after parsing. Cause: The splitting logic for multi-level BOMs and tabular documents of this category is not adapted, and the default chunking rule truncates hierarchical association information.
- Phenomenon: Parsing speed is too slow, and parsing time for a single 100-page assembly document exceeds the preset threshold. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted, or the parsing mode adapted to tables and hierarchical structures is not enabled, causing the parsing engine to repeatedly attempt to split complex structures. - Phenomenon: The document parsing function fails after upgrading to FastGPT v4.9.0, and works normally when switching back to an older version. Cause: The
customPdfParseparameter is not configured correctly, or the parameter rules of the new version's parsing module are not adapted, causing the old version compatibility logic to not take effect.
How to Confirm the Configuration Is Correct
- Upload a typical multi-level BOM document, and check that parsed chunks retain hierarchical headings and corresponding component information, with no cross-level splitting.
- Upload a torque record document with units, and confirm that split text fragments contain complete numerical values and units, with no truncation.
- Adjust the
PARSE_FILE_TIMEOUT_SECONDSparameter, submit a large-volume assembly document, and confirm that no timeout error appears in the parsing status. - Compare parsing results for different types of assembly documents (process cards, inspection sheets), and confirm that the chunking logic adapts to different document layout structures.
The values provided are common starting points and should be measured against the reader's own samples.