What this category of data looks like
Daily documents for fixture and gauge manufacturing include fixture requirement documents, positioning datum schemes, fixture drawings, gauge operation manuals, acceptance reports, calibration records, and change histories. Data sources include product 3D models and GD&T, process routes, coordinate measuring machine (CMM) data, on-site trial feedback, and metrology systems. Updates occur per individual project, product changes, and regular calibrations. Most document structures include technical parameters, process steps, metrology data, and change nodes. Fields cover technical indicators with clear units such as positioning tolerance, clamping force, and calibration cycle.
What constraints do these characteristics impose on document parsing and chunking
This category includes vector drawings, annotated PDFs, structured tables, and change records. Core information distribution varies significantly across different document types. Project numbers and change identifiers across documents enable correlation of multi-source data. Chunking must retain contextual associations to avoid breaking technical logic. Frequently updated change histories and calibration records require precise version node identification during parsing, to prevent cross-version data mixing. Large-scale fixture drawings and CMM data files have large sizes. Parsing and chunking workflows must support large file processing.
How to set up the configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the typical length of technical parameter sections and process descriptions in this category of documents, avoids splitting core technical logic |
chunkSplitter | Split by paragraph and header/footer markers | Adapts to document structures with fixture drawing annotations and calibration footers, retains chapter integrity |
customPdfParse | Enabled | Supports parsing PDF files with embedded fixture drawings and GD&T annotations, extracts vector annotation information |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Meets upload requirements for large-scale fixture drawings and CMM data files |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Reserves sufficient processing time for complex CMM data parsing, prevents timeout interruptions |
chunkSettingMode | Manual configuration | Allows adjusting chunking rules for different document types, adapts to multiple types of business documents |
Three common configuration errors
- After uploading an Excel dataset with embedded fixture drawings, some metrology fields are empty. The cause is that
customPdfParseis not enabled, resulting in failure to correctly parse and extract embedded PDF drawings and CMM data from the Excel file. - When parsing an acceptance report that includes change histories, chunking omits change nodes. The cause is that
chunkSplitteris configured to the default character-based split, which does not recognize change marker paragraphs in the document. - Uploading a large-scale fixture drawing PDF triggers a parsing timeout. The cause is that the
PARSE_FILE_TIMEOUT_SECONDSsetting is lower than the actual required parsing duration, failing to adapt to large file processing needs.
How to confirm the configuration is correct
- Upload a single fixture drawing PDF, check if the parsed chunks retain core chapter content such as positioning datum, clamping scheme, and tolerance annotations.
- View the parsing log of the uploaded file, confirm that no timeout errors occur, and that the
PARSE_FILE_TIMEOUT_SECONDSconfiguration value matches the processing duration shown in the log. - Export the current knowledge base's chunking configuration, import it to a new server, compare the configuration items for consistency with the original configuration, and confirm that parameters such as
chunkSizeandchunkSplitterhave not been reset. - Upload a document with continuous change histories, check if chunks are split by change time nodes, with no merged chunks spanning change nodes.
The values provided are common starting points and should be measured against the reader's own samples.