What data for this category looks like
Daily documents for this category include tooling requirement documents, positioning benchmark plans, 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 follow individual project timelines, are triggered by each product change, and are updated during regular calibration cycles. Most documents contain engineering drawing parameters, metrology values, and change traceability fields. Some documents include embedded CAD attachments or raw CMM inspection data.
What constraints do these characteristics impose on document parsing and chunking
Document characteristics for this category impose multiple constraints on parsing and chunking. Structured content with engineering drawing parameters and metrology values must retain the association between fields and their corresponding traceability information. This prevents benchmark plans and acceptance data from being split into separate chunks. Documents with embedded CAD attachments or raw CMM data require parsing logic adapted for non-text formats. This prevents loss of structured inspection data. The update pattern tied to projects, product changes, and calibration cycles requires chunked data to link to corresponding version identifiers. This supports subsequent audit traceability. Mixed multi-source document structures require avoiding chunking confusion across content types. This ensures consistent audit logic for inspection and process records.
How to set configurations
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category’s documents include engineering parameters and metrology data. This range preserves the integrity of individual process or inspection records and avoids splitting critical fields. |
chunkSplitter | Split by title + paragraph | Daily documents often include hierarchical tooling plans and operation steps. Splitting by title preserves the independence of logical units, and adapts to segmented verification needs during audits. |
customPdfParse | Enabled | Most documents for this category include engineering drawings and embedded attachments. Enabling custom PDF parsing adapts to non-standard layout engineering documents and preserves the association between charts and parameters. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Some documents include large CMM data or CAD attachments. This duration covers parsing time for complex documents and prevents mid-process timeout interruptions. |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Project documents for this category may include multiple attachments uploaded in a package. This upper limit covers storage and parsing needs for most project-level documents. |
chunkSettingMode | Manual configuration | Document structures vary significantly across different projects. Manual configuration allows flexible adjustment of chunking rules, adapting to customized inspection and audit workflows. |
Three common mistakes
- Phenomenon: When parsing documents containing raw CMM data, the log reports a slow operation timeout error. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted, and the default duration is insufficient for documents with large structured data. - Phenomenon: After uploading a docx document with embedded engineering drawings, the image fields in the parsing result are empty or show as invalid. Cause:
customPdfParseis not enabled to adapt to non-standard layout engineering documents, so embedded drawing images cannot be correctly identified. - Phenomenon: When parsing large fixture design documents, CPU usage is high but GPU resources are not fully utilized. Cause: Parsing rules adapted for hardware acceleration are not configured, and the default full-process parsing uses CPU, so idle GPU resources cannot be called.
How to confirm configurations are correctly set
- Upload a calibration record document containing CMM data. Check that the parsing log has no timeout errors, and verify that the
PARSE_FILE_TIMEOUT_SECONDSconfiguration matches the actual parsing time of the document. - Upload a fixture drawing document with embedded engineering drawings. Check that the image fields in the parsing result load normally, and confirm that
customPdfParseis enabled. - Upload multiple packaged project documents. Verify that the upload size does not trigger the
UPLOAD_FILE_MAX_SIZElimit, and confirm that storage and parsing processes work normally. - Adjust the
chunkSizeparameter. Check that the chunked result retains the integrity of a single process record, and confirm that the chunking rule adapts to the document structure.
The values provided are common starting points and should be measured against the reader's own samples.