This document is written based on FastGPT v4.17.0.
What the data for this category looks like
Daily documents for tooling, fixture and gauge manufacturing include tooling 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 data, on-site trial feedback, and measurement systems. Updates are driven by project progress, triggered by product changes, and completed via periodic calibration updates. Documents have dedicated fields such as structured measurement values, geometric dimensioning and tolerancing parameters, and calibration cycles. Some documents, like fixture drawings, are CAD-exported PDFs with structures that differ significantly from general office documents.
What constraints do these characteristics impose on document parsing and chunking
Documents in this category have large structural differences. CAD drawing PDFs, technical documents, and calibration records follow different format logic, so targeted parsing rules are required. Structured coordinate measuring machine data and GD&T parameters must retain full field integrity, and must not be split arbitrarily. The update rhythm tied to projects and changes requires chunks to be associated with project metadata, to avoid mixing content across projects. Parsing large-scale tooling CAD drawings takes a long time, so a longer parsing duration threshold must be used to prevent mid-process failures.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the typical paragraph length of tooling requirement documents and acceptance reports, while accommodating segmented content from coordinate measuring machine data, and avoids splitting critical technical parameters |
chunkSplitter | Custom delimiters (section titles, line breaks) | Meets the structured segmentation needs of change histories and gauge operation manuals, and avoids splitting GD&T technical parameters and positioning datum descriptions |
customPdfParse | Enabled | Targets the non-standard PDF format of fixture CAD drawings, adapts the extraction logic for layers and professional technical text, and improves parsing completeness |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Addresses the long parsing time of large-scale tooling CAD drawing PDFs, and prevents parsing failures caused by insufficient default duration |
chunkSettingMode | Manual configuration | Adjusts chunking parameters for different document types such as calibration records and change histories, to adapt to the content structure of specific scenarios |
Three Common Mistakes
- Uploading a large-scale tooling CAD drawing PDF results in a parsing failure, with a timeout error returned in the logs. The cause is failure to adjust the
PARSE_FILE_TIMEOUT_SECONDSparameter; the default duration is insufficient to complete text and layer extraction for large drawings. - Configuring custom post-parsing logic without matching the category's document structure leads to incorrect splitting of GD&T parameters and calibration record fields. The cause is failure to enable
customPdfParseand configure targeted document recognition rules. - Knowledge base recall results include unrelated old-version change history content. The cause is failure to configure chunking with project-associated metadata, leading to mixed chunking of content across projects.
How to Confirm Configurations Are Set Correctly
- Upload a single fixture drawing PDF under 100MB, check whether the parsing status completes within the set
PARSE_FILE_TIMEOUT_SECONDSduration, with no timeout errors. - Import a tooling requirement document containing GD&T parameters, check whether the chunking result retains complete geometric dimensioning and tolerancing descriptions, and is not split incorrectly.
- After enabling
customPdfParse, verify that the step text of gauge operation manuals is correctly extracted, with no missing or garbled content. - Upload multiple documents from the same project, check whether the chunking result retains project-associated metadata, and no mixed content across projects appears.
The values provided are common starting points and should be measured against the reader's own samples.