What the data for this category looks like
The core data of this category comes from business outputs of tooling departments of automotive, aerospace, mechanical assembly enterprises and dedicated tooling suppliers. Daily documents include tooling requirement documents, positioning benchmark 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 metrology systems. Update frequency adjusts with project progress, product changes or regular calibrations. Documents include fields such as GD&T tolerance annotations, positioning clamping parameters, calibration cycles and measured values. Dimension units are mostly millimeters. Record-type documents include change time and responsible person information.
Constraints on Document Parsing and Chunking
Documents in this category have diverse formats, including vector engineering drawings, structured tables and professional technical texts. This requires adaptable parsing capabilities. Professional terms such as GD&T and coordinate measuring machine parameters must be fully retained to avoid semantic breaks from incorrect splitting. The update rhythm tied to projects and changes requires chunking logic to bind project context, preventing cross-project document content from being mixed. For time-series documents such as calibration records and change histories, chunk content must be split by time nodes, while retaining the association between parameters and corresponding verification results. Engineering drawings contain precise dimension annotations. When chunking, ensure that text descriptions linked to graphics are not split. This ensures complete positioning benchmark information can be retrieved fully in later steps.
The following configurations apply to FastGPT v4.17.0.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Single project documents in this category contain professional parameters and text associated with engineering drawings. This range preserves the integrity of parameter associations and avoids semantic breaks |
chunkSplitter | Split by header/footer + paragraph | Documents include structured content such as tooling requirement documents and calibration records. Splitting by document structure preserves the logical integrity of single-type documents |
customPdfParse | Enabled | Documents in this category include vector engineering drawings and GD&T annotations. Enabling custom PDF parsing preserves text metadata associated with graphics, avoiding loss of professional annotation information during parsing |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Single project documents may contain multiple fixture drawings and calibration records. A longer timeout ensures complete parsing of large PDFs or multi-file packaged documents |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Project-level documents may include multiple engineering drawings and data attachments. This upper limit covers the storage and parsing requirements of conventional project documents |
chunkSettingMode | Manual configuration | Documents in this category have large format differences. Manual configuration allows adjusting chunking rules for different document types, adapting to the retention requirements of professional parameters |
Three Common Mistakes
- Phenomenon: After a fixture engineering drawing PDF is uploaded, the GD&T annotation text associated with the graphics is empty, and the model cannot recognize dimension parameters. Cause: The
customPdfParseconfiguration is not enabled. The default parsing only extracts plain text, losing metadata associated with vector graphics. - Phenomenon: The number of chunks after parsing a single bid document is far higher than expected. Retrieval cannot obtain a complete positioning benchmark scheme. Cause: The
chunkSizevalue is too small, splitting a complete positioning benchmark scheme into multiple independent chunks and destroying the parameter association logic. - Phenomenon: Garbled Chinese characters appear when scanned gauge operation manuals are parsed, and operation steps cannot be recognized. Cause: Chinese-adapted OCR parsing configuration is not enabled. The default parsing engine has insufficient character recognition accuracy for professional terms.
How to Verify Correct Configuration
- Upload a single fixture drawing PDF, check if the parsed text includes GD&T tolerance annotations and positioning dimension parameters, confirming the
customPdfParseconfiguration is active. - Select a document fragment containing a positioning benchmark scheme, verify if the chunked content retains complete parameter associations, confirming the
chunkSizevalue is adapted to the document structure. - Upload a document containing multiple pages of calibration records, check if the parsing duration meets expectations, confirming the
PARSE_FILE_TIMEOUT_SECONDSconfiguration is sufficient to cover parsing time. - Import calibration data in Excel format, check if field mapping corresponds to parameters from the metrology system, confirming custom parsing rules are adapted to structured table formats.
The values provided are common starting points and should be measured against applicable sample data.