What Data for This Category Looks Like
Daily documents for this category include BOMs and coordinate files, line change sheets, stencil and reflow oven temperature records, first article confirmation reports, SPI/AOI inspection reports, maintenance records, and more. Data sources include customer-provided BOMs, Gerber files, coordinate data, plus real-time collected data from MES systems, mount machines, SPI/AOI equipment, and reflow ovens. Update frequency is triggered per work order and each line change; some device data is synced in real time. Most documents are PDF files with tables or structured text, containing fields such as part number, reference designator, component lot number, reflow temperature parameters, coordinate values, etc. Units include millimeters, degrees Celsius, standard part number codes, and others.
Constraints on Document Parsing and Chunking
The multi-source heterogeneous documents, high-frequency update rhythm, and demand for precise fields for this category impose multiple constraints on document parsing and chunking. Tabular SPI/AOI reports, BOM files, and similar documents require retention of table structure and field associations, to avoid table misalignment or field breakage after parsing. Real-time synced temperature and device data require fast parsing workflows to prevent delays in work order processing. Document layouts vary significantly across different work orders, so flexible chunking rules must be supported. Cross-document field association logic must also be retained, to ensure that subsequent audits can link key information such as corresponding part numbers and coordinates for verification work.
How to Configure
This configuration applies to FastGPT v4.17.0, optimized for SMT mounting and electronic assembly scenarios.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the paragraph length of tabular content in SMT documents, avoiding splitting associated fields such as part numbers and coordinates |
chunkSplitter | Separate by paragraph + table | Matches the structured layout of BOMs and inspection reports, retaining the association between tables and their corresponding explanatory text |
customPdfParse | Enabled | Adapts to the parsing of complex tables and embedded parameters in SMT documents, avoiding table misalignment caused by default parsing rules |
PARSE_FILE_TIMEOUT_SECONDS | 60–90 seconds | Addresses the high-frequency document parsing requirements of real-time collected data, preventing work order process interruptions caused by timeouts |
UPLOAD_FILE_MAX_SIZE | 50 MB | Adapts to the scenario of batch uploading multiple documents per work order, meeting the upload needs of bulk BOMs and inspection reports |
chunkSettingMode | Manual configuration | Adjust chunking rules for different document types such as BOMs and SPI reports, to meet the requirement of retaining precise fields |
Three Common Configuration Mistakes
- Phenomenon: The table in uploaded SPI/AOI inspection report PDFs is misaligned and field layout is chaotic after parsing. Cause: The
customPdfParseconfiguration is not enabled, and the default parsing rules cannot adapt to complex table structures. - Phenomenon: Some files fail to parse when uploading work order documents in bulk, and a timeout prompt is displayed in the interface. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted, and the parsing duration of real-time data documents collected at high frequency exceeds the default threshold. - Phenomenon: In the parsed BOM documents, part number and corresponding reference designator information are split into different chunks. Cause: The
chunkSizesetting is too small, splitting complete paragraphs containing associated fields.
How to Verify Proper Configuration
- Upload a standard SPI/AOI inspection report PDF, and check whether the parsed content retains the complete table structure and field associations.
- Upload a compressed package containing multiple BOM files, verify that the uploaded file size does not exceed the
UPLOAD_FILE_MAX_SIZElimit, and there are no upload failure prompts. - Trigger a parsing task for real-time collected reflow oven temperature record documents, check whether the parsing duration is within the range set by
PARSE_FILE_TIMEOUT_SECONDS, and there are no timeout error messages. - Adjust the
chunkSizeparameter, compare the chunking results under different values, and confirm that associated fields such as part numbers and reference designators are not split into different chunks.
The values provided here are common starting points and should be evaluated against the reader's own samples.