What Data for This Category Looks Like
Daily documents for this category include winding process cards, insulation treatment records, dynamic balance reports, dielectric strength insulation test reports, and type and factory inspection test reports. Data sources include electromagnetic design data, copper wire and insulating material batch records, winding equipment operation data, balancing machine test results, electrical test bench test data, and production batch information from ERP/MES systems. Data is updated once per motor or product batch. All associated documents are updated synchronously when any type change occurs. Documents contain specific fields such as winding turns, insulation class, test voltage, and temperature rise values, paired with corresponding physical units.
What Constraints Do These Characteristics Impose on Document Parsing and Chunking
Documents in this category combine structured process parameters and complex test data, and their fields have dedicated physical units. Parsing and chunking must avoid splitting core associated information. Documents generated from multiple data sources have varying formats: both structured Word process cards and PDF test reports with tables and formulas. Parsing rules must be adapted to different formats. Each product’s documents are strongly bound to its production batch. When chunking, context association for documents from the same batch must be retained, to avoid splitting that prevents matching corresponding batch material and equipment data. If key parameters are lost during parsing or chunking is done improperly, the accuracy and compliance of subsequent tender quotations will be directly affected.
How to Configure Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of core parameter segments in winding process cards and test reports, avoids splitting key associated fields |
chunkSplitter | By paragraph + semantic boundary | Retains context association for process steps and test data, avoids splitting winding turns and wiring instructions |
customPdfParse | Enable MinerU adaptation mode | Adapts to complex table and formula layouts in motor test reports, improves parsing accuracy |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Addresses parsing time for large type test reports, prevents timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports batch upload of entire batch test reports, fits bulk tender scenarios |
chunkSettingMode | Manual configuration | Adjust chunking parameters for different document types (process cards/test reports) to adapt to multi-format documents |
Three Common Configuration Errors
- Image links in parsed Markdown are missing domain names after uploading Word documents: This occurs because the image external link configuration for
customPdfParseis not enabled, causing local image paths to not be replaced with accessible public domain prefixes. - Timeout errors occur when parsing large type test reports: This happens because
PARSE_FILE_TIMEOUT_SECONDSis set too low, failing to match the parsing time required for complex documents. - Associated information between winding turns and wiring is lost after chunking: This is caused by configuring
chunkSplitterto split by fixed characters, without retaining semantic context association.
How to Verify Configuration is Correct
- Upload a winding process card document, check if the parsed text retains core fields such as winding turns, insulation class, and their corresponding physical units.
- Upload a type test report containing tables and formulas, confirm that the parsing progress does not show a timeout interruption, and that the
PARSE_FILE_TIMEOUT_SECONDSconfiguration is active. - Adjust the
chunkSizeparameter, check that the chunking results do not split a single segment of process description or test data. - Verify that image external links in documents have accessible domain prefixes, confirming that the
customPdfParseconfiguration is enabled.
The values provided are common starting points and should be measured against the reader's own samples.