What the data for this category looks like
Typical subjects in this category are battery pack, energy storage cabinet, and BMS system manufacturers. Daily documents include grouping rules, BOMs, welding and torque records, BMS flashing records, insulation and voltage withstand reports, EOL test reports, and system configuration lists. Data sources include cell capacity grading data, MES systems, welding equipment, torque tools, BMS software versions, and EOL test benches. Updates occur once per pack or per cabinet, and are synchronized whenever the software version changes. Documents fall into two categories: structured tables and unstructured text. Fields include business content with clear units, such as material codes, torque values (unit N·m), and insulation resistance values (unit MΩ).
Constraints Imposed on Document Parsing and Chunking
Documents for this category mix structured tables and unstructured text. Parsers must retain both fields and their associated units to prevent information loss. Documents for a single pack or cabinet are independent business units. Chunking must split strictly along business boundaries to avoid cross-pack context confusion. Data updates are frequent, and single-document volumes are concentrated. This requires adaptation to batch parsing performance requirements. Some documents include embedded on-site images. Full access paths for these images must be retained to ensure proper rendering in conversations.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field length and business unit boundaries of documents for this category, avoids splitting single torque records or BOM entries |
chunkSplitter | Split by paragraph + table block | Adapts to document types that mix text grouping rules and structured BOMs, test reports, preserves table integrity |
customPdfParse | Enable structured parsing mode | Accurately extracts fields and units for fixed-format PDF documents such as insulation and voltage withstand reports, EOL test reports |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Adapts to the parsing duration of large BOMs or multi-page EOL reports, avoids timeout failures |
chunkSettingMode | Auto-group by business unit | Groups documents by single pack/single cabinet collections, ensures context within each chunk corresponds to a single business unit |
Three Common Mistakes
- Phenomenon: After uploading a Word-format welding record document, the domain name prefix is missing from embedded on-site photo links, and the links cannot render properly in conversations. Cause: Global upload image path replacement rules are not configured, only relative paths are retained.
- Phenomenon: When parsing a BOM document with multi-page structured tables, a parsing timeout error is returned. Cause: The set
PARSE_FILE_TIMEOUT_SECONDSduration is insufficient, not adapting to the parsing time of large documents. - Phenomenon: Chunk results include grouping rule content for two different battery packs. Cause:
chunkSettingModefor grouping by business unit is not enabled, chunks are split only by character length, and business boundaries are not recognized.
How to Verify the Configuration
- Upload a single-pack BOM document, check that the parsed text retains fields such as material codes, quantities, and their corresponding units.
- Upload a Word-format EOL test report that includes embedded images, check that image links carry complete domain name prefixes.
- Submit multiple single-cabinet document collections, check that chunk results are grouped by single cabinet, with no cross-cabinet content mixing.
- Upload an insulation and voltage withstand report with more than 10 pages, check that the parsing process does not trigger a timeout error.
The values provided are common starting points and should be measured against the reader's own samples.