What the Data for This Category Looks Like
The main entities in this category are power 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 version records, and EOL test benches. Documents are updated once per pack or cabinet, and updated alongside software version changes. Some device log files have large capacity, and include structured fields such as material codes, torque values, and insulation resistance parameters.
Constraints on Document Parsing and Chunking
Documents in this category include both structured tables (such as BOMs and EOL reports) and semi-structured logs (such as welding records). Update frequency is high. Parsed data must be accurately retrievable for quality audits. File formats from multi-source equipment are inconsistent. Different parsing logic must be supported. Single-file capacity varies widely. Parsing timeouts or lost critical parameters must be avoided. Quality audits rely on accurate parameter matching. Chunking operations cannot split associated parameter groups, such as material codes and corresponding supplier information in BOMs.
Configuration Setup
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the structured field length of multiple document types in this category, avoids splitting associated parameter groups |
chunkSplitter | Split by paragraph + table boundaries | Preserves the integrity of tables in documents such as BOMs and EOL reports, prevents incorrect splitting of table content |
customPdfParse | Enable OCR mode | Processes scanned insulation and voltage withstand reports and EOL test records, extracts text and test parameters embedded in images |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Adapts to large-capacity cell capacity grading data and device log files, avoids parsing interruptions due to timeout |
UPLOAD_FILE_MAX_SIZE | 200 MB | Supports uploading multiple batches of welding equipment log files in a single upload, meets the upload requirements for batch audits |
chunkSettingMode | Manual configuration | Adjusts chunking parameters for different document types (such as BOMs and test reports) to meet the accurate retrieval requirements of quality audits |
Three Common Configuration Errors
- Phenomenon: Only native text is returned after uploading a PDF document. Image-embedded insulation withstand voltage values are not extracted. Cause: The OCR mode for
customPdfParseis not enabled. Only the document’s native text layer is parsed. - Phenomenon: Material codes and corresponding quantities in BOM tables are split into different chunks. Cause:
chunkSplitteris configured to split by fixed characters. Table boundaries are not adapted, leading to incorrect truncation of table content. - Phenomenon: Parsing timeout errors trigger when uploading large-capacity EOL test reports in batches. Cause:
PARSE_FILE_TIMEOUT_SECONDSis set too low. It does not adapt to the parsing duration of single large-capacity files.
How to Confirm Proper Configuration
- A BOM document with embedded tables may be uploaded. Chunking results may be checked for complete material rows and corresponding parameters.
- A scanned insulation withstand voltage report may be uploaded. Parsing results may be confirmed to include test values and timestamps from the report.
- A welding equipment log file with a capacity exceeding 100 MB may be uploaded. The parsing process may be checked for triggered timeout errors.
- The
chunkSizeparameter may be adjusted. Retrieval results under different chunk lengths may be compared. Critical parameters may be confirmed to not be split into different chunks.
The values provided are common starting points and should be measured against the reader's own samples.