What data for this category looks like
This category's data comes from the electrode, assembly, formation, and quality departments of power, energy storage, and consumer lithium-ion cell manufacturing. Daily documents include batching records, coating and calendering parameter sheets, winding/stacking ledgers, liquid injection records, formation and grading data reports, OCV and safety test reports. Data sources include raw material batch ledgers, workshop MES systems, slurry and coating testing equipment, dew point monitoring devices, formation and grading cabinets, and laboratory testing systems. Updates occur per battery batch, per individual cell, and in real-time operating conditions. Most document fields have clear units, such as slurry solid content, coating thickness (μm), charging capacity (Ah), internal resistance (mΩ), and the like. A single document typically corresponds to a single batch production process or one day of equipment operating data.
Constraints on document parsing and chunking
The data characteristics of this category impose multiple constraints on document parsing and chunking. First, most data comes from structured PDFs exported by industrial equipment. Some documents have fixed equipment numbers and detection time headers and footers, requiring precise identification and skipping of irrelevant areas to avoid mixing invalid content. Second, updates occur per batch, per individual cell, and in real time. The volume of documents generated per production batch grows rapidly as production capacity increases. Chunking must balance business logic integrity, and must not split parameters from different production stages into the same chunk. In addition, the electrochemical systems used in lithium-ion cell manufacturing have extremely high requirements for batch consistency. Chunking must strictly retain complete parameter groups for the same batch, avoiding mixing data across batches. Most document fields have clear units and strong business relevance, so chunking must retain complete parameter groups for a single process. For example, group coating calendering thickness, tension, and speed as a single chunk. Finally, documents exported by some older equipment have messy line breaks and misaligned fields, requiring parsing logic adapted to non-standard formats to complete extraction.
How to configure settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
customPdfParse | Enabled | Adapts to non-standard PDF formats exported by industrial equipment, skips irrelevant areas such as headers and footers, and accurately extracts structured fields |
chunkSize | 800–1200 characters | Matches the length of complete parameter groups for a single process production ledger, avoiding splitting core detection data from a single batch |
chunkSplitter | By paragraph + semantic boundary | Preserves business logic coherence within the document, avoiding splitting parameters from the same paragraph into multiple chunks |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | A single production ledger document typically contains multiple sets of detection data, requiring sufficient time to complete full parsing, avoiding empty data processing due to timeout |
UPLOAD_FILE_MAX_SIZE | 500 MB | Adapts to the need to upload packaged multi-documents for a single full batch production process, avoiding interception due to oversized files |
chunkSettingMode | Manual configuration | Adjusts chunking parameters manually to match business logic and adapt to dedicated document formats, based on the structured document characteristics of lithium-ion cell manufacturing |
Three common configuration errors
- Symptom: After uploading a PDF exported by industrial equipment, data processing is empty, and no matching content appears in search tests. Cause: The
customPdfParseconfiguration is not enabled, and fixed headers and footers included with the equipment are not skipped, causing the parsing engine to fail to identify valid structured fields. - Symptom: Some PDF files parse normally, while others have no content after parsing. Cause: Non-standard PDF formats are not adapted. Text layers embedded in PDFs exported by older equipment may have misalignment or undisclosed encryption rules, and custom parsing logic is not used to extract content.
- Symptom: After uploading a large packaged multi-document file, parsing times out and fails. Cause: The
PARSE_FILE_TIMEOUT_SECONDSconfiguration value is too low, and insufficient time is reserved for the parsing process of multiple data sets.
How to confirm configurations are correctly set
- Upload a single-batch batching record PDF, view the parsed text content, and confirm that irrelevant areas such as headers and footers have been removed, and core fields such as batch number and slurry solid content have been correctly extracted.
- Upload a coating calendering parameter document, view the chunking results, and confirm that a single set of parameters is not split into multiple independent chunks.
- Upload a packaged document within 500 MB, wait for parsing to complete, check the data processing progress, and confirm that no timeout error prompt appears.
- Perform a search test, enter keywords for core fields in the document, confirm that the corresponding chunked content can be matched, and the search results are not empty.
The values provided are common starting points and should be measured against the reader's own samples.