What the data for this category looks like
Daily documents for complete home appliance manufacturing include configuration BOMs, assembly work instructions, refrigerant charging records, electrical safety test reports, performance sampling inspection reports, and repair records. Data sources include ERP/MES systems, component batch ledgers, tightening and charging device logs, safety tester exported data, performance lab test results, and after-sales fault feedback ledgers. Updates to data occur per individual product, shift production batch, and sampling plan execution. Document structures combine structured fields and unstructured descriptions. Field units include production-specific metrics such as torque, charging volume, and test thresholds.
Constraints imposed on document parsing and chunking
Multi-model mixed-line production requires document parsing to retain model-associated fields, to avoid mixing data across models. High-frequency updated production and after-sales data require parsing processes to support real-time processing, to prevent delays that reduce defect analysis efficiency. Multi-source structured and unstructured documents require chunking rules to adapt to different document formats, to retain logical connections between fields and paragraphs. Severe potential error consequences require parsing and chunking accuracy to meet required standards, to prevent loss of key process parameters or batch information. Such loss disrupts subsequent quality traceability and recall investigations.
How to configure the settings
The following configurations are based on FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field density of home appliance manufacturing documents, avoids splitting associated structured fields such as material codes and their corresponding specifications |
chunkSplitter | Split by paragraph + fixed headings | Matches the fixed process and heading structure of assembly work instructions and test reports, retains logical connections between paragraphs |
customPdfParse | Enable structured parsing mode | Adapts to tabular formats exported by ERP BOMs and device logs, accurately extracts key fields such as batch numbers and process parameters |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Covers parsing durations for large-capacity safety test reports and batch repair records, prevents task termination due to mid-run timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports batch uploads of device logs for a single shift’s production batch, adapts to bulk data submission requirements |
chunkSettingMode | Dynamically adjust based on document type | Differentiates chunking rules for BOMs, charging records, and repair records, avoids generic chunking that disrupts document logic |
Three common misconfigurations
- Phenomenon: After uploading PDF-format charging records or repair reports, parsing status shows completed but no corresponding chunks appear in the knowledge base, and no error logs appear in the backend. Cause: The
customPdfParseconfiguration is not enabled. The default generic parsing cannot recognize the tabular PDF structure exported by devices, resulting in no valid content being extracted. - Phenomenon: After uploading a packaged file containing multiple safety test reports, the parsing task times out and fails. Cause: The
PARSE_FILE_TIMEOUT_SECONDSsetting is too short, failing to cover the parsing duration of large-capacity batch documents. - Phenomenon: Chunking results split the material code and corresponding specification fields in BOMs. Cause: The
chunkSizesetting is too small, splitting originally associated structured fields into different chunks and disrupting their logical connections.
How to confirm configurations are correct
- Upload a single typical document, such as the refrigerant charging record for a single air conditioner, and verify that parsed chunks retain key fields including charging time, charging volume, and batch number.
- Upload a large-capacity test report, and check that the parsing task execution duration stays within the preset
PARSE_FILE_TIMEOUT_SECONDSthreshold. - Upload a mixed package of multiple document types, and verify that the
chunkSplittercompletes chunking according to document structure, with no cross-type data mixing. - Call the API to upload a test file, and confirm that chunked content synchronizes normally to the knowledge base, matching results from interface-based uploads.
The values provided are common starting points and should be measured against in-house samples.