What data for this category looks like
Daily documents for this category include winding and core assembly records, vacuum drying records, insulating oil reports, partial discharge and withstand voltage test reports, and nameplate parameter tables. Data sources cover design BOM, silicon steel sheet and insulating material data, vacuum and drying equipment, oil testing, and test stations. The update frequency is once per device or per batch of materials. Most documents contain structured business fields such as insulating water content, partial discharge amplitude, test voltage, and more, with corresponding units including ppm, pC, kV, etc. The set of documents associated with a single device must maintain relevance for full-process traceability.
What constraints these characteristics impose on the "document parsing and chunking" link
Since data is updated per device or per batch of materials, documents must maintain contextual coherence for a single batch or single device. Chunking must not split key records across devices. Document types include structured tables and segmented records. Parsing must retain field alignment and format integrity, and avoid splitting core business parameters. Insulation parameters and test data are core fields for quality traceability. Losing key information during chunking will reduce the accuracy of subsequent defect analysis. The set of documents associated with a single device may include multiple files of different formats, so the parsing process must adapt to the format requirements of multiple document types.
How to set configurations
The following configurations are formulated based on the logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of a single business record for this category (such as a partial discharge test data segment) to avoid splitting key parameters |
chunkSplitter | Custom delimiter + paragraph marker | Adapts to the fixed line breaks and segmented formats of winding records and test reports |
customPdfParse | Enabled | Retains table alignment and field structure in insulating oil reports and withstand voltage test reports |
UPLOAD_FILE_MAX_SIZE | 500 MB | Limits the total upload size of documents associated with a single device to avoid parsing timeouts |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the full parsing time required for large test reports |
chunkSettingMode | Manual mode | Allows custom delimiters to match the unique document typesetting of this category |
Three common mistakes
- After uploading an Excel document for an insulating oil report, the knowledge base parsing result is empty or core fields are missing. Cause: Parsing support for corresponding structured table documents is not enabled, and the default configuration does not cover the dedicated document format of this category.
- After uploading a 300 MB withstand voltage test report PDF, the parsing task times out and fails. Cause: The
PARSE_FILE_TIMEOUT_SECONDSsetting value is too low, and does not match the full parsing time required for large test documents. - A custom delimiter is set to a line break, but the chunking result either merges multiple test paragraphs or splits a single partial discharge data entry. Cause: The linkage logic of
chunkSizeandchunkSplitteris not configured at the same time, orchunkSettingModeis not switched to manual mode to match the custom rules.
How to confirm the configuration is correct
- Upload a PDF of the winding and assembly records for a single device, and check whether the parsed text retains the complete structure of process numbers, operating personnel, and assembly parameters.
- Upload an Excel insulating oil report, and confirm that core fields such as water content and breakdown voltage can be extracted after parsing, with no format disorder.
- Upload a withstand voltage test report with a single size of 400 MB, and check whether the parsing task is completed within 10 minutes with no timeout errors.
- Adjust
chunkSizeto 1000 characters, input a segment of test records containing multiple line breaks, and check whether the chunking result is split by the custom delimiter and the length of each chunk meets the set value.
The values provided are common starting points and should be measured against the reader's own samples.