Data Characteristics for This Category
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 include design BOMs, silicon steel sheet and insulating material procurement data, vacuum drying equipment logs, oil test reports, and high-voltage test records from the test station.
Data updates are performed per single product or batch of materials. Each document includes fixed structured fields: insulating oil reports note water content and breakdown voltage, partial discharge reports record partial discharge magnitude values, nameplates list rated capacity and voltage rating. All fields have dedicated attached units.
Constraints Imposed on Document Parsing and Chunking
Full-process documents for a single product require traceable association. Chunking must retain inter-document association identifiers to avoid losing context after splitting.
Structured fields have dedicated units. Parsing must preserve the binding between fields and their units. Chunking must not separate fields from their corresponding values.
High-frequency updated test documents require fast parsing. Chunk length must accommodate the difference between short test reports and long assembly records.
Insulation-related data is a core control item. Parsing must prioritize extracting key fields such as insulating oil water content and partial discharge abnormal values. Chunking must aggregate associated fields to avoid omissions during subsequent summary.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Accommodates mixed documents of this category, including long assembly records and short test reports, to avoid splitting key fields |
chunkSplitter | Split by paragraph + fixed field delimiter | Preserves the binding relationship between structured fields such as insulating oil water content and partial discharge values, meeting business data association requirements |
customPdfParse | Enabled | Optimizes structured extraction effects for fixed-format PDF documents such as insulating oil reports and test reports, adapting to the custom parsing rules of FastGPT v4.17.0 |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Prevents parsing timeouts when processing large-capacity withstand voltage test reports or multiple documents uploaded in batches |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates the total size of multiple archived documents associated with a single product, meeting batch upload requirements |
chunkSettingMode | Manual configuration | Adjusts chunking rules for different document types, adapting to the large structural differences of documents in this category |
Three Common Misconfigurations
- Phenomenon: Feishu Knowledge Base fails to synchronize transformer documents in PPT or PDF format, or displays no valid content after parsing. Cause: The
customPdfParseconfiguration is not enabled, or the parsing rules are not adjusted for the fixed format of documents in this category, causing the engine to fail to recognize structured fields. - Phenomenon: After deploying
marker, logs show anocr errormessage. Cause:PARSE_FILE_TIMEOUT_SECONDSis not adjusted to a reasonable range, or the uploaded high-resolution insulating oil scan report exceeds the default OCR processing threshold. - Phenomenon: When using the Claude 3.7 model, normal chat works normally, but parsing errors occur after passing in transformer test reports. Cause: The association between partial discharge values and test time is not retained during chunking, causing the model to fail to associate context to complete the summary logic.
How to Verify Correct Configuration
- Upload an insulating oil report PDF, check whether the parsed text retains fields such as water content and breakdown voltage along with their corresponding units.
- Check the
PARSE_FILE_TIMEOUT_SECONDSconfiguration item to confirm that the value falls within the preset range, and upload a large-capacity withstand voltage test report to verify that no timeout error occurs. - Synchronize PPT-format assembly records in Feishu Knowledge Base, check whether they can be loaded normally and content related to the winding process can be parsed.
- Call the chunking interface to verify that the generated chunked content includes both partial discharge abnormal values and their corresponding test time in the same chunk.
The values provided are common starting points and should be measured against the reader's own samples.