What Data Looks Like for This Category
This category of data includes winding and core assembly records, vacuum drying records, insulating oil test reports, partial discharge and withstand voltage test reports, and nameplate parameter tables. Data sources are design BOMs, silicon steel sheet and insulating material incoming inspection data, vacuum drying equipment logs, oil test reports, and test station inspection records. Data is updated per single product or per batch of materials. Core parameters include equipment number, batch number, insulation resistance (unit: MΩ), partial discharge volume (unit: pC), test date, and other key fields.
Constraints for Source Citation and Traceability
Single products require full-process traceability. Citation and traceability must be accurately bound to equipment numbers and batch numbers. This prevents mixing data across products. Multi-source data structures require configuration to cover data source associations across design, production, testing, and other links. Data is updated per single product or per batch. Retrieval must prioritize the latest test and assembly records. Differences in core parameter units require unified standardization during traceability. This avoids audit deviations from inconsistent units. Abnormal high-voltage test data requires rapid traceability. This means locating corresponding batch materials and equipment logs quickly to support fault analysis.
How to Configure
The following configurations are based on the interface and parameter logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 1500–2000 token | Matches the total character count of single product inspection records, ensuring complete citation of core content from test reports and assembly records |
| Maximum Context | 8000–12000 characters | Covers full-process data including design BOMs, multi-stage production logs, and test reports, supporting complete traceability |
| Similarity Threshold | 0.82–0.88 | Accurately matches equipment numbers and batch numbers, preventing unrelated batch inspection data from being included in traceability results |
responseData | Enable source field return | Clearly displays the source path and metadata of cited documents, meeting traceability compliance requirements for inspection audits |
detail | Enable detailed traceability mode | Displays original fragments of cited content and the generation time of corresponding documents, facilitating verification of parameter timeliness |
Three Common Misconfigurations
First Misconfiguration
Phenomenon: After triggering an inspection and audit query, only cited document fragments are returned. No structured audit conclusion is included. Cause: Basic logic for answer generation is not configured. Only traceability retrieval is enabled.
Second Misconfiguration
Phenomenon: Some insulating oil reports or withstand voltage test reports are split and stored as original text blocks in the knowledge base. No corresponding question-answer pairs are generated. Cause: Chunk length exceeds the range where the model can generate question-answer pairs. Or question optimization configuration is not enabled.
Third Misconfiguration
Phenomenon: A configuration error prompt pops up during model testing. Traceability results are empty during formal operation. Cause: Multi-stage data source association rules are not correctly bound. Retrieval fails to match records for the corresponding equipment number.
How to Confirm Correct Configuration
- Enter the equipment number of a single transformer. Check if the returned results include cited fragments of the corresponding batch’s insulating oil report and withstand voltage test report.
- View the source field in the answer interface. Confirm that the complete path and metadata information of the cited documents are displayed.
- Simulate a partial discharge anomaly query scenario. Check if the corresponding batch’s silicon steel sheet incoming inspection records and vacuum drying logs can be associated.
- Test document splitting with different chunk lengths. Confirm that all inspection records can generate retrievable content blocks or question-answer pairs.
The values provided are common starting points and should be measured against the reader's own samples.