What Data Looks Like for This Category
Data for this category comes from winding, insulation, final assembly, and testing departments of electric motor factories and power equipment enterprises. Data sources include electromagnetic design data, copper wire and insulation material batch records, winding equipment, balancing machines, electrical test benches, and ERP/MES systems. Documents include winding process cards, insulation treatment records, dynamic balance reports, dielectric strength insulation test reports, type and factory inspection test reports. Fields include number of turns, insulation class, balance accuracy, dielectric strength value, test date, and others. Update frequency is per unit, per batch product, or each time a type change occurs.
Constraints on Knowledge Base Retrieval and Recall
Diverse document types contain electromagnetic and mechanical professional parameters. Retrieval must accurately match the correspondence between process parameters and test results, to avoid confusing different batches of materials and processes during supplier audits. Multi-batch data updated at high frequency must support incremental updates, to prevent full retraining from consuming excessive resources and interrupting daily audit workflows. Exclusive test data for single units or batches must be distinguished from general process data during recall, to prevent cross-batch parameter misuse that causes production abnormalities. Strong constraint fields such as dielectric strength, temperature rise, and dynamic balance must be used as retrieval weight items, to ensure rapid location of matching records during abnormal analysis.
Configuration Settings
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the parameter accuracy of professional documents such as winding processes and test reports, filters irrelevant general mechanical content |
limit | Top 8 entries | Covers multi-dimensional audit documents for a single motor batch, while controlling context length to fit model input limits |
searchMode | mixedRecall | Balances precise matching of professional terms such as "turn-to-turn insulation" and "dynamic balance accuracy" and parameter values |
usingReRank | Enabled | Re-ranks recall results by business relevance, prioritizing core audit items such as dielectric strength tests and type tests |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms related to process parameters, covering professional expressions such as "impregnation drying" and "insulation system" |
collectionIds | Collection IDs grouped by document type | Distinguishes general process libraries and batch-specific test libraries to avoid confusion of cross-batch data |
Three Common Configuration Errors
- Performing a full knowledge base retraining with no change in retrieval results. The cause is failing to correctly trigger the incremental update trigger condition, or interrupting the retraining task before completion.
- Importing a Markdown file with local images, with images not displaying in retrieval results. The cause is failing to upload local images to a supporting image bed and update the links within the file.
- Exported knowledge base files are only hundreds of KB and have incomplete content. The cause is only exporting metadata for some collections, not selecting the full document export option, or exporting large files before parsing is complete.
How to Verify Proper Configuration
- Upload a test winding process card, search for "winding turn count verification", and check if the recall results include this document and rank highly.
- View the knowledge base task logs, confirm that incremental update tasks trigger normally with no failed errors.
- Export the knowledge base file, verify that the file size matches the total size of uploaded documents with no obvious missing content.
- Enter a test query containing professional terms, such as "causes of dielectric strength insulation test failure", and check if the recall results include the corresponding test reports.
The values provided are common starting points and should be measured against the reader's own samples.