What data looks like for this category
Core data for this category comes from winding, insulation, final assembly, and testing departments of motor factories and power equipment enterprises. Daily documents include winding process cards, insulation treatment records, dynamic balance reports, dielectric strength test reports, and type and factory inspection reports. Data is sourced from electromagnetic design data, copper wire and insulation material batch records, winding equipment, balancing machines, electrical test benches, and ERP/MES systems. Updates trigger per unit or batch production, and each type change. Data fields include winding turns, paint dipping process parameters, test voltage, temperature rise values, batch numbers, and more. Corresponding units include turns, megapascals, degrees Celsius, volts, and others.
What constraints do these characteristics impose on knowledge base retrieval and recall
This category's data characteristics impose multiple constraints on the retrieval and recall link. Documents cover multiple professional links including winding processes, insulation treatment, and test and inspection. Retrieval scope must be divided by process scenarios to avoid mixing irrelevant cross-process content. Data updates per unit, batch, or type change. Latest version batch data must be prioritized for recall to prevent use of outdated parameters. Professional fields such as winding turns, test voltage, and temperature rise values require exact term matching. Fuzzy retrieval often returns invalid results. Multi-source data comes from different systems including devices and ERP/MES. Heterogeneous field formats must be integrated to ensure field consistency in retrieved content.
How to set configurations
This explanation is based on the configuration logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Meets precision requirements for matching professional terms such as winding turns and test voltage, filters low-relevance content |
limit | Top 8–12 entries | Covers core retrieved data volume across multiple process links, avoids exceeding processing limits due to overly long context |
usingReRank | Enabled | Reorders retrieved results to raise priority of dynamic balance reports and dielectric strength test data |
searchMode | fullTextRecall | The query examples emphasize identifiers or specialist terms; test this mode with representative data, not as a product default. |
datasetSearchUsingExtensionQuery | Enabled | Adds synonym expansion for professional terms to cover different expressions of process parameters |
collectionIds | Collection IDs grouped by winding, insulation, and testing links | Limits retrieval scope, only retrieves production ledger data for the corresponding links |
Three common mistakes
- Symptom: After uploading PDFs of winding process cards or test reports, search tests return empty results or show errors. Cause: Failed to limit the retrieval collection for the corresponding process link, or set the
similaritythreshold too high, filtering matched professional content. - Symptom: After splitting the knowledge base into chunks of conventional length, complete winding turns and wiring information cannot be fully retrieved. Cause: Chunking truncates context containing complete parameters, without adjusting chunking strategy for long professional documents.
- Symptom: When using an Ollama local model, knowledge base question answering performance is worse than API models. Cause: The embedding accuracy of the local model does not meet professional term matching requirements, and
usingReRankfor optimizing retrieved result ranking is not enabled.
How to confirm configurations are set correctly
- Upload documents for the corresponding process link, perform a search test, and verify that core professional fields such as winding turns and test voltage can be retrieved normally.
- Adjust the
similaritythreshold range, compare the relevance of retrieved results to confirm alignment with precision requirements for the business scenario. - Enable
usingReRankanddatasetSearchUsingExtensionQuery, and verify the ranking and expansion effects of professional parameters. - Perform retrieval grouped by process link, confirm that only production ledger data for the corresponding link is retrieved, with no irrelevant cross-process content included.
The values provided are common starting points and should be measured against the reader's own samples.