What the data for this category looks like
Business data sources for SMT mounting and electronic assembly include BOMs, Gerber files, and component coordinates provided by clients, plus production node data from MES systems, placement machines, SPI/AOI inspection equipment, and reflow oven temperature recorders. Daily produced documents include line change sheets, stencil parameter records, first article confirmation reports, repair traceability labels, and component batch ledgers. Data updates occur once per work order, and synchronize each time production lines are changed. Some device data pushes in real time. Document fields cover part numbers, reference designators, component specifications, reflow temperature curve parameters, and inspection defect records. Some files use structured table formats.
Constraints for knowledge base retrieval and recall
The business data characteristics of SMT mounting and electronic assembly create multiple constraints for knowledge base retrieval and recall. Frequent real-time updates require retrieval systems to support incremental synchronization and real-time indexing. This prevents inconsistencies between old and new production data. Numerous specialized fields such as part numbers, reference designators, and reflow temperature curve parameters require retrieval to balance precise keyword matching and semantic understanding. This ensures accurate process data is cited in bid documents. Single work orders use exclusive production data. Retrieval scopes must be limited to the document set for that work order, to avoid irrelevant data interfering with quotations and bid content. Structured table documents make up a notable share of the dataset. Retrieval must adapt to field-level matching logic to improve recall accuracy.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75-0.85 | Matching SMT specialized terminology and process data requires a high similarity threshold, to avoid irrelevant results being included in bid reference content |
limit | Top 8 entries | The volume of reference data required for bid documents is moderate; controlling the number of returned results avoids context length exceeding processing limits |
usingReRank | Enabled | Semantic matching for specialized process documents requires reranking to optimize result order, improving the priority of core reference data |
searchMode | mixedRecall | Balances precise keyword requirements for part numbers, reference designators, etc., and semantic understanding requirements for process descriptions, adapting to multi-type queries in bid scenarios |
datasetSearchUsingExtensionQuery | Enabled | Bids require expanding queries to include associated derivative data such as line change sheets and temperature records, improving the completeness of reference content |
collectionIds | Bind to the knowledge base collection for the corresponding work order, no more than 5 | Precise limitation of retrieval scope, avoiding interference from cross-work order irrelevant production data with quotation and bid content |
Three Common Misconfigurations
- Issue: Retrieval results do not return expected BOM part numbers or reflow temperature parameter data. Cause:
collectionIdsis not configured correctly, so the retrieval scope covers document sets from non-target work orders. - Issue: The first query returns no results, but the second query successfully recalls the corresponding data. Cause: There is a delay in knowledge base incremental synchronization. The newly uploaded work order document has not completed indexing during the first query.
- Issue: Original HTTP request data is not correctly identified as reference content. Cause: The request data is not converted to a standard knowledge base document chunk structure, and extended query adaptation is not enabled via
datasetSearchUsingExtensionQuery.
How to Verify Correct Configuration
- Upload a test SMT work order BOM document. Run a query with specific part number keywords, and confirm the returned results include the document’s corresponding content.
- Adjust the value range of the
similarityparameter. Observe changes in retrieval result relevance, and confirm the matching logic meets business requirements. - Simulate uploading a new line change sheet document. Run a query with a line change number, and verify the newly uploaded document can be recalled in real time. This confirms the synchronization configuration is active.
- Configure multiple knowledge base bindings. Run a cross-collection query, and verify
collectionIdscorrectly limits the retrieval scope to only return content from the target collection.
The values provided here are common starting points and should be measured against the reader's own samples.