What the Data for This Category Looks Like
Business data for SMT mounting and electronic assembly comes from customer BOM, Gerber coordinate files, MES systems, solder paste and component batch ledgers, mounter operation logs, SPI/AOI inspection reports, reflow oven temperature records, and repair traceability tags. Data updates take effect in real time for each work order and each line change. Documents associated with a single work order include part number lists, reference designator mapping tables, line change parameter tables, defect inspection records, and more. Fields include component batch numbers, oven temperature curve parameters, defect coordinates, repair stations, and other data. Some content uses structured tables, while other content uses plain text logs.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall?
Real-time updated business data requires knowledge base retrieval to support incremental synchronization, to avoid recalling expired line change parameters or component batch information. Mixed-format document structures require the retrieval system to support both structured field matching and unstructured text semantic recall. High-frequency search terms such as part numbers and reference designators need enhanced exact match weighting. The concentrated associated data volume per work order requires retrieved results to be limited to the current work order’s associated dataset, to avoid interference from invalid cross-work order information. Additionally, numeric fields with units such as defect coordinates and oven temperature curves need support for numeric range matching retrieval.
How to Configure the Settings
This configuration is based on the knowledge base retrieval function of FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Threshold for exact search terms such as part numbers and reference designators, balances recall rate and accuracy |
limit | Top 8 entries | Number of core retrieval results from the current work order’s associated dataset, prevents information overload |
searchMode | mixedRecall | Use mixed semantic/vector and full-text retrieval. Exact-field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Performs secondary ranking of initial recall results by semantic relevance, optimizes result ranking for defect analysis queries |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands search terms to supplement associated semantics such as alternative components and line change standards |
collectionIds | Bind the associated dataset ID of the current work order | Limits retrieval scope to the current work order’s BOM, inspection reports, and other data |
Three Common Misconfigurations
- Retrieval results display garbled text or empty fields. Cause: Documents such as BOM and line change tables were imported without unified UTF-8 encoding, or document encoding formats were not configured correctly, leading to abnormal parsing of Chinese fields.
- Retrieval results cannot match the table of contents structure of Word documents. Cause: The configuration to split Word-format structured documents by table of contents was not enabled, causing long documents to be randomly truncated and losing semantic associations linked to the table of contents.
- Custom Qdrant vector database cannot be properly connected for retrieval. Cause: Connection parameters were not configured according to FastGPT’s dataset access specifications, and the vector database collection corresponding to
collectionIdswas not specified, preventing retrieval from reading data in the custom database.
How to Confirm the Configuration Is Valid
- Import the BOM file and SPI inspection report for the current work order, run the search "solder paste batch number for this work order", and verify that the returned results include the corresponding batch ledger information.
- Upload a Word line change table with a table of contents structure, run the search "stencil parameters during line change", and verify that the returned results match the chapter content corresponding to the table of contents.
- Connect the custom Qdrant vector database, bind the corresponding
collectionIds, run the search "mounter operation logs", and verify that the returned results include log data from the custom database. - After enabling
usingReRank, run the search "analysis method for solder joint defects", and verify that the returned results are sorted to prioritize professional content related to defect analysis.
The values provided here are common starting points and should be measured against the reader's own samples.