What data for this category consists of
Daily data for SMT mounting and electronic assembly covers multiple types of professional documents. These include BOM and coordinate files, changeover lists, stencil and reflow oven temperature records, first article inspection reports, SPI/AOI inspection reports, maintenance records, and traceability labels. Data sources include customer-provided BOMs, Gerber files, coordinate files, MES systems, solder paste and component batch information, and real-time collected data from placement machines, SPI/AOI equipment, and reflow ovens. Data updates run in real time per work order or line change. Some device-collected data updates at second-level intervals. Document fields include part number, reference designator, component specification, X/Y/Z axis coordinates (unit: mm), temperature curve parameters (unit: degrees Celsius, seconds), and additional relevant fields.
What constraints these characteristics impose on knowledge base retrieval and recall
The multi-source, high-frequency update, and precise field characteristics of documents in this category create multiple constraints for knowledge base retrieval and recall. First, real-time updates per work order or line change require the knowledge base to support incremental synchronization and scheduled full refreshes. This prevents data lag from disrupting cross-review. Second, the need for precise fields across multiple document types requires retrieval to distinguish between different data types such as BOM part numbers, coordinate parameters, and furnace temperature curves. This avoids cross-category interference. Third, high accuracy requirements for review scenarios such as solder joint defect detection and part number verification require controlling the proportion of low-relevance recall results. It also covers extended retrieval needs for professional terminology.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precise field requirements of SMT documents, prevents recall of low-relevance non-part number or coordinate data |
limit | Top 8 entries | Covers the quantity requirements for multiple document types such as BOMs, changeover lists, and AOI reports involved in a single work order |
searchMode | fullTextRecall | Use full-text retrieval for text content. Field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Improves ranking accuracy for professional documents such as solder joint defects and furnace temperature curves, reduces false recall probability |
datasetSearchUsingExtensionQuery | Enabled | Supplements extended review questions for users, for example, expands "part number verification" to "BOM part number and component batch matching" |
collectionIds | Collection IDs grouped by document type | Differentiates documents of different categories such as BOMs, AOI reports, and furnace temperature records, avoids cross-category data interference |
Three common mistakes
- Symptom: Retrieved <Reference> fields do not include matching BOM or AOI report content. Cause: Failed to associate the knowledge base collection for the target category of documents, or did not enable the
datasetSearchUsingExtensionQueryextended retrieval function. - Symptom: Cannot export the list of file names in the knowledge base via the interface. Cause: Did not enable document metadata indexing configuration, and did not enable metadata recall in retrieval parameters.
- Symptom: 504 timeout error occurs in a single retrieval task. Cause: The
limitsetting exceeds 8 entries andusingReRankis not enabled, leading to a large number of documents participating in similarity calculation, exceeding the processing threshold of FastGPT v4.17.0.
How to confirm the configuration is correct
- Upload a test BOM file, retrieve "verify the reference designator for part number A123", check if the returned results include the corresponding content of this BOM.
- Check if the knowledge base metadata index switch is enabled, attempt to retrieve any document, confirm that the returned results include the file name field.
- View the FastGPT v4.17.0 logs, confirm that the retrieval request carries the correct
collectionIdsandsimilarityparameter values. - Initiate a retrieval containing the keyword "furnace temperature curve", confirm that the returned results include the corresponding reflow oven temperature record documents.
The values provided are common starting points and should be measured against the reader's own samples.