What the data for this category looks like
Business data for SMT mounting and electronic assembly comes from customer BOMs, Gerber files, coordinate files, MES systems, mounters, SPI/AOI testing equipment, and reflow oven temperature records. Daily documents include BOM tables, line change sheets, stencil and oven temperature records, first article confirmation sheets, SPI/AOI reports, maintenance records, and traceability labels. Data updates occur per work order, each line change, and in real time. Document fields include part number, reference designator, component batch number, oven temperature parameters, coordinate values, and more. Corresponding units are pieces, degrees Celsius, millimeters, batch numbers, and work order numbers.
Constraints imposed by these characteristics on the source citation and traceability link
This category’s data updates frequently and has precise fields. The traceability link must match the latest work order and equipment data in real time, and avoid using expired documents. Multiple document types coexist, so multi-format indexing and precise field recall are required to locate key information such as part numbers and reference designators. Errors directly affect complete machine quality. The traceability link must ensure cited sources match actual production data exactly, with no mismatching or omissions. Individual documents have long content, so cited fragment length must be controlled reasonably to avoid interference from irrelevant information in results.
How to configure the system
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Adapts to the core content length of a single SMT document, avoiding truncation of key traceability fields such as part numbers and batches |
responseData | Enabled | Forces return of original data from cited sources to meet traceability requirements for inspection records |
detail | Full field display | Displays all key fields in documents to facilitate verification of part numbers, component batches, and other information |
| Single recall limit | Top 5 entries | Matches the single-batch data volume of SMT work orders, avoiding redundant recall that interferes with core results |
| Document chunk length | 800 characters | Adapts to the content structure of BOM tables and SPI reports, ensuring complete field associations are retained after chunking |
| Variable reference format | System preset standard format | Adapts to FastGPT’s variable calling rules, ensuring correct binding of data sources |
Three common configuration errors
- Returning cited content without enabling the citation switch: The page shows cited fragments have been returned, but the global citation switch is disabled. This occurs because default configurations changed after the FastGPT v4.17.0 version upgrade, and the global setting item was not checked.
- Mismatched knowledge base cited files and actual business documents: Query results return traceability files that are not the latest documents for the current work order. This occurs because indexes were not updated by work order dimension, and real-time synchronized MES and equipment data sources were not bound.
- Invalid variable reference format: Entering the
[{datasetId: xxx}]format fails to call the target knowledge base. This occurs because non-standard variable syntax was used, and system preset variable calling rules were not followed.
How to verify correct configuration
- A test SPI report is uploaded, a query containing the report’s defect description is initiated, and returned results are checked for the report’s file name and corresponding fields.
- The
quoteMaxTokenparameter is viewed via the configuration page, and its value is confirmed to fall within the 800–1200 token range. - The preset variable reference format is entered, a test query is initiated, and correct calling of bound data source content is confirmed.
- A line change scenario is simulated to update the corresponding document, a query including line change requirements is initiated, and recall of the latest version of the line change sheet is confirmed.
The values provided are common starting points and should be measured against the reader's own samples.