What Data for This Category Looks Like
Typical entities for this category are engineering, production, quality control, and laboratory departments of printed circuit board factories. Daily processed documents include production work instructions, Gerber parsing records, drilling and lamination parameter files, chemical solution analysis reports, AOI inspection reports, electrical test reports, and sectioning reports. Data sources include Gerber files, drill tapes, and BOM files provided by customers, working drafts generated by CAM, operating parameters collected by equipment, bath solution test results, and laboratory inspection data. Update frequency follows production batches, shifts, or continuous collection. Document fields include coordinate values, aperture parameters, defect sizes, chemical solution concentrations, and more. There is no unified fixed format, with significant variation across document types. Unlike the SMT category under electronic manufacturing, the core of this category’s documents revolves around interconnection parameters for multi-layer graphics and holes, focusing on parameter control for production processes and quality inspection.
How These Characteristics Impact Citation Sources and Traceability
PCB manufacturing documents have multiple sources and are updated frequently. Citation traceability must accurately match production batches and document versions to avoid incorrect parameter references across batches. Fields vary widely across different document types. Differentiated recall rules must be configured for different document types, such as engineering parameters and inspection reports. Continuously collected real-time data requires traceability systems to support near-real-time vector database updates, ensuring cited content aligns with current production status. When a parameter error occurs, the original document and production link must be quickly located. Traceability information must include complete metadata and file identifiers to enable tracing of specific production steps and reduce abnormal risks.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Single search limit | Top 6 entries | Covers the core number of documents per production batch, avoids recalling irrelevant historical batch data |
| Similarity threshold | 0.75–0.85 | Matches the precision of PCB parameters, prevents recall of incorrect drilling aperture or lamination temperature data |
| Maximum context length | 1500–2000 characters | Retains complete production work instructions and Gerber parsing snippets |
quoteMaxToken | 900–1100 tokens | Adapts to the length of single production parameter segments, prevents truncation of critical chemical solution concentration or AOI defect descriptions |
responseData | Includes source_id, file path, inspection time | Ensures traceability information includes traceable production batches and document versions |
detail | Enabled | Retains original document metadata, such as CAM file generation time, device operation log timestamps |
Three Common Configuration Errors
- Phenomenon: After adjusting the similarity threshold and search limit, the number of recalled content remains unchanged at a fixed value. Cause: Context caching was not disabled, so each call reuses previously recalled results and cannot re-retrieve documents from the current production batch.
- Phenomenon: Output content retains Markdown syntax symbols and does not render as readable format. Cause: The prompt did not explicitly require formatted output, or relevant content formatting configuration was not enabled.
- Phenomenon: Traceability cannot find database field information for the corresponding document. Cause: Metadata extraction was not enabled when uploading documents, so source_id cannot be associated with file fields in the database.
How to Verify Correct Configuration
- Upload a test Gerber parsing record and AOI report, submit a test query that includes specific parameters, check if the returned result includes traceability information such as source_id and file name.
- Adjust the value of
quoteMaxToken, observe changes in the length of returned content to confirm the parameter takes effect. - Upload documents of the same type from different production batches, submit a query to confirm that recalled results only match documents from the current batch, with no cross-batch interfering data.
- Check FastGPT operation logs to confirm that each call generates new traceability information, with no cache reuse markers.
The values provided are common starting points and should be measured against the reader's own samples.