What data looks like for this use case
Data in PCB manufacturing scenarios covers daily documents from four core teams in printed circuit board factories: engineering, production, quality control, and laboratory teams. Documents include engineering production instructions, Gerber parsing records, drilling and lamination parameters, chemical solution analysis reports, AOI reports, electrical test reports, and sectioning reports.
Data sources include customer-provided Gerber, drill tape, and BOM files, CAM design files, equipment operating parameters, bath solution test results, AOI and electrical test data, and laboratory sectioning analysis records.
Update frequencies include per production batch, per shift, and real-time continuous collection. Each production batch generates multi-format, multi-dimensional process and exception-related documents.
Constraints on vector models and indexing workflows
Multi-source, heterogeneous data formats and sources require the indexing module to support vector extraction and synchronization for multiple document types, including binary files, tables, and plain text.
High-frequency, batch-level production data updates require incremental synchronization for indexing, to avoid performance losses caused by full reindexing.
The strong correlation between process parameters and exception reports requires vector recall to accurately match field associations, to ensure targeted retrieval results.
Differences in document lengths, ranging from single drilling parameter entries to complete sectioning reports, require chunking strategies adapted to different text units, to avoid truncation of critical associated information.
Configuration settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Covers complete process parameter sections or exception report paragraphs, to avoid truncation of critical information |
| Overlap Length | 100–200 characters | Retains overlapping content between adjacent chunks, to ensure cross-chunk associated information can be fully matched |
maxToken | Set to the selected embedding model/service maximum supported input tokens | maxToken is the input-token maximum supported by the selected embedding model/service; verify it in that provider configuration. See the v4.17.2 model-field reference. It is not retrieval/LLM context or a FastGPT-wide fixed cap. |
defaultToken | Set based on actual measurement | Matches the maximum single-data length limit of local vector storage |
| Similarity Threshold | 0.75–0.85 | Filters low-relevance non-target documents, to ensure retrieval results are strongly correlated with process and exception scenarios |
| Single Search Limit | Top 10 results | Controls the total length of retrieval results, to avoid exceeding context window limits |
Three common configuration errors
- Symptom: Retrieving exception reports returns a large number of unrelated bath solution test data. Cause: The similarity threshold is set too low, failing to filter low-relevance non-target documents, leading to incorrect matching of non-process-related content. This may cause process adjustment deviations, resulting in scrap of an entire production batch.
- Symptom: After importing a Gerber binary file, the vector index generation log shows an unsupported format error. Cause: Text extraction configuration for binary files is not enabled, so Gerber files cannot be converted to vectorizable text content.
- Symptom: After triggering an incremental update, process files for new production batches cannot be retrieved. Cause: Incremental synchronization trigger rules are not configured. Only a one-time full index build is performed, failing to cover real-time updated production data.
How to verify correct configuration
- Import a standard drilling parameter document, check that the vector index generation log has no format errors.
- Enter exception keywords from an AOI report, verify that the retrieval results return associated lamination process documents.
- Trigger an incremental synchronization task, confirm that new production batch data is correctly included in the index scope.
- Call a test interface to perform a retrieval, verify that the returned content meets preset associated matching requirements.
The values provided here are common starting points, and should be evaluated against the user’s own samples.