What This Category of Data Looks Like
Data sources for this category include BOM, Gerber, and coordinate files provided by customers, factory MES systems, solder paste and component batch ledgers, placement machines, SPI/AOI inspection equipment, and reflow oven temperature records. Daily documents cover changeover sheets, stencil archives, first article confirmation sheets, defect inspection reports, maintenance records, and traceability labels. Data updates occur in real time per work order and each changeover. Single data entries include fields such as part number, position number, component batch, furnace temperature parameters, and defect coordinates. Units include degrees Celsius, millimeters, and others.
Constraints Imposed by These Characteristics on Vector Models and Indexes
The high-frequency, real-time update feature of this category of data requires vector indexes to support incremental synchronization, avoiding performance loss caused by full index reconstruction. The presence of multi-dimensional fields such as component coordinates, furnace temperature parameters, and defect positions requires vector models to capture cross-field semantic associations, ensuring that retrieval can correlate and match relevant production process data. Document types include structured ledgers and unstructured inspection reports, so indexes must be compatible with text parsing logic for different formats. Traceability data has high requirements for field relevance, so indexes must balance semantic retrieval and precise field matching capabilities to prevent mismatching production data from different work orders.
How to Set Configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the semantic integrity of single segments of SMT ledgers and inspection reports, avoiding confusion of cross-work order information |
| Overlap Length | 50–80 characters | Retains contextual association at changeover and work order handover points, preventing semantic breaks |
| Similarity Threshold | 0.72–0.85 | Matches the semantic similarity of part numbers and defect descriptions, balancing recall rate and accuracy |
| Single Search Limit | Top 20 entries | Covers inspection data from multiple devices associated with a single work order, avoiding redundant results |
defaultToken | Set from representative sample documents | defaultToken is the default token count for direct text chunking; tune it on representative source documents. See the v4.17.2 model-field reference. It is not a vector dimension, embedding-model input cap, or retrieval/LLM context. |
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. |
Three Common Misconfigurations
- Phenomenon: Vector database timeout errors occur during repeated queries of production ledgers. Cause: No caching mechanism is configured, and each query triggers full vector retrieval, leading to excessive resource consumption under high-frequency query loads.
- Phenomenon: No index update prompt appears when adding data to a collection after creating a training order. Cause: The trigger logic for training orders and collection data addition is confused. Training orders are used for batch preprocessing of unstructured documents, and single-data indexing does not require triggering via a training order.
- Phenomenon: Retrieval results are empty or have extremely low matching accuracy after uploading defect images. Cause: No dedicated image vectorization configuration for SMT defect images is set up. Only the default text vectorization model is used to process image data, which cannot extract image features.
How to Confirm the Configuration Is Correct
- Initiate a simulated work order query, verify whether the returned result fields include core production information such as part numbers and position numbers, and confirm that the semantic matching logic is active.
- View the vector index update log, confirm whether high-frequency updated changeover data completes index synchronization within a reasonable time, and verify that the incremental update configuration is effective.
- Test retrieval results for different document types such as BOM tables and SPI inspection reports, confirming that parsing and vectorization logic is compatible with different data formats.
- Adjust the similarity threshold and observe changes in the number of retrieval results, confirming that the threshold configuration meets the matching requirements of the current business.
The values listed above are common starting points and should be evaluated against the reader's own samples.