What the data for this category looks like
Data for this category comes primarily from multiple business systems and production equipment at EMS factories, consumer electronics factories, and automotive electronics factories. Daily documents include BOM and coordinate files, line change lists, stencil and furnace temperature records, first article confirmation reports, SPI/AOI inspection reports, maintenance records, and traceability labels. Data sources cover customer-provided BOMs, Gerber files, coordinate data, MES systems, solder paste and component batch ledgers, placement machines, SPI/AOI inspection equipment, and reflow oven temperature collection systems. Data updates run in real time per work order and each line change. Some batch ledgers synchronize per production cycle.
What constraints these characteristics impose on knowledge base retrieval and recall
Frequently updated real-time business data requires retrieval and recall to support low-latency incremental synchronization. Outdated data can cause errors in process documents and exception reports. Multiple types of structured documents—including BOM tables, furnace temperature curve reports, and defect inspection logs—require retrieval to adapt to varied data formats. It prioritizes recall of part numbers, reference designators, or defect types matching the current work order. Data scattered across multiple datasets requires configuration to support cross-dataset retrieval and quick range limitation. Work order-level update rhythms require retrieval and recall to quickly switch associated knowledge base content by production batch. This avoids interference from cross-work order data.
Configuration Settings
The following configurations are based on FastGPT v4.17.0, and are adapted to the process document and exception report drafting scenario for SMT mounting and electronic assembly:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches retrieval requirements for precise fields such as part numbers and reference designators in SMT scenarios, and filters low-relevance non-production documents |
limit | Top 10 entries | Covers the recall volume for multiple types of process reference documents under a single work order, and prevents context length from exceeding model processing limits |
searchMode | mixedRecall | Adapts to mixed retrieval scenarios for structured documents and unstructured logs, and recalls both exact matching part numbers and semantically relevant defect analysis content |
usingReRank | Enabled | Reorders initial recall results by relevance, and corrects misjudgments of defect types and coordinate positions by single semantic models |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms, supplements synonyms for alternative parts and common defects, and matches alternative part configurations in BOMs and standard terminology in exception reports |
collectionIds | List of dataset IDs bound to production work orders | Limits retrieval scope to exclusive datasets associated with the current work order, such as BOMs and line change lists, and avoids interference from cross-work order data |
Three Common Mistakes
- Symptom: After uploading Excel-format BOMs or line change lists, part numbers and reference designator fields are empty in retrieval results. Cause: Failed to enable table structured parsing configuration, and direct plain text chunking caused loss of field information.
- Symptom: Cannot retrieve content from customer-provided Gerber files by specifying a document URL. Cause: Failed to configure dataset permissions to allow external URL retrieval, or failed to correctly enable URL content parsing for
datasetSearchUsingExtensionQuery. - Symptom: Initial recall results have an overly high proportion of non-production documents unrelated to the current work order. Cause: Failed to limit
collectionIdsto datasets bound to the current work order, and retrieval scope covers the full knowledge base.
How to Verify Correct Configuration
- Upload a single BOM table, retrieve a specified part number, and check if returned results include corresponding reference designator and alternative part information.
- Trigger an exception report drafting task, retrieve a specified defect type, and check if recall results include the latest SPI/AOI inspection report data.
- Adjust the
similarityparameter value, observe changes in recall result relevance, and confirm the threshold meets the accuracy requirements of the current scenario. - After enabling
usingReRank, compare result order before and after reordering, and confirm the reordering function is active.
The values provided are common starting points and should be measured against the reader's own samples.