What the Data for This Category Looks Like
Data for this category originates from slab or billet batches, heating furnaces, mill automation control systems, online thickness and width measurement equipment, surface inspection devices, and MES systems. Daily documents include rolling plans, heating schedules, rolling specifications, shape and thickness records, surface quality reports, and quality certificates. Document fields include coil or section batch numbers, heating temperature, reduction rate, rolling speed, shape deviation, thickness tolerance, and surface defect level. Units include degrees Celsius, millimeters, megapascals, meters per second, and other standard industrial units. Data updates in real time for each coil or individual product batch.
Constraints on Knowledge Base Retrieval and Recall
This category’s data updates in real time and is generated per coil or single product batch. This requires the knowledge base retrieval and recall system to support incremental synchronization, to avoid resource consumption from full database updates. Multi-dimensional process parameters and quality fields require retrieval to support combined multi-condition matching, to accurately locate rolling and quality documents for specific batches. Traceability requirements for individual coils mean recall results must link unique identifier fields such as coil or section batch numbers. Bid documents must match process requirements for specific product specifications, so the recall scope must be limited to the document set for this category. This prevents mixing process data from unrelated categories and ensures the accuracy of bid content.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Steel rolling process parameters have high precision requirements. Low-similarity irrelevant documents must be filtered to ensure the accuracy of process data used for bids |
limit | Top 6–10 results | Bid documents need to cover multi-dimensional process requirements, while controlling context length to avoid exceeding model processing limits |
searchMode | mixedRecall | Process parameters must be matched by both semantics and keywords. For example, "12mm cold-rolled sheet rolling specification" matches both semantic and keyword fields |
usingReRank | Enabled | Retrieval results for multi-dimensional parameters require secondary ranking, to prioritize documents that better match bid specifications |
datasetSearchUsingExtensionQuery | Enabled | Specification parameters in bid requirements can be expanded into retrieval terms to improve recall coverage. For example, "Q235 section steel" can be expanded to rolling specification keywords for the corresponding material |
collectionIds | Bind exclusive dataset IDs for hot rolling / cold rolling / section steel / steel pipe | Limit the retrieval scope to the document set for this category, to avoid mixing irrelevant data from other categories |
Three Common Misconfigurations
- Issue: Retrieval results include chemical composition documents from steelmaking categories, and are not limited to rolling and finishing documents. Cause:
collectionIdsis not configured correctly, and the retrieval scope is not bound to the exclusive dataset for this category. - Issue: Returned document fields lack coil or section batch numbers, which cannot support the traceability requirements of bids. Cause: Multi-field matching rules are not configured, and only text semantics are retrieved without associating batch identifier fields.
- Issue: Too many results are returned by a single retrieval, causing model output to exceed context length limits. Cause: The
limitparameter is not set reasonably, and the number of recalled entries exceeds the processing range.
How to Confirm Proper Configuration
Based on FastGPT v4.17.0, perform the following verification steps:
- Check the
collectionIdsparameter in the retrieval configuration, confirm that only the dataset ID for this category is bound, and no irrelevant datasets are associated. - Enter a test query containing process parameters for this category, verify that returned result fields include traceability information such as coil or section batch numbers.
- Adjust the
similarityparameter range, compare recall results across different values, confirm that matching accuracy meets business requirements. - Test recall result ranking after enabling
usingReRank, confirm that documents matching the current bid specifications are returned first.
The values listed here are common starting points and should be evaluated against in-house sample data.