What Data Looks Like for This Category
Data for this category comes from slab or billet batch ledgers, furnace operating parameters, mill automatic control logs, online thickness and width measurement data, surface inspection results, and MES system records. Daily documents include rolling plans, heating schedules, rolling specifications, shape and thickness records, surface quality reports, and quality certificates. Fields include batch number, thickness and width values, surface defect level, furnace temperature and rolling speed parameters, and more. Data updates per coil or individual steel piece. Key process and quality parameters are synced in real time.
Constraints Imposed on Knowledge Base Retrieval and Recall
Data for this category updates per coil or individual piece and is synced in real time. Retrieval and recall must quickly match the latest batch process and quality records, and avoid calling outdated data. Documents include multi-dimensional process parameters and quality inspection results. Retrieval must accurately associate content such as heating schedules and shape records corresponding to the current audit batch. Batch numbers and specification parameters vary significantly between different steel coils. Retrieval must use clear field filtering to avoid interference from irrelevant documents. Quality certificates and surface quality reports are strongly linked to downstream inspection and audit work. Recall must prioritize matching exclusive documents for the current batch.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches strongly correlated parameters such as rolling process and quality records, and avoids recalling documents from unrelated batches |
limit | Top 8–12 entries | The number of related documents for a single steel piece batch is limited. Excessive recall will exceed the context window |
searchMode | fullTextRecall | The query examples emphasize identifiers or specialist terms; test this mode with representative data, not as a product default. |
usingReRank | Enabled | Secondary sorts recall results by relevance to the current audit batch, and prioritizes displaying core documents such as heating schedules and shape records |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms such as batch number and specification parameters to supplement matching hidden related process documents |
collectionIds | Bind the rolling document collection for the current production line | Restrict retrieval scope to exclusive documents such as rolling plans and quality records for this production line, and avoid interference from cross-production line data |
Three Common Misconfigurations
- The interface displays a full knowledge base search entry by default, and cannot limit the current production line scope. Cause:
collectionIdsis not configured to bind the exclusive document collection for the current production line, causing the system to load full knowledge base content. - After online deployment, garbled characters appear in imported knowledge base documents, while local deployment has no issues. Cause: The online environment is not configured with the same document encoding format as the local environment, or documents are not imported in the structured format required for rolling documents.
- When initiating a query for a specific steel coil batch, the recall results include a large number of non-current batch documents. Cause: A reasonable
similaritythreshold is not set, ordatasetSearchUsingExtensionQueryis not enabled, leading to irrelevant data being recalled via fuzzy matching.
How to Verify Successful Configuration
- Check if the
collectionIdsconfiguration only includes exclusive document collections such as rolling plans and quality records for the current production line. - Initiate a query for a specific steel coil batch number, and verify that the recall results only include documents corresponding to the heating schedule, shape records, etc., for this batch.
- Check if the
similaritythreshold has been empirically calibrated based on the process parameter correlation degree of the current production line. - Confirm that
usingReRankis enabled, and the sorted recall results prioritize displaying documents directly related to the current audit task.
The values provided here are common starting points, and should be evaluated against the reader's own samples.