Data Profile for This Category
Core data for this category comes from hot metal and scrap steel ledgers, L2 process control systems, laboratory test results, MES production systems, and energy medium and equipment condition monitoring modules. Daily generated documents include furnace plans, batching lists, converter smelting records, composition reports, continuous casting process records, and slab judgment sheets. These documents contain fields such as furnace identification, carbon content, tapping temperature, and casting speed, with corresponding testing and process units attached. Update frequency follows these patterns: Smelting records are generated per furnace, process control data is synchronized in real time, and scheduling and quality audit documents are aggregated per shift.
Constraints for Knowledge Base Retrieval and Recall
Real-time and furnace-level update rhythms require knowledge base retrieval to support quick filtering and recall of results by furnace identification and production time period. This prevents irrelevant historical data from being included in results. Multi-source and heterogeneous document structures, which include both structured parameters and semi-structured records, require retrieval modes that combine field-level matching and full-text search. Data is strongly tied to production operations, and errors can lead to furnace downgrade or continuous casting interruption. This means the relevance threshold for recall results must strictly align with the business scenario. The strong coupling between furnace and casting sequence requires retrieval and recall to support queries linked to the full production chain, covering data from steelmaking to continuous casting.
Configuration Settings
This configuration is based on FastGPT v4.17.0, and is adapted to the professional data characteristics of iron and steel smelting and continuous casting scenarios.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Iron and steel smelting uses dense professional terminology, so low-correlation results must be filtered to avoid incorrect recall of records from non-corresponding furnaces |
limit | Top 8 entries | The number of valid documents associated with a single furnace is limited. Too many recalls will exceed the context window and reduce audit efficiency |
searchMode | fullTextRecall | Use full-text retrieval for text content. Field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Semantic similarity of multi-source heterogeneous data requires secondary calibration to improve the accuracy of professional term matching |
datasetSearchUsingExtensionQuery | Calibrated based on actual measurements | Professional queries in smelting scenarios need to be optimized and expanded to avoid missing associated process parameters |
collectionIds | Bind dedicated datasets for ironmaking, steelmaking, continuous casting, and quality audit | Non-category knowledge base content must be isolated to ensure that recall results accurately match the business scope |
Three Common Mistakes
- Phenomenon: A
Validation failed: name: Pathnameis required.error is returned when creating a knowledge base or generating a dataset via a static link. Cause: The name field in the basic knowledge base configuration is not filled in, or the required information of the associated collection is not correctly specified in the static link generation process. - Phenomenon: The input and response reference content of the knowledge base search is forcibly displayed in the answer, and only the audit conclusion cannot be shown. Cause: The reference display switch in the answer is not turned off, or the corresponding parameter for hiding tool output is not configured when calling the workflow.
- Phenomenon: A large number of historical data from non-current furnaces are included in the recall results, interfering with audit judgments. Cause: The logic for filtering recall results by furnace identification or production time period is not configured, or the
limitparameter is set too large, causing irrelevant data to be recalled.
How to Verify Successful Configuration
- Execute a single-furnace inspection and audit query, check whether returned results only include relevant documents for the corresponding furnace, with no irrelevant historical data mixed in.
- Trigger a workflow call or direct knowledge base retrieval, view the answer content, and confirm whether the reference module of the knowledge base search is displayed or hidden as required.
- Attempt to create a knowledge base and static link dataset, confirm that there are no error messages about missing required fields.
- Adjust the value range of the
similarityparameter, verify whether the relevance of the recall results meets the requirements of business audits.
The values provided are common starting points and should be measured against samples relevant to the specific deployment.