What the Data for This Category Looks Like
Data for this category comes from iron melt and scrap steel ledgers, L2 process control systems, laboratory test reports, manufacturing execution systems, energy medium monitoring platforms, and equipment condition monitoring systems. Daily documents include heat plans, batching lists, converter smelting records, composition test reports, continuous casting process parameter records, and slab quality judgment sheets. Data update frequency is per heat, real-time, or per shift. Data fields include heat number, cast number, hot metal temperature, final smelting composition, continuous casting pulling speed, defect type code, and more. Temperature is measured in degrees Celsius, and time is measured in minutes or hours.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
The high update frequency and multi-source heterogeneous nature of this category’s data require knowledge base retrieval to support real-time incremental synchronization of multi-system data, to avoid retrieval results lagging behind actual production. The tightly coupled business logic between heats and casts requires retrieval and recall to associate fields such as heat number and cast number, to ensure retrieval results match the current production batch. Documents with a high proportion of structured fields require retrieval to support precise matching of numeric fields. Relying only on text keywords cannot cover all key information. Meanwhile, the wide scope of impact from consequences involved in quality problem traceability requires retrieval and recall results to have high reliability, to avoid incorrect association of historical data.
How to Configure Settings
This page is written based on the configuration rules of FastGPT v4.17.0.
| Config Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Steel smelting data contains a large number of structured numeric fields. This threshold filters low-correlation non-production documents while retaining matching results for key process parameters. It can be adapted to different heat types after calibration based on actual measurements. |
limit | Top 8–12 entries | The number of documents related to a single heat is limited. This range ensures the context window fits while covering all key associated documents. |
searchMode | embedding | Use semantic/vector retrieval. Exact-field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Quality traceability requires precise matching of defect types and process parameters. Reranking improves the sorting priority of highly relevant results and reduces the probability of incorrect associations. |
datasetSearchUsingExtensionQuery | Enabled | It can automatically expand retrieval to documents related to casts, batching, and more, based on heat numbers, adapting to the tightly coupled business logic between heats and casts. |
collectionIds | Bind corresponding knowledge base collections by production workshop/process | Data from each steel smelting process is independent and has different update frequencies. Binding corresponding collections avoids retrieving historical data from unrelated processes. |
Three Common Mistakes
- Phenomenon: Retrieval results include smelting records from unrelated workshops. Cause: The knowledge base collection for the corresponding process was not bound via
collectionIds, resulting in the retrieval scope covering all production data. - Phenomenon: Multi-level directory continuous casting process documents are not fully recalled. Cause: The full directory recursive parsing rule for the knowledge base was not configured, only document files in the top-level directory were recognized.
- Phenomenon: A
common:core.chaterror is returned after triggering retrieval. Cause: The knowledge base collection bound viacollectionIdswas not adjusted according to the real-time updated heat batch, resulting in the referenced collection not existing or having abnormal permission configuration.
How to Confirm the Configuration Is Correct
- Manually enter the heat number and temperature parameters of a specific heat, check whether the retrieval results include the smelting record and slab judgment sheet for that heat.
- View the
collectionIdslist configured in the knowledge base, confirm that it only contains the process document collection for the current workshop. - After enabling
usingReRank, compare the sorting of retrieval results before and after enabling, confirm whether the ranking of highly relevant documents has improved. - Test uploading continuous casting process documents in multi-level directories, confirm that documents at all levels are included in the retrieval scope.
The values provided are common starting points and should be measured against the reader's own samples.