What this category of data looks like
Business data sources for iron and steel smelting and continuous casting include iron melt and scrap steel data, L2 process control data, laboratory composition reports, MES systems, energy medium systems, and equipment status data. Data update frequencies cover three levels: per heat, real-time, and per shift. Daily documents include heat plans, batching lists, converter smelting records, continuous casting process records, and slab determination sheets. Fields include business indicators with clear units such as carbon content, tapping temperature, and casting speed. Most individual documents exceed 1,000 characters in length.
What constraints these characteristics impose on workflow orchestration
Data is grouped by heat and cast sequence, with varied update frequencies. Workflow trigger nodes must bind to heat events or real-time data update signals. This avoids invalid execution caused by fixed-period triggers.
Multi-source heterogeneous data and long-document structures are present. Workflows must support parallel data pulling across multiple nodes. They must also limit the file set scope for knowledge base searches.
Highly coupled smelting and continuous casting processes require workflow orchestration. Nodes for steelmaking and continuous casting stages must be linked. This avoids process fragmentation.
High-risk consequences come with errors. Workflows must include failure retry and result verification links. This reduces losses from abnormal heats.
How to set configurations
| Configuration Item | Recommended Setting | Rationale for This Setting |
|---|---|---|
| Knowledge base search scope | Only include the file set associated with the current heat | Iron and steel smelting and continuous casting data is archived by heat and cast sequence. Limiting the scope can exclude irrelevant process documents |
| Number of search results returned per query | Top 8 | Individual process documents are long in length, and too many results will exceed the context capacity limit |
| Batch execution node loop trigger | Based on the maximum number of heats per shift | Matches the production processing scale of a single shift, avoiding invalid loops |
variables | Bind heat ID, cast sequence number, composition indicators | Used to accurately retrieve process records and standards for the corresponding heat |
nodeId | Bind the exclusive knowledge base node ID for process specifications | Ensure workflow calls point to the knowledge base where smelting and continuous casting standards have been uploaded |
| Timeout threshold | 600 seconds | Adapts to the time requirements of multi-source data pulling and long-document retrieval |
| Maximum context length | 8000 token | Adapts to the long-text retrieval requirements of smelting process documents |
Three common mistakes
- Workflows cannot obtain the associated ID of the target knowledge base, and the corresponding field displays as empty on the interface. The cause is that the access permission for the corresponding iron and steel smelting and continuous casting process knowledge base is not bound in the workflow configuration, or the
nodeIdparameter is not correctly configured to point to the target knowledge base node. - The batch execution node can complete the full process during online debugging, but cannot execute all loops when called via API. The cause is that the API request does not configure the timeout threshold for batch execution, or the number of concurrent requests exceeds the system's carrying capacity, resulting in incomplete processing of some heats.
- The large model temperature parameter setting button disappears in variable reference mode, and direct configuration is not possible. The cause is that in FastGPT v4.17.0, the temperature parameter must be passed via
variablesin variable reference mode, and direct setting on the node interface will not take effect.
How to confirm the configuration is complete
- Trigger the workflow, then check the return results of the knowledge base search node, and confirm that only process documents associated with the current heat are included.
- Call the API to execute the workflow, then check the running logs of the batch execution node, and confirm that all heat loops are completed and no timeout errors occur.
- Check whether the
variablesparameter correctly binds business fields such as heat ID and cast sequence number, and confirm that the temperature parameter for large model calls has been passed via variables. - View the workflow status monitoring panel, and confirm that the node running time does not exceed the preset 600-second timeout threshold.
The values provided here are common starting points, and should be evaluated against the reader's own samples.