What this category of data looks like
Data sources for this category include molten iron and scrap steel charging data, L2 process control data, laboratory composition reports, MES production data, energy medium system parameters, and equipment condition monitoring data. Data update frequencies cover per heat, real-time, and shift-based intervals. Daily documents include heat plans, batching lists, converter smelting records, continuous casting process records, and slab judgment sheets. Fields include heat number, steel grade, smelting duration, endpoint temperature, casting speed, defect type, and others. Formats vary slightly based on production lines.
Constraints on Model Access and Configuration
Data sources for this category are scattered, and update frequencies vary significantly. The model access link must support parallel reading of multiple data sources, and adapt to mixed input of real-time streams and batch archived data. Daily document formats are inconsistent. Parsing rules with customizable field mapping must be configured to avoid hardcoding for single document types. The coupling between heats and casting sequences is strong. Model input must associate related IDs from upstream and downstream processes to ensure complete quality traceability links. The high-temperature continuous production process cannot be shut down arbitrarily. Configuration changes must support gray release and rapid rollback to avoid disrupting production rhythms.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the total length of multiple documents associated with a single heat, covering core fields of converter, continuous casting, and slab records |
maxTokens | 4000 token | Limits the output length of a single analysis round to avoid exceeding the necessary information scope for steel process analysis |
quoteMaxToken | 2000 token | Restricts the length of referenced source document fragments, ensuring only process parameters directly related to quality defects are retained |
| Workflow/API tool selection | auto | Automatically selects tools to meet multi-data source reading and associated query requirements |
temperature | 0.2–0.4 | Reduces randomness to ensure process analysis results align with actual production rules |
AIPROXY_API_ENDPOINT | Fill in the proxy address deployed for the production line | Adapts to access requirements of enterprise internal network isolation |
Three Common Configuration Mistakes
- Retrieval results only return data for a single heat, and cannot associate upstream and downstream casting sequence information. The root cause is that associated field mapping rules for heat numbers and casting sequence numbers are not configured, leading to broken data traceability links.
- External tool calls fail, returning status code 400. The root cause is failure to adapt to the tool's SSE protocol format, only using a generic call script.
- Model analysis results show deviations that do not conform to steel process logic. The root cause is that the
temperatureparameter value is outside the reasonable range, resulting in excessive analysis randomness.
How to Verify Successful Configuration
- Import a single heat's converter smelting record and slab judgment sheet, confirm whether the model can associate all related data for the corresponding heat number.
- Call the configured external tool to verify that query results for process parameters can be returned normally.
- Adjust the
temperatureparameter, observe changes in the randomness of analysis results to confirm they meet expected behavior. - Check the FastGPT v4.17.0 system logs to confirm there are no error messages related to data source reading or tool calls.
The configuration values listed above are common starting points. They should be evaluated against the reader's own production samples.