What this category's data looks like
Heat treatment scenario data originates from material grade standards, customer technical requirements, temperature control instruments, furnace temperature uniformity testing equipment, hardness testers and metallographic equipment, and manufacturing execution systems (MES). Daily documents include heat treatment process cards, charging charts, furnace temperature curves, hardness and metallographic reports, equipment inspection forms, and exception handling sheets.
Data is updated per furnace run or batch. Single entries contain structured fields such as heating temperature (unit: ℃), holding time (unit: h/min), cooling medium, hardness values (unit: HV or HRC), alongside time-series furnace temperature curve data.
What constraints these characteristics impose on the vector models and indexes step
High-frequency updates per furnace run require indexes to support incremental synchronization, avoiding performance loss from full index reconstruction. Multiple document types include text, tables, and time-series furnace temperature curves. This requires adapting different chunking logic to preserve the semantic integrity of process parameters and curve segments.
Fields with specific units (such as ℃, HV, h) require vector encoding to retain the semantic meaning tied to the unit, avoiding confusion during parameter matching. Single furnace run data has a small individual volume but a large cumulative total. Reasonable planning of index sharding is needed to ensure retrieval efficiency.
The business requirement for precise matching of process parameters means recall relevance thresholds must be strictly adapted to this scenario.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the single-section content length of heat treatment process cards and furnace temperature curves, avoiding chunking across process parameters |
| Overlap Length | 100–200 characters | Covers time node continuity for furnace temperature curves, preventing loss of continuous temperature sequences across chunks |
| Similarity Threshold | Calibrated based on actual measurements | Meets the business requirement for precise verification of process parameters, avoiding mismatched heating and holding regimes |
maxToken | 8192 | Adapts to the maximum input length of Doubao-embedding-large, compatible with long furnace temperature curve text |
| Single Search Limit | Top 8 results | Covers the recall requirement for multiple sets of process parameters per furnace run, avoiding omission of key verification items |
defaultToken | 1024 | Sets the default chunk token limit, balancing index storage and recall accuracy |
Three common configuration errors
- Direct errors when testing a custom model, with the interface returning "request failed" or "apikey invalid". The cause is incorrect configuration of the custom request address and apikey, or failure to bind model permissions to the corresponding knowledge base.
- The knowledge base page still displays "No available index model detected" after refresh. The cause is failure to associate the configured vector model in knowledge base settings, or that the model configuration was not saved and activated.
- Table documents (such as charging charts and equipment inspection forms) cannot be indexed correctly. The cause is failure to enable multi-vector support, or failure to adapt chunking to the multi-column field structure of tables, resulting in truncated table content or incorrect semantic splitting.
How to confirm the configuration is complete
- Enter the knowledge base settings page of FastGPT v4.17.0, and check if the configured Doubao-embedding-large appears in the vector model dropdown list.
- Upload a single heat treatment process card or furnace temperature curve document. After indexing completes, manually enter process parameter keywords to verify that the recall results include the corresponding document fragments.
- View the index management page, confirm that the generated index files have no abnormal error logs, and that the file size matches expected values.
- Create a new conversation agent, enter queries related to heat treatment processes or exception analysis, and verify that returned results are associated with correct document content.
The values provided here are common starting points, and should be evaluated against the reader's own samples.