What Data for This Category Looks Like
Data for this category originates from material grade standards, customer technical specification documents, temperature control instrument collection data, furnace temperature uniformity test reports, hardness tester and metallographic equipment output reports, and work order records from manufacturing execution systems (MES). Daily generated documents include heat treatment process cards, furnace loading layout diagrams, real-time furnace temperature curves, hardness and metallographic test reports, equipment inspection sheets, and exception handling documents. Data is updated per furnace run or production batch. Documents contain fields such as heating temperature, holding duration, cooling rate, hardness value, and metallographic structure type. Units include professional parameters such as degrees Celsius, minutes, HV, and percentage.
Constraints on Knowledge Base Retrieval and Recall From These Characteristics
The high update frequency and per-furnace or per-batch data generation require the knowledge base to support incremental updates and rapid synchronization, to prevent retrieval of expired process parameters. Documents exist in multiple formats including structured numerical tables, time-series curves, and text specifications. This requires the retrieval system to distinguish field attributes across different document types, and perform precise matching for process parameters. Multi-source data fields have professional terminology differences. For example, the same process parameter may use different industry expressions. This requires the retrieval system to support terminology expansion. Errors directly affect the mechanical properties of parts. This requires recalled results to accurately cover key process parameters, and avoid omitting or incorrectly recalling irrelevant data.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirement for heat treatment process parameters, and avoids recalling irrelevant surface treatment or process documents from other workshops |
limit | Top 6 entries | The number of valid documents related to a single furnace run typically ranges from 5 to 7. Excessive recall increases the context processing load on the large model |
searchMode | fullTextRecall | The query examples emphasize identifiers or specialist terms; test this mode with representative data, not as a product default. |
usingReRank | Enabled | Performs a secondary ranking of initially recalled results by relevance, corrects deviations in initial retrieval, and improves the recall priority of key parameters |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands industry synonyms related to heat treatment to cover more professional query scenarios |
collectionIds | Bound to the current production line | Limits the retrieval scope to the knowledge base collection of the current workshop or production line, reducing interference from irrelevant data across production lines |
Three Common Configuration Mistakes
- Phenomenon: Single retrieval takes more than 15 seconds, and logs show vector retrieval timeout. Cause: The
collectionIdsconfiguration is not used to limit the retrieval scope, and the search covers the full knowledge base, leading to excessive vector database scan load. - Phenomenon: Recalled documents include irrelevant surface coating process content, and the large model output does not meet heat treatment scenario requirements. Cause:
searchModeis set to fuzzy matching, and precise retrieval is not implemented for heat treatment professional parameters, introducing cross-category irrelevant data. - Phenomenon: Unable to upload CSV or PDF files of furnace temperature curves, and the upload progress stalls before displaying a failure prompt. Cause: Structured table or industrial document parsing plugins are not enabled, resulting in incorrect file splitting and vectorization.
How to Verify Proper Configuration
- Upload a single heat treatment process card, run a query that includes process parameters, and check if the number of recalled results matches the
limitsetting. - Enter a query that includes specific furnace temperature parameters, and verify that the relevance ranking of recalled results aligns with the effect after enabling
usingReRank. - Switch to the knowledge base collection of another production line, and confirm that retrieval results only include documents from the current production line, with no cross-production line data included.
- Construct a non-standard process query that does not exist in the knowledge base, and verify that the processing logic meets preset requirements when no retrieval results are returned.
The values provided are common starting points, and should be validated against relevant operational samples.