What data for this category looks like
Data for this category originates from the electrode sheet, assembly, formation, and quality departments of power, energy storage, and consumer battery enterprises. Daily documents include batching records, coating and rolling parameters, winding or stacking records, liquid injection records, formation and grading data, OCV and safety test reports. Data is sourced from raw material batch ledgers, MES systems, slurry and coating detection data, dew point monitoring records, data collected from formation and grading cabinets, and laboratory test results. Updates occur in real time for each batch and individual cell. Fields include batch number, parameter value, detection time, qualification determination, and more. Units include electrochemical-specific units such as voltage, capacity, and dew point temperature.
Constraints on knowledge base retrieval and recall
Real-time updated batch data requires retrieval chains to support high-frequency data synchronization, to avoid recalling outdated batch parameters. Electrochemical-specific detection fields such as dew point temperature, internal resistance, and capacity require matching precise semantic and numerical associations, rather than relying solely on keyword recall. Multi-source heterogeneous data including MES, laboratory, and cabinet data requires retrieval to support cross-data source associated recall, ensuring parameters cited in tender documents come from authoritative records of corresponding production links. Long-term formation and grading data requires complete archiving of historical batches, to avoid missing key consistency verification data during retrieval. Compliance requirements for tender scenarios mean recalled content must strictly match the current batch’s qualification determination standards, to prevent citation of unqualified parameters.
How to configure settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precise semantics of electrochemical parameters, to avoid recalling unrelated coating or liquid injection data |
limit | Top 8 entries | Covers necessary parameters across multiple links including batching, coating, and formation, controls context length to fit tender document length requirements |
searchMode | mixedRecall | Balances semantic matching for professional terminology and precise recall of keywords such as batch numbers and parameter units |
usingReRank | Enabled | Performs secondary ranking on recalled multi-source data, prioritizes returning high-priority documents such as formation and grading, safety test reports |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands professional terms in tender documents, such as associating "areal density" with coating and rolling parameter documents |
collectionIds | Knowledge base collection IDs grouped by business link | Precise limitation of retrieval scope, to avoid mixing in general electronic assembly documents unrelated to cell manufacturing |
Three common misconfigurations
- Only text content is returned in retrieval results, without detection report images. The cause is that the knowledge base’s image parsing plugin is not configured, and the prompt does not explicitly require embedding image reference identifiers.
- AI-generated knowledge base content is overly modified, and the units and field formats of original parameters are not retained. The cause is that the configuration item for directly outputting the original knowledge base text is not enabled, and the prompt does not specify retaining the original text.
- The number of retrieval results exceeds expectations, or unrelated consumer electronic assembly documents are mixed in. The cause is that
collectionIdsis not correctly configured to limit the retrieval scope, or thesimilaritythreshold is set too low, leading to recall of redundant data.
How to verify correct configuration
- Submit a test query containing coating and rolling parameters, check whether the retrieval results include documents from the corresponding link, and whether the parameter units match the original records.
- Upload a test report document containing images, initiate a test query, confirm that the returned results include image reference links or placeholder identifiers.
- Switch between knowledge base collections for different business links, verify that retrieval results only include content from the corresponding link, and do not mix in documents from other categories.
- After enabling
usingReRank, check whether the ranking of recall results prioritizes returning high-priority documents such as formation and grading, safety test reports.
The values provided here are common starting points and should be measured against the reader's own samples.