What the Data for This Category Looks Like
Daily documents for this category include batching records, coating and calendering parameters, winding or lamination records, liquid injection records, formation and grading data, OCV and safety test reports. Data sources include raw material batch ledgers, production MES systems, slurry and coating inspection data, workshop dew point monitoring records, data collected from formation and grading cabinets, and laboratory test reports. Data is updated in real time for each production batch and each individual cell. Business scenarios focus on electrode preparation, assembly, formation, and quality control processes. The electrochemical system has extremely high requirements for production environment moisture, cleanliness, and batch consistency, which differs significantly from general electronic assembly.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
Real-time updated batch-level data requires the retrieval pipeline to support low-latency recall, to avoid process parameter matching errors caused by data lag. Multi-source and heterogeneous document types and fields, such as slurry solid content, workshop dew point values, and cell internal resistance parameters, require the retrieval model to adapt to matching logic for different data structures. Exclusive production records for each individual cell must be accurately associated with batch IDs to prevent cross-batch parameter confusion. Strict production consistency requirements require recall results to strictly match the units and precision of process parameters, to avoid process adjustment errors caused by unit deviations. The long-cycle nature of formation and grading data requires the retrieval pipeline to support cross-cycle associated queries of historical data.
How to Set the Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements of lithium-ion cell process parameters, prevents irrelevant data with low similarity from being included in retrieval results |
limit | Top 10 entries | Controls the number of documents recalled per single retrieval, adapts to the scale of documents associated with a single batch |
searchMode | embedding | Adapts to multi-source heterogeneous process documents, prioritizes semantic matching logic |
usingReRank | Enabled | Performs secondary sorting on recall results, optimizes relevance sorting for long-cycle formation and grading data |
datasetSearchUsingExtensionQuery | Enabled | Supplements process-related extended queries, covers parameter associations across multiple links such as batching and coating |
collectionIds | Configured according to the target process document collection | Accurately limits the retrieval scope, avoids interference from cross-category documents |
Three Common Errors
- When invoking a knowledge base search, the prompt "No matching knowledge base found" is displayed. This occurs because the
collectionIdsparameter is not configured correctly, and a non-existent knowledge base collection ID is provided. - The number of retrieval results does not match expectations. This occurs because the retrieval scope is not limited by production batch, resulting in recall of redundant cross-batch data.
- Knowledge base question answering splitting fails. This occurs because long parameter fields of lithium-ion cell documents are not adapted, such as multiple sets of coating and calendering parameters, and the complete process parameter association relationship is not retained during splitting.
How to Confirm the Configuration Is Correct
This page is written based on the retrieval framework of FastGPT v4.17.0.
- Import process documents for a single batch, perform a retrieval test, and check whether the recall results include all associated documents for that batch.
- Adjust the
similarityparameter range, and verify whether the relevance of the recall results meets the matching precision requirements of process parameters. - After enabling
usingReRank, compare the result order before and after secondary sorting, and confirm whether the relevance sorting of long-cycle data is reasonable. - Check the
collectionIdsconfiguration, confirm that only the document collection for the current process link is included, with no cross-category documents mixed in.
The values provided are common starting points and should be measured against the reader's own samples.