What Data for This Category Looks Like
The data for this category comes from raw material batch ledgers, workshop MES systems, slurry and coating testing equipment, dew point monitoring devices, formation and grading cabinets, and laboratory test reports. Daily documents include batching records, coating and rolling parameter tables, winding or lamination work orders, liquid injection ledgers, formation and grading data, OCV and safety test reports. Data is updated in real time per production batch and individual cell production node. Document fields include slurry solid content, coating areal density, rolling thickness, dew point temperature, charge-discharge current, and other metrics, with corresponding units such as g/㎡, μm, ℃, A, and others.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall?
The real-time nature and batch-binding characteristics of data for this category require that knowledge base retrieval accurately match production batches and corresponding process parameters, to prevent quality issues caused by misuse of cross-batch parameters. Multi-source and heterogeneous document formats require retrieval to support both structured tables and unstructured text, and filter irrelevant content by document type. Highly sensitive electrochemical production requirements demand that retrieval results strictly link to original data sources, to avoid safety risks such as internal short circuits and bulging caused by parameter deviations. Frequently updated single-cell data requires that recall results be limited to valid documents for the current production batch, to prevent calls to expired data.
Configuration Settings
This configuration scheme is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for lithium-ion cell process parameters, avoiding low-similarity irrelevant parameters from being included in recall results |
limit | Top 8 entries | Covers core parameter items for a single batch process, while controlling context length to fit model input limits |
searchMode | mixedRecall | Balances precise matching of structured parameters and semantic recall of unstructured documents, adapting to multi-source document formats |
usingReRank | Enabled | Performs secondary ranking on recall results, corrects semantic deviations from vector retrieval, and strengthens strongly relevant matching for process parameters |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonymous expressions for process questions, such as matching "coating areal density" to documents related to "coating thickness uniformity" |
collectionIds | Bound to the process knowledge base collection for the current production workshop | Limits the retrieval scope to documents from the workshop associated with the current batch, avoiding confusion from cross-workshop data |
Three Common Misconfigurations
- Retrieval results do not display traceability metadata such as the raw material batch and testing equipment number of the corresponding document. Cause: The metadata display rule for retrieval results is not configured, and core traceability fields of documents are not included in the recall return scope.
- Images in retrieval results fail to load or display properly. Cause: The knowledge base's image analysis and external link hosting configuration is not enabled, and direct reference to local image paths leads to inaccessibility.
- Recall results include expired formation and grading data. Cause: The recall collection is not filtered by production batch, resulting in calls to historical process documents from non-current batches.
How to Verify Correct Configuration
- A coating and rolling parameter document for the current batch is uploaded, a retrieval for "coating areal density parameters" is initiated, and returned results are checked to confirm they only include this document and similar documents from the same workshop.
- A retrieval containing professional terminology, such as "winding alignment abnormality handling", is initiated, and relevant lamination work documents are checked to confirm they are automatically matched through query expansion.
- The display items of retrieval results are reviewed, and traceability metadata such as document batch number and testing equipment number is confirmed to be included.
- A dew point monitoring screenshot from the coating process is uploaded, a retrieval for "dew point control requirements" is initiated, and the results are checked to confirm the image is displayed normally.
All provided configuration values are common starting points, and should be validated against operational samples from the target deployment environment.