What Data for This Category Looks Like
The primary data users are the electrode, assembly, formation, and quality departments of power, energy storage, and consumer battery enterprises. Daily documents include batching records, coating and rolling parameters, winding or lamination records, liquid injection records, formation and grading data, OCV and safety test reports. Data comes from raw material batches, MES systems, slurry and coating testing equipment, dew point monitoring devices, formation and grading cabinets, and laboratory datasets. Update frequency is real-time, per batch and per individual cell. Data fields include material batch numbers, process parameter values, test values, timestamps, and additional relevant fields. Units cover standard manufacturing and electrochemical units such as weight measurements, millimeters, seconds, millivolts, and similar units.
How These Characteristics Create Constraints for Knowledge Base Retrieval and Recall
Real-time, high-frequency batch data requires the knowledge base to support incremental synchronization mechanisms. This avoids delays caused by full reindexing, ensuring process data for each cell batch can be retrieved promptly. Multi-source heterogeneous data sources include directly uploaded device parameters and laboratory reports. These require parsing rules adapted to different formats. Specifically, PDF safety test reports need accurate extraction of table and numerical fields. Electrochemical processes have strict requirements for parameter precision. Retrieval must balance semantic matching and exact numerical matching, to avoid retrieving process data from unrelated batches. Long-duration formation and grading data can produce large individual document sizes. It is necessary to limit the context length returned by retrieval, to avoid exceeding the model's supported range.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Similarity Threshold | 0.75–0.85 | Matches the precision requirements for electrochemical process parameters, filters low-relevance non-batch data |
| Single Search Limit | Top 10 results | Balances retrieved information volume and context length, adapts to associated queries for single-batch process data |
| Search Mode | Semantic + exact matching | Meets the needs of semantic understanding of process parameters and exact matching of batch numbers and numerical values |
| Reranking Switch | Enabled, rerank top 5 results | Performs secondary sorting on retrieved sets of process parameter results, improves the recall priority of core parameters |
| Query Optimization Switch | Enabled, expand 2–5 synonyms | Automatically expands synonyms related to processes, adapts to real-time updated batch data retrieval |
| Collection ID Binding | Bind corresponding collections by production link | Precise limitation of retrieval scope, avoids mixing manufacturing data from other categories |
This configuration is based on FastGPT v4.17.0.
Three Common Misconfigurations
- Issue: Newly uploaded batching records remain unavailable for retrieval within the specified time frame. Cause: Incremental synchronization trigger rules are not configured, and only full updates are performed, resulting in real-time data failing to sync promptly.
- Issue: When retrieving PDF-format OCV test reports, areal density and compaction parameters in tables cannot be recognized properly. Cause: Dedicated parsing configuration for table datasets is not enabled, leading to table content being incorrectly split or omitted.
- Issue: After switching the embedding model, recall results for multi-language process parameters decrease significantly. Cause: Batch re-embedding operations are not performed, and old data still uses the vector representation from the original model, which does not match the vector space of the new model.
How to Verify Proper Configuration
- Run a query for process parameters of a single battery cell batch, check whether the number of retrieval results matches the single search limit setting.
- Upload a new batching record, run the same query after waiting the specified time, confirm that the new data can be retrieved normally.
- Upload an OCV test report containing tables, check whether core parameters in the table are fully extracted and included in the retrieval scope.
- Switch the embedding model and perform batch re-embedding operations, check whether the vector update progress completes normally.
The configuration values provided are common starting points, and should be evaluated against your own specific samples and use cases.