Data for This Category
Typical entities in this category include power battery pack manufacturers, energy storage cabinet manufacturers, and BMS system manufacturers. Daily documents cover cell grouping rules, BOMs, welding and torque records, BMS flashing records, insulation and voltage withstand reports, EOL tests, and system configuration lists. Data sources include cell capacity grading data, MES, welding equipment, torque tools, BMS software versions, and EOL test benches. Data updates per pack or cabinet. Each software version change also triggers an update. Document fields include professional parameters such as cell capacity, internal resistance, material codes, torque values, and BMS version numbers. Formats include structured tables and unstructured logs.
Constraints for Knowledge Base Retrieval and Recall
Since data updates per individual pack or cabinet, retrieval must strictly limit scope to avoid cross-batch data interference. Professional fields such as torque values and insulation withstand parameters require higher matching precision, and need to adapt to semantic associations of professional terminology. Mixed structured and unstructured data from multiple sources requires retrieval modes that balance precise matching and format compatibility. A single retrieval must cover core quality inspection documents for a single batch to avoid context overflow affecting subsequent audit workflows.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.85–0.90 | Match the terminology precision of professional quality inspection documents, filter low-relevance generic content |
limit | Top 8 entries | Cover the number of core quality inspection documents for a single batch of PACKs, avoid context overflow |
searchMode | embedding | Adapt to semantic matching requirements for professional fields such as cell parameters and torque values |
usingReRank | Enabled | Perform secondary sorting on recall results to filter low-confidence mismatched records |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand professional terminology based on quality inspection questions, for example, expand "grouping" to "cell capacity grading grouping rules" |
collectionIds | Batch-bound dataset IDs | Limit retrieval scope to only exclusive documents for the current supplier or batch, avoid cross-batch data interference |
Three Common Configuration Mistakes
- Phenomenon: Retrieval results include welding torque records from non-current batch PACKs, and the number of returned entries exceeds expectations. Cause: Failed to bind the exclusive dataset for the current supplier or batch via
collectionIds, so the retrieval scope covers the full knowledge base. - Phenomenon: The knowledge base collection tag entry cannot be found in the interface. Cause: Sub-scenario permission configuration was not enabled in the dataset management page of FastGPT v4.17.0, or the knowledge base settings module of the corresponding workflow was not accessed.
- Phenomenon: The knowledge base associated ID returned during the chat process does not match the current quality inspection task. Cause: The
collectionIdsparameter was not correctly bound in the workflow, and the associated dataset ID does not correspond to the documents of the current batch.
How to Confirm Correct Configuration
- Access the dataset management page of FastGPT v4.17.0, verify that the dataset bound to the
collectionIdsparameter includes core documents such as the grouping rules and BOMs for the current batch. - Initiate a simulated quality inspection query, for example, "Check whether the insulation withstand of the current batch of energy storage cabinets meets requirements", verify that the
similarityscores of the retrieval results fall within the preset range of 0.85–0.90. - After enabling the
usingReRankfunction, verify that the secondary sorted results prioritize displaying high-confidence EOL test records and BMS flashing records. - Access the workflow editor, confirm that the RAG retrieval node has bound the correct
collectionIdsparameter and retrieval configuration, and no context overflow or timeout errors occur.
Note: The configuration values listed above are common starting points. Adjust them based on your own samples and testing.