What This Category’s Data Looks Like
Daily documents for this category include configured BOM, assembly work instructions, refrigerant charging records, electrical safety test reports, performance spot check reports, and rework records. Data is sourced from ERP/MES systems, component batch ledgers, tightening and charging equipment, safety compliance testers, performance laboratories, and after-sales fault databases. Updates run per individual product, shift production batch, and spot check plan schedules. Document fields include component batch numbers, process parameter values, test result codes, and similar fields. Corresponding units include torque units, volume units, voltage units, and other standard measurement units.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
Multi-source heterogeneous data structures create index alignment requirements. Field formats for equipment logs, test reports, and after-sales records must be unified to ensure recall matching accuracy. The high-frequency update rhythm requires an incremental recall mechanism, to avoid excessive system resource usage from full index refreshes. Mixed-line assembly production scenarios require retrieval results to be strictly bound to the corresponding product model, preventing incorrect recall of cross-model process documents. Consumer-facing product requirements demand high confidence in safety compliance and performance recall data, to avoid compliance risks from retrieval deviations.
Configuration Settings
This configuration is adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches home appliance process terminology and parameter precision, ensuring relevance of recall results |
limit | Top 6 entries | Controls the number of documents returned per retrieval round, adapting to the multi-dimensional reference needs of process documents and exception reports |
searchMode | mixedRecall | Combines keyword and vector retrieval, adapting to mixed matching scenarios of model parameters and process requirements in mixed-line production |
usingReRank | Enabled | Performs secondary sorting on recall results, filtering model matching deviations in mixed-line scenarios |
datasetSearchUsingExtensionQuery | Enabled | Expands home appliance-specific professional terms in the original query, improving recall coverage of process-related documents |
collectionIds | Configured by product model group | Complete home appliance manufacturing has mixed-line production scenarios for multiple models. Grouping by model prevents incorrect recall of cross-model documents |
Three Common Misconfigurations
- Symptom: An error occurs in the question-and-answer workflow after adding a knowledge base, but independent search tests on the knowledge base return normal results. Cause: The
collectionIdsparameter is not configured correctly, and the document collection corresponding to the target product model is not bound, leading to cross-model retrieval conflicts. - Symptom: Retrieved documents do not include process parameters for the target home appliance model. Cause:
datasetSearchUsingExtensionQueryis not enabled, so home appliance professional terms cannot be expanded, resulting in failure to recall some process documents. - Symptom: A timeout error occurs during a single round of knowledge base retrieval. Cause: An incremental update index mechanism is not used, and full retrieval of multi-model massive process documents exceeds the interface response threshold.
How to Verify Successful Configuration
- Access the knowledge base retrieval test interface, enter a process query term for the corresponding model, and check whether the document type and product model of the returned results match.
- Adjust the
similarityparameter range, and verify whether the relevance of recall results meets the precision requirements for process drafting. - After enabling
usingReRank, check whether the secondary sorted results prioritize documents for the current production model. - Review interface logs to confirm that retrieval requests carry the correct
collectionIdsparameter, and that no cross-model retrieval occurs.
The values provided are common starting points, and should be evaluated against the reader's own samples.