What the Data for This Category Looks Like
Typical entities in this segmented category are passenger vehicle, commercial vehicle, and new energy vehicle complete vehicle manufacturers. Daily generated documents include production orders, VIN lists, assembly work instructions, tightening data, offline inspection reports, rework records, and conformity and release documents. Data sources cover ERP systems, MES systems, VIN and configuration systems, supplier batch libraries, tightening tools, EOL inspection lines, and quality traceability systems. Data update frequencies are per vehicle, real-time, and per shift. Core fields include VIN code, torque value, configuration items, and batch number. Each vehicle's VIN links multiple types of business data. In mixed-line production scenarios, the same knowledge base stores configuration information for multiple vehicle models.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
The mixed-line production model centered on the VIN as the single vehicle mainline requires retrieval to accurately match the VIN field, to avoid recalling configuration data for non-target vehicle models. Real-time and per-shift updated tightening data and rework records require retrieval nodes to quickly obtain the latest business data, ensuring that parameters used in tender documents comply with current production standards. The coexistence of multiple types of documents requires retrieval to cover both structured fields and unstructured text, balancing accuracy and recall completeness. The high risk of error consequences requires recalled results to have sufficient confidence, to prevent mismatched or missing content from entering the tender process, avoiding risks such as subsequent production line shutdowns or vehicle recalls.
How to Configure the Settings
| Config Item | Recommended Approach | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Configuration parameters for vehicle final assembly have high precision requirements, to avoid irrelevant data with low matching scores being included in tender documents |
limit | Top 10–15 entries | Tender documents need to cover multiple types of data including vehicle configuration, tightening standards, traceability requirements, and excessive content will exceed the context length limit |
searchMode | mixedRecall | Need to match both structured fields such as VIN codes and unstructured documents such as assembly work instructions, balancing accuracy and recall completeness |
usingReRank | Enabled | Configuration similarity is high in mixed-line production scenarios, reranking can filter irrelevant results from non-target vehicle models and improve recall accuracy |
datasetSearchUsingExtensionQuery | Enabled | Terminology in tender documents may differ from the original fields in the knowledge base, extended queries can match more relevant business content |
collectionIds | Bind the knowledge base collection corresponding to the final assembly workshop | Isolate configuration data from different production lines, avoid retrieving documents from irrelevant vehicle models, and narrow the retrieval scope |
Three Common Misconfigurations
- Phenomenon: Retrieval results are empty, and the text content extracted by the knowledge base reference in the workflow is empty. Cause: The
collectionIdscorresponding to the final assembly workshop are not bound, and the retrieval scope covers knowledge base collections of irrelevant vehicle models, so target data cannot be matched. - Phenomenon: Non-target vehicle model configuration data is mixed in retrieval results. Cause: The
similaritythreshold is set too low, orusingReRankis not enabled, so low-matching similar content cannot be filtered. - Phenomenon: Similar queries are incorrectly recalled, such as similar business problems from different systems being incorrectly matched. Cause: Accurate retrieval is not set for exclusive fields such as VIN codes, and
datasetSearchUsingExtensionQueryis not enabled to correct terminology differences, leading to incorrect recall of cross-scenario similar queries.
How to Verify Correct Configuration
- Enter the knowledge base management interface of FastGPT v4.17.0, confirm that the knowledge base collection corresponding to the final assembly workshop has been bound, and check that the collection includes required documents such as production orders, VIN lists, and tightening data.
- Test retrieval of the configuration parameters corresponding to a specific VIN code, verify whether the matching degree of the returned results meets the
similaritythreshold requirements, and confirm that irrelevant results are filtered afterusingReRankis enabled. - Trigger the knowledge base retrieval node in the workflow, check whether the returned text content completely covers the configuration, standards and other information required for the tender document, and confirm that no null values are returned.
- Adjust the
similaritythreshold, verify changes in the number of results and matching accuracy, and confirm that the configuration parameters take effect.
The values provided here are common starting points and should be measured against the reader's own samples.