What this category's data looks like
The typical subjects of this category are passenger car, 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 consistency and release documents. Data sources cover ERP/MES systems, VIN and vehicle configuration systems, supplier batch ledgers, tightening tools, EOL inspection lines, and quality traceability systems. Data update frequency follows per individual vehicle: some real-time collected tightening and inspection data is updated in real time, while batch documents are synchronized per shift. Documents use VIN as the core associated field, including standardized fields and corresponding units such as part numbers, torque parameters, inspection thresholds, and supplier batch numbers.
What constraints do these characteristics impose on the knowledge base retrieval and recall link?
The VIN-centric association logic requires retrieval to precisely match individual vehicle configurations to avoid cross-vehicle configuration confusion. Real-time updated tightening and inspection data requires the retrieval link to support low-latency recall, otherwise it cannot match the latest verification standards of the current workstation. Multi-source batch documents need to support dataset filtering by business scenario to avoid irrelevant data interfering with retrieval results. Mixed-line production beat requires that the number and length of single retrieval results adapt to workstation operation time; overly long retrieval results will occupy workstation operation windows. At the same time, quality traceability documents need to retain the original association relationship, and the content retrieved and recalled must be traceable back to the original data source and update time.
How to set the configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Adapts to the accuracy requirements of VIN matching, avoiding low-match irrelevant configuration data from being included in results |
limit | Top 10–15 entries | Adapts to the workstation operation time window, returns necessary documents sufficient to cover the current assembly process, avoiding excessive content occupying operation time |
searchMode | mixedRecall | Balances precise matching of structured fields such as VIN and semantic matching of unstructured content such as work instructions |
usingReRank | Enabled | Re-ranks the initial recalled results, prioritizing documents strongly related to the current workstation process, such as tightening torque requirements |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval for relevant supplier batches, part lists and other associated data based on the user-input VIN, covering full-process traceability requirements |
collectionIds | Configured by process scenario grouping | Filters the dataset required for the current workstation, for example, the final assembly chassis workstation only calls tightening tool data and chassis assembly work instructions |
Three common misconfigurations
- Phenomenon: Irrelevant supplier batch documents appear in retrieval results, and the current complete vehicle's supplier information cannot be accurately matched. Cause: The
collectionIdsconfiguration is not used to limit the dataset for the corresponding process, resulting in redundant cross-scenario data being recalled. - Phenomenon: The download address is abnormal after clicking to view the original content, and the original document cannot be obtained. Cause: The storage path of the knowledge base files is not configured correctly, or the access permissions of the corresponding API interface are not enabled.
- Phenomenon: The globally configured VIN information cannot be automatically brought in during retrieval, resulting in the inability to match the exclusive documents of the corresponding vehicle. Cause: The automatic assignment logic of global variables is not configured correctly, or the global variable parameters are not included in the retrieval expansion rules.
How to verify a successful configuration
- Enter a single vehicle VIN for retrieval, and check whether the returned results include the part list and assembly work instructions corresponding to that VIN.
- Simulate real-time pushing of a tightening data document, confirm that the document can be quickly recalled during retrieval, and verify that the real-time updated retrieval link is functioning properly.
- Switch between different process scenarios, and check that the dataset configured via
collectionIdsonly returns relevant documents for the corresponding workstation, with no redundant content. - After enabling
usingReRank, confirm that the returned results are sorted to prioritize matching the core requirements of the current process, such as torque parameter documents appearing at the top.
The values provided here are common starting points and should be measured against the reader's own samples.