What the Data for This Category Looks Like
Data sources for process and standard data in mass production of automotive parts include customer drawings, special characteristic lists, ERP/MES systems, inspection and measuring tool records, equipment process parameters, and supplier quality certificates. Data is updated per production batch, during daily shift patrols, and through continuous collection of equipment parameters. Daily documents cover control plans, process flowcharts, work instructions, first article and shift patrol records, SPC reports, PPAP materials, and 8D reports. Fields include special characteristic numbers, process parameters, deviation ranges, equipment numbers, and other relevant fields. Units include millimeters, newton meters, batch numbers, and other process-specific units.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall?
Data sources are scattered, with both structured parameters and unstructured text. This requires retrieval to cover both keyword matching and vector features. Frequently updated data needs incremental synchronization to prevent standard documents from becoming outdated. Recalls must meet high confidence levels due to strict constraints from vehicle customers, requiring precise matching of special characteristics. Mass production consistency demands that retrieval results cover all process standards. Missing process parameters from core processes may trigger batch quality issues.
Configuration Settings
These settings are compatible with FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Automotive part special characteristics require high confidence; this range filters low-match irrelevant standards and avoids incorrect recall |
limit | Top 8 entries | Mass production processes cover multiple processes; recalling 8 entries at a time covers the standard content for core processes while controlling context length |
searchMode | mixedRecall | Balances keyword matching for text standards and vector features for parameter values, adapting to mixed-format process documents |
usingReRank | Enabled | Re-ranks initial recall results by relevance to ensure high-value results related to special characteristics are displayed first |
datasetSearchUsingExtensionQuery | Enabled | Extended queries can cover associated context such as equipment and batches in process questions, improving recall coverage |
collectionIds | Configured by process group | Split knowledge base collections by production process to precisely limit retrieval scope and prevent irrelevant cross-process standards from being recalled |
Three Common Misconfigurations
- Phenomenon: Images cannot be displayed normally in search results after importing Feishu documents. Reason: The knowledge base image parsing configuration was not enabled, and only text content was extracted, meaning process parameters associated with images cannot be retrieved.
- Phenomenon: An error occurs when creating a new knowledge base after privately deploying FastGPT, with a prompt indicating missing configuration fields. Reason: The vector database connectivity check was not completed, or the data source association parameters for the dataset were not configured correctly.
- Phenomenon: Search results cannot exclude specified knowledge base collection content. Reason: The
collectionIdsparameter was not configured correctly, or an incorrect collection ID list was passed, leading to failure to precisely limit the retrieval scope.
How to Verify Correct Configuration
- Upload a work instruction document containing special characteristic parameters, perform targeted retrieval, and confirm that the recall results include the core content of the target document.
- After configuring
collectionIds, retrieve documents from non-specified collections, and confirm that these results are not included, verifying that the collection filtering function works. - After enabling
usingReRank, compare the recall ranking before and after enabling to confirm that high-match results are displayed first. - After importing Feishu documents, check the attachment parsing status in the knowledge base to confirm that both images and text have been successfully extracted.
The values provided are common starting points and should be measured against samples from the specific deployment.