What the Data for This Category Looks Like
Daily documents for this category include mold design task sheets, DFM reports, part drawings, machining process cards, trial mold reports, mold modification notices, and mold history logs. Data sources include customer product models, mold flow or forming simulation files, CAD/CAM source files, machining and measurement data, and test sample inspection records. Updates are completed per individual project, with synchronized updates after each trial mold or mold modification. Most documents include dedicated fields such as mold number, material grade, machining tolerance, and trial mold batch. Some drawing documents include vector graphics and annotation parameters.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Process
The multi-type, high-update-frequency, and dedicated-field characteristics of this category impose multiple constraints on the retrieval and recall process. Mixed-format documents, including vector drawings and simulation files, require the retrieval pipeline to support cross-format parsing and recall. The project-based update rhythm requires isolating knowledge base data from different projects via tags to avoid cross-project retrieval interference. Dedicated fields such as mold number and machining tolerance require retrieval to support field-level filtering for precise matching of business needs. A high proportion of long documents requires adaptation to long-text chunking rules to prevent key process parameters from being truncated.
How to Set the Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | The matching accuracy of mold design process parameters directly affects trial mold results. This similarity range filters interference from general manufacturing documents and retains results matching the current project’s mold specifications and materials |
limit | Top 8 entries | The number of core documents for a single project’s knowledge base is limited. 8 entries cover all materials required for key cross-review, including trial mold reports, process cards, and drawings |
searchMode | fullTextRecall | Cross-review for mold design requires associating multi-dimensional information. Full-text retrieval covers scattered content such as drawing annotations, process steps, and trial mold defect records |
usingReRank | Enabled | Semantic dispersion may occur after long documents are chunked. Reranking corrects semantic deviations from initial retrieval and improves matching accuracy for mold modification schemes corresponding to trial mold defects |
datasetSearchUsingExtensionQuery | Enabled | Mold terminology is mostly industry-specific. Enabling query optimization expands user review questions into search terms that include professional terminology |
collectionIds | Specified via current project tags | Avoid cross-project data interference. Only recall knowledge base collections corresponding to the current mold project to ensure accurate data sources for cross-review |
Three Common Configuration Errors
- Phenomenon: Embedded images in the knowledge base, such as trial sample photos and CAD annotation drawings, fail to load normally after retrieval. Cause: Extended parsing configuration for multi-format documents is not enabled, or the uploaded image format is not supported by FastGPT v4.17.0, resulting in lost image association information during parsing.
- Phenomenon: Retrieval results include cross-project mold documents, leading to confusion between different batches of mold modification records during cross-review. Cause: The current project’s knowledge base collection is not specified via
collectionIds, so the retrieval scope covers the full knowledge base. - Phenomenon: Key parameters such as machining tolerance in long documents are not recalled, leading to missed key information during cross-review. Cause: The chunking logic does not adapt to long drawing documents, truncating complete annotated parameter segments into incomplete fragments, resulting in semantic loss.
How to Verify the Configuration Is Set Correctly
- Upload a trial mold report labeled with a dedicated mold number, search for the mold number, and verify that returned results only include documents from the current project.
- Upload a document containing vector drawings, search for professional terms such as "demolding angle", and verify that the results include parameter information from the drawing annotations.
- After enabling
usingReRank, search for "shrinkage defect mold modification scheme", and verify that reranked results prioritize matching corresponding records from trial mold reports. - Check the tag configuration of the knowledge base collection, confirm that only documents from the current project are included in the retrieval scope, and no cross-project data is mixed in.
The values provided are common starting points and should be evaluated against sample datasets relevant to the specific deployment context.