This document is compatible with FastGPT v4.17.0.
What this category’s data looks like
Daily documents for this category include injection molding process cards, raw material drying records, mold history logs, first-piece inspection reports, patrol inspection forms, and material change and machine cleaning records. Data sources include resin batch property sheets, injection molding machine parameters, mold temperature controllers, MES systems, and dimensional and appearance inspection data. Documents are updated per production batch, mold change or material change event, and each shift. Fields include process parameters such as fill, hold, and cooling window durations, mold temperature, raw material moisture content, mold usage count, and other fields with clear engineering units.
What constraints these characteristics impose on knowledge base retrieval and recall
Since data is updated in real time per production batch, mold change, or material change event, knowledge base retrieval must support filtering recall scope by batch and mold ID. This prevents interference from invalid cross-batch data. Documents include process parameters with engineering units. Retrieval must match parameter units to query conditions to avoid parameter confusion. Frequently updated patrol and material change records require the retrieval pipeline to support incremental synchronization. This prevents expired data from being recalled. Additionally, injection molding fault analysis requires rapid recall of associated mold and raw material data. This imposes clear requirements on retrieval latency.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for injection molding process parameters, prevents low-relevance fault analysis data from being included |
limit | Top 6 entries | Controls the number of documents recalled per round, adapts to scenarios where single documents such as injection molding process cards and mold history logs have relatively long content |
searchMode | embedding | Prioritizes semantic matching of process parameters and equipment parameters, supports cross-document fault association retrieval |
usingReRank | Enabled | Rearranges recall results by relevance, optimizes retrieval ranking for injection molding faults such as shrinkage and short shots |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms, supplements synonyms related to injection molding processes, such as associating "shrinkage" with "shrink mark" and "dimensional deviation" |
collectionIds | Dataset IDs bound to mold/batch | Limits retrieval scope to the knowledge base associated with the current production batch, avoids interference from cross-batch data |
Three common misconfigurations
- Phenomenon: The knowledge base retrieval returns matching results, but there is no valid output when passed to the large model. Cause:
datasetSearchUsingExtensionQueryis not configured, causing recalled documents to not meet the large model context requirements, or thelimitvalue is set too high, causing context overflow. - Phenomenon: When creating a knowledge base directory, a parent ID is passed in, but the directory still appears under the root directory. Cause: The dataset ID of the corresponding parent directory is not bound in the
collectionIdsconfiguration, or the parent ID format does not meet FastGPT requirements. - Phenomenon: Retrieval latency for a single-document knowledge base exceeds 10 seconds. Cause:
usingReRankis not enabled, resulting in too many recalled results, or thesimilaritythreshold is set too low, resulting in a large number of irrelevant documents being recalled.
How to verify that the configuration is correct
- Retrieve specific injection molding process parameters, check whether the recalled results only include documents from the current production batch, to verify that the
collectionIdsconfiguration takes effect. - Input the query "shrinkage", check whether the recalled results include associated raw material drying records and mold temperature data, to verify that
datasetSearchUsingExtensionQuerytakes effect. - Check whether the number of returned retrieval results matches the
limitconfiguration, to verify that the parameter settings are accurate. - Test retrieval across mold IDs, confirm that only knowledge base data for the corresponding mold is recalled, to verify that the filtering logic is correct.
The values provided here are common starting points, and should be evaluated against the user’s own samples.