What Data for This Category Looks Like
Data for this category primarily comes from raw material quality certificates, heating furnace systems, press collection data, MES systems, non-destructive testing equipment, and customer forging drawings. Daily generated documents include forging process specifications, cutting cards, heating furnace records, forging process cards, mold life records, heat treatment entrustment forms, and ultrasonic testing reports. Data updates follow the rhythm of individual forgings, batches, or furnace heats. Fields include furnace batch number, forging ratio, fiber direction, testing defect type, and others. Some data is structured from industrial systems, while some is unstructured content in paper or electronic formats.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Data sources are scattered and formats are inconsistent. Structured industrial system data is mixed with unstructured documents. This requires retrieval systems to support multi-source data adaptation. Update frequency is high, and updates are tied to batches. The system must support quick location of associated documents by furnace batch number or forging batch. A single batch of data is linked to multiple document types. Retrieval must match the relational connections between forging processes, testing results, and production records to avoid retrieving irrelevant content. Some real-time data comes from industrial systems. Retrieval response speed must align with production rhythms. The system must also accurately identify professional fields such as forging ratio and fiber direction to ensure retrieved content matches business needs.
Configuration Settings
The following configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches forging professional terminology and batch fields to avoid retrieving irrelevant general process documents |
limit | Top 6 entries | Covers multiple types of documents associated with a single batch of forgings, while controlling the total context length |
searchMode | mixedRecall | Balances structured furnace batch number and batch number data with unstructured process specifications and testing report content |
usingReRank | Enabled | Reorders retrieval results by business relevance, prioritizing matching core fields such as forging ratio and fiber direction |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms related to batches and furnace numbers based on the original query to improve retrieval accuracy |
Three Common Misconfigurations
- Symptom: Retrieval results are empty or only return a small number of irrelevant documents. Cause: The
similaritythreshold is not configured for forging professional fields, ordatasetSearchUsingExtensionQueryis not enabled to expand batch retrieval terms. This prevents matching core identifiers such as furnace batch numbers and forging drawings. - Symptom: In private deployment scenarios, API calls with the
limitparameter set above 6 result in no response or return errors. Cause: The system's default single retrieval upper limit configuration is not adjusted. The default limit matches the number of documents per batch, and exceeding this threshold triggers parameter verification failure. - Symptom: Importing heat treatment entrustment forms or ultrasonic testing reports larger than 1 MB fails. Cause: The system's single file size threshold is not adjusted. Forging documents often contain large amounts of testing data, and the default threshold cannot accommodate this.
How to Verify Correct Configuration
- Upload the cutting card, heating furnace record, and ultrasonic testing report for a single batch of forgings. Initiate a retrieval that includes the furnace batch number, and confirm all uploaded documents are included in the retrieval results.
- Call the API with
limitset to 8, verify the number of returned results matches expectations, and no parameter errors occur. - Import a single 2.5 MB ultrasonic testing report, confirm the import is successful and normal retrieval is possible.
- Initiate a retrieval that includes "fiber direction", and confirm the retrieval results prioritize returning process specification documents matching this field.
The values provided are common starting points and should be measured against the reader's own samples.