What the Data for This Category Looks Like
Data sources for this category include raw material quality certificates, heating furnace systems, press collection data, MES systems, non-destructive testing instruments, and customer forging drawings. Daily documents cover forging process specifications, cutting cards, heating furnace records, forging process cards, mold life records, heat treatment work orders, and ultrasonic testing reports. Data updates occur per individual piece, batch, or furnace heat. Fields include furnace batch number, forging batch number, heating temperature, forming pressure, defect type, and mold usage count. Units include degrees Celsius, megapascals, and counts, among others.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
This category’s data is updated frequently by production batch or furnace heat. Cross-document matching of associated fields such as furnace batch number and forging batch number is required to ensure the accuracy of traceability results. Daily documents contain multiple types of content: process parameters, testing records, production logs, and more. Field formats vary, and some documents use non-standard terminology and units. Retrieval and recall must accurately match batch-related fields, adapt to the field characteristics of multiple document types. It must also address the risk of index synchronization delay caused by frequent updates, to prevent retrieval results from being out of sync with current production data. Additionally, retrieval weight allocation across different document types must be balanced, to avoid unbalanced priority between process documents and testing documents.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precise fields such as batch numbers and defect types for this category, balances recall coverage and result relevance, and adapts to the precise traceability needs of forging scenarios |
limit | Top 8–12 results | Adapts to the characteristic that a single batch has a large number of associated documents, avoiding context overload that reduces analysis efficiency |
searchMode | mixedRecall | Meets retrieval needs for multiple document types including process parameters, testing records, and production logs |
usingReRank | Enabled | Optimizes sorting results for high-weight fields such as defect types and process parameters, improving the targeting of defect analysis |
datasetSearchUsingExtensionQuery | Enabled | Supplementary retrieval for associated fields such as batch numbers and furnace batch numbers, improving cross-document matching accuracy |
collectionIds | Bind the forging document collection for this category | Limits retrieval scope, excluding irrelevant knowledge base content from other categories |
Three Common Mistakes
- Phenomenon: Multiple duplicate index entries for the same document block appear in retrieval results. Multiple recall records for the same document block can be seen in logs. Cause: No deduplication rule is configured for multi-index associated document blocks, leading to repeated recall of the same content.
- Phenomenon: Non-batch forging documents are mixed into retrieval results. The interface returns production records from other furnace heats. Cause:
collectionIdsis not bound to limit the retrieval scope, ordatasetSearchUsingExtensionQueryis not enabled to supplement matching of batch-related fields. - Phenomenon: The number of retrieval results is much lower than expected. Only 2-3 results are returned per single retrieval. Cause: The
similaritythreshold is set too high, filtering out some low-match documents that meet the required relevance, such as process records containing non-standard terminology.
How to Confirm Configuration Is Complete
- Check the knowledge base collection binding configuration, confirm that
collectionIdsspecifies the forging production document collection for this category, and no irrelevant collections are included. - Test retrieval with forging batch number keywords, verify whether
datasetSearchUsingExtensionQuerysupplements matching results for associated fields such as furnace batch numbers and forging numbers, and that the recall scope covers cross-document associated content. - Adjust the
similaritythreshold, observe changes in the relevance of recall results, and confirm that the value meets the field matching needs of this category, avoiding thresholds that are too high or too low. - After enabling
usingReRank, check whether the sorted results prioritize high-weight fields such as defect types and process parameters, which aligns with the business logic of quality traceability.
The values provided are common starting points and should be measured against the reader's own samples.