What this category of data looks like
Data for this category originates from raw material quality certificates, heating furnace system data, press collection data, MES, non-destructive testing instruments, and customer forging part drawings. Daily documents include forging process specifications, cutting cards, heating furnace records, forging process cards, mold life records, heat treatment work orders, and ultrasonic testing reports. Data updates follow the rhythm of individual pieces, batches, or furnace charges. A single document typically includes fields such as furnace batch number, forging part batch number, process parameters, and testing results. Some equipment collection data uses time-series format, with no uniform fixed single document length.
Constraints imposed on knowledge base retrieval and recall
This category has high data update frequency, with new data generated for each batch or furnace charge. Retrieval must support high-frequency incremental synchronization to avoid recalling outdated data. Data exists in multi-source heterogeneous formats: structured time-series collection data, standardized testing reports, and unstructured process documents. Retrieval must adapt to content matching across mixed formats. Data includes strongly associated fields such as furnace batch number and forging part batch number. Inspection and review scenarios require precise matching of batch-related information, and cannot rely solely on semantic similarity to recall irrelevant content. Some equipment collection data has strong time-series characteristics, so retrieval must support filtering recall results by time range to meet process record traceability needs.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category requires precise matching of strongly associated fields such as furnace batch number and forging part batch number to avoid irrelevant recalls with low similarity |
limit | Top 8–12 entries | Inspection and review need to cover all process records for a single batch. Too many results increase review burden, while too few may miss key information |
searchMode | mixedRecall | This category of data includes structured time-series data and unstructured process documents. Hybrid retrieval balances semantic matching and precise field matching |
usingReRank | Enabled | Semantic similarity varies widely across multi-source heterogeneous data. Re-ranking optimizes the accuracy of result sorting |
datasetSearchUsingExtensionQuery | Enabled | Forging processes and testing use specialized terminology. Extended queries can cover recall of synonymous expressions |
collectionIds | Bind the knowledge base collection corresponding to the current forging part batch | Limit the retrieval scope to avoid recalling irrelevant documents from other batches |
This configuration applies to FastGPT v4.17.0.
Three Common Errors
- Phenomenon: Retrieval responses time out, with a duration exceeding 15 seconds. Cause:
collectionIdsis not configured to limit the retrieval scope. Default full retrieval of the entire knowledge base loads redundant unrelated batch data. - Phenomenon: Excel import fails or field parsing errors occur. Cause: Fields such as furnace batch number and forging part batch number in Excel are not configured as precisely matched retrieval fields, leading to loss of key associated information during chunking.
- Phenomenon: Vectorization takes more than 300 seconds after uploading a PDF document. Cause: The retrieval and recall scope is not adjusted for long-text forging process specifications, leading to loading of redundant content.
How to Verify Successful Configuration
- Enter the specified forging part batch number. Retrieval results only include documents such as heating furnace records and ultrasonic testing reports corresponding to that batch.
- After enabling
usingReRank, the first entry in the retrieval results has the highest matching degree with the query’s process parameters or batch information. - Upload a single forging process specification document, and confirm that incremental synchronization time meets expectations.
- Access the application release interface, and verify that the download link for retrieval results displays normally.
The values provided are common starting points and should be measured against the reader's own samples.