What the data for this category looks like
Forging and die forging production ledgers cover multiple document types, including forging process specifications, cutting lists, furnace logs, forging process cards, mold life records, heat treatment work orders, and ultrasonic testing reports. Data sources include raw material quality certificates, furnace systems, press-collected data, MES systems, non-destructive testing instruments, and customer forging drawings. Data is updated once per part, batch, or furnace run. It includes fields such as furnace batch number, forging ratio, fiber direction, heating temperature, and mold life count. Units include ℃, MPa, and counts. The data structure includes both structured tables and unstructured text.
Constraints on knowledge base retrieval and recall
This category has high-update-frequency, multi-source heterogeneous data. Retrieval systems must support incremental indexing and adapt to multiple document types. Structured ledgers with multiple fields (such as Excel-format cutting lists) require precise field matching. This avoids losing key fields when using plain text chunking. Data correlation requirements are strict. Retrieval for a single batch of forgings must cover multiple documents such as furnace logs and testing reports, while excluding interference data from other batches. This sets clear requirements for retrieval scope restriction and recall accuracy. Additionally, business error consequences are severe. Retrieval result accuracy must be ensured to avoid missing key parameters.
How to set configurations
All configurations below are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for structured fields in forging ledgers, filters low-relevance non-target documents |
limit | Top 8 entries | Covers the volume of associated ledger data for a single furnace run or batch of forgings, avoids context overflow |
searchMode | fullTextRecall | Use full-text retrieval for text content. Field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Improves ranking accuracy for key business fields such as forging ratio and furnace batch number |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms to cover terminology differences across documents such as cutting lists and furnace logs |
collectionIds | Bind corresponding collections by forging category | Accurately recalls exclusive ledger data for the current batch of forgings, avoids interference from cross-category data |
Three common misconfigurations
- Uploaded Excel-format ledgers cannot extract key fields such as furnace batch number and forging ratio during retrieval, resulting in low matching between retrieval results and queries. The cause is failure to configure indexing rules for structured table data, only processing the data as plain text chunking, which prevents structured fields from being effectively identified.
- Retrieval results only return a single matching entry, failing to cover associated documents such as furnace logs and ultrasonic testing reports for the same furnace run. The cause is that the
limitparameter is set too small, not reserving enough recalled entries to cover all associated data. - Retrieval results include ledger data from other batches of forgings, interfering with report generation for the current batch. The cause is failure to configure the
collectionIdsparameter, not restricting the retrieval target collection, leading to recall of irrelevant cross-category data.
How to confirm configurations are correctly set
- Upload a typical forging cutting list Excel spreadsheet, test retrieval for "furnace batch number of this batch of forgings", and check if the specified fields in the spreadsheet can be accurately extracted.
- After configuring the
collectionIdsparameter to bind the collection for the current forging category, retrieve keywords for cross-category forging ledgers, and confirm results only return documents within the currently bound collection. - After enabling the
usingReRankfunction, compare retrieval ranking results before and after enabling, and confirm that the ranking priority of key business fields such as forging ratio and furnace batch number has been improved. - Test retrieval for multiple associated documents, and confirm that the number of entries recalled by the
limitparameter covers all ledger data related to the current batch.
The values provided here are common starting points, and should be evaluated against the reader's own samples.