What Data Looks Like for This Category
Data for this category comes from daily business documents of foundries, die casting workshops, and quality departments. These documents include casting process cards, alloy batching lists, heat records, mold histories, first article inspection records, material reports, and nonconforming product review forms. Data sources cover ERP material master data, melting furnace thermometers, spectrometers, MES heat records, mold management systems, and customer technical drawings. Data updates occur per heat, per batch, or daily. Each single data entry includes fields such as alloy composition, pouring temperature, mold times, and defect type. Units include temperature, weight, percentage, and other professional foundry parameters.
Constraints Imposed on Knowledge Base Retrieval and Recall
The data characteristics of this category create multiple constraints for retrieval and recall workflows. Multi-source heterogeneous data covers equipment collection, system records, and business documents. This requires adaptation to indexing rules for different formats. The high update frequency requires the retrieval pipeline to support incremental synchronization, avoiding excessive resource usage from full reindexing. Single data entries contain professional foundry fields with close associations. For example, a heat entry corresponds to batching ratio, pouring temperature, and defect type. Retrieval must support multi-field associated recall. Units vary across different documents. Preprocessing must standardize formats to prevent matching errors. Process quality audits require tracing associations between defects, heats, and mold times. Retrieval must accurately recall historically associated records.
How to Configure Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Casting data has high professional specificity, so low-match irrelevant documents must be filtered. This range balances recall coverage and accuracy. |
limit | Top 8–12 entries | The number of relevant documents per heat is limited. Too many results exceed the context window, while too few may miss key associated records. |
searchMode | mixedRecall | Professional terminology keywords and semantically related content must both be matched. This adapts to the professional expressions of casting processes. |
usingReRank | Enabled | Multi-source data may produce semantically similar but irrelevant results. Reranking improves retrieval accuracy. |
datasetSearchUsingExtensionQuery | Enabled | Expanded queries can cover synonymous expressions of casting professional terms. For example, "return charge" matches "scrap material ratio". |
collectionIds | Bind exclusive collections for casting processes, heat records, and quality reviews | Isolates non-casting data to avoid interference from irrelevant content in retrieval results. |
Three Common Configuration Mistakes
- Phenomenon: A large number of non-casting documents appear in retrieval results, leading to low accuracy. Cause: The
collectionIdssetting is not used to bind exclusive knowledge base collections, so the retrieval scope covers irrelevant data. - Phenomenon: Duplicate heat or mold time records appear after merging multiple casting-related datasets. Cause: No deduplication rule based on heat number and mold time fields is configured, so duplicate entries are not filtered during index merging.
- Phenomenon: Segmented casting knowledge base content cannot be exported for local migration. Cause: Server original files are copied directly without retaining segmented metadata, or the official export process is not used to export structured knowledge base fragments.
How to Confirm Correct Configuration
- Manually input the core parameters of a heat record, and check that retrieval results prioritize returning associated documents for the same heat.
- Verify that the retrieval configuration binds exclusive casting knowledge base collections, confirming that irrelevant data is not included in the retrieval scope.
- Validate the deduplication logic after merging datasets, confirming that duplicate heat and mold time records have been filtered.
- Test the retrieval function connected to an external casting technical website, confirming that page content can be correctly recalled.
The values provided are common starting points and should be measured against the reader's own samples.