What data in this category consists of
Data comes from the mining, raw material, calcination, grinding, quality control, and energy departments of cement enterprises. It covers business documents including raw material batching plans, kiln operation records, laboratory reports, quality control notices, equipment inspection records, and energy reports. Data sources are online composition analysis systems, DCS control systems, laboratories, kiln and grinding equipment, raw and fuel ledgers, and energy metering devices. Update frequencies include real-time, per-shift, and daily. Document fields include material composition values, equipment operating temperatures, inspection cycles, and energy consumption values. Units include degrees Celsius, tons, kilowatt-hours, and other standard industrial units.
Constraints on knowledge base retrieval and recall
Differences in data formats across multiple sources require unifying field and unit identifiers before documents are added to the repository. This prevents loss of valid information due to format mismatches during retrieval. Real-time, per-shift, and daily update frequencies require retrieval scopes to cover datasets with different update cycles. This ensures that core documents such as the latest kiln operation records and quality control notices are recalled. Documents related to equipment inspection and kiln/grinding equipment make up a large share of the total. Retrieval must prioritize matching keywords for equipment inspection and maintenance to improve the recall rate of relevant results. Documents including core business materials such as raw material batching plans require accurate retrieval results. Incorrect recalls could disrupt production links including clinker quality control and kiln system operation.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.78 | Most documents in this category use professional industrial terminology. This threshold balances the matching accuracy of professional terms and recall scope, and is compatible with the semantic retrieval logic of FastGPT v4.17.0 |
limit | Top 8 entries | The number of documents related to equipment inspection and maintenance in this category is moderate. This setting avoids overloading the model with excessive results while ensuring coverage of valid information |
searchMode | embedding | This category of documents contains a large number of professional operation records and terms. Semantic retrieval is better able to match user queries related to equipment inspection and maintenance than keyword retrieval |
usingReRank | Enabled | Multi-source data may contain documents with similar semantics. Reranking filters results that best match the query, solving the problem of cluttered recall results |
datasetSearchUsingExtensionQuery | Enabled | User queries about equipment inspection and maintenance may contain colloquial expressions. Extended queries can match more accurate document content |
collectionIds | Bind the exclusive dataset ID for cement clinker and grinding | The retrieval scope must be limited to avoid mixing in documents from other categories, ensuring that retrieval results only cover business documents of this category |
Three common mistakes
- Symptom: After setting
similarityto 0.77, some documents matched by professional terms are judged as irrelevant, and the model returns no relevant information. Cause: The semantic similarity calculation for professional terms in this category has deviations. A threshold of 0.77 is too high, filtering out some documents with sufficient matching degrees. - Symptom: Semantic retrieval returns more than 10 results, and the content most matching the query is not filtered out. Cause: The
usingReRankconfiguration is not enabled, and no reranking sorting is performed on the recall results, leading to cluttered results. - Symptom: After uploading equipment inspection records in Excel format, the table content cannot be read during retrieval. Cause: The knowledge base is not configured to support parsing of Excel files, and the index switch for the corresponding file type is not enabled.
How to confirm the configuration is correct
- Upload typical documents of this category, perform queries related to equipment inspection, and check whether the returned results include exclusive documents such as raw material batching plans and kiln operation records.
- Adjust the
similarityconfiguration, compare recall results under different thresholds, and confirm that the matching degree meets business requirements. - After enabling
usingReRank, verify that the order of recall results matches the relevance to the query. - Upload equipment inspection records in Excel format, perform a query, and confirm that the table content can be correctly retrieved.
The values given are common starting points and should be measured against the reader's own samples.