What This Category's Data Looks Like
Data for the pulp and paper category comes from wood pulp and waste paper batches, DCS/QCS systems, on-paper machine measuring equipment, laboratory testing, and energy and water treatment systems. Daily documents include pulp preparation sheets, beating records, paper machine operation logs, online quality curves, laboratory reports, and sheet break and product change records. Data update frequencies fall into three categories: real-time, per shift, and per roll. Fields cover pulp ratio, beating degree, paper machine speed, grammage, moisture content, and more. Units include tons, seconds, °SR, g/㎡, %, and others.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall?
This category's data characteristics create multiple constraints for the knowledge base retrieval and recall process. Heterogeneous data from multiple sources includes DCS/QCS curves and paper-to-electronic operation records. This requires retrieval to support mixed-format recall. Real-time paper machine operation data needs frequent knowledge base index refreshes. This prevents retrieval of expired information. Exclusive sheet break and product change records for each roll must be associated and retrieved by production batch. This improves matching accuracy. Professional fields such as beating degree and grammage must retain their original units and parameter names. This avoids understanding discrepancies across departments.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Semantic similarity threshold for matching pulp and paper professional terminology, to avoid retrieving irrelevant pulp preparation or sheet break records |
limit | Top 8 entries | Covers key shift parameters for a single paper machine while controlling context token usage |
searchMode | mixedRecall | Combines semantic retrieval and keyword matching to support retrieval of professional fields with units such as beating degree and grammage |
usingReRank | Enabled | Re-ranks recall results by business relevance, prioritizing high-frequency issues such as wet end and dryer section balance, sheet break handling |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands professional term synonyms, such as matching "beating" to documents related to "fiber defibration" |
collectionIds | Bind corresponding knowledge base collections by department | Isolates dedicated data for preparation, pulping, papermaking and other departments to avoid cross-department retrieval interference |
Three Common Misconfigurations
- Garbled text appears in retrieval results when using a thinking model. This occurs when semantic encoding rules for the embedding model and thinking model are not unified, leading to abnormal parsing of professional terms.
- Images exported from on-paper machine measuring equipment cannot be automatically indexed in a local deployment environment. This happens when parsing plugins for the corresponding image formats are not configured, or images exported from DCS/QCS systems are not included in the index scope.
- Prompt words stored in the knowledge base cannot be dynamically loaded. This occurs when the extended query matching function is not enabled, or prompt word documents are not marked as dedicated retrieval sources.
How to Confirm Proper Configuration
- Submit the query "Adjust beating degree to 35°SR". Verify that recall results include corresponding beating records and pulp preparation sheets, and that rankings align with business priorities.
- View the retrieval log. Confirm that the similarity scores of returned results fall within the
0.72–0.80interval, with no abnormally low or high scores. - Test uploading a paper machine online quality curve image in the local deployment environment. Confirm that the index status shows "Completed" with no parsing failure prompts.
- Submit a query containing professional terms. Confirm that synonyms automatically expanded by
datasetSearchUsingExtensionQueryare correctly matched to relevant documents.
The values provided here are common starting points and should be measured against the reader's own samples.