What this category of data looks like
Pulp and paper manufacturing business data primarily comes from wood pulp and waste paper batch ledgers, DCS/QCS control systems, on-paper machine measurement equipment, laboratory testing systems, and energy and water treatment systems. Daily documents include pulp blending sheets, beating records, paper machine operation logs, online quality curves, laboratory reports, and sheet break and product change records. Data update cycles cover real-time (paper machine operating parameters, pulp blending adjustments), shift-based (work section summaries), and per roll (production batch archiving). Document fields include pulp ratio, beating consistency, paper machine speed, dryer section temperature, grammage, moisture content, thickness, sheet break duration, and more. Some documents link production batch numbers and roll IDs. Units include percentage, ton, meters per second, degrees Celsius, and others.
What constraints these characteristics impose on knowledge base retrieval and recall
The multi-source, high-frequency update data characteristics of pulp and paper manufacturing require knowledge base retrieval to support precise filtering by production batch and roll number. This prevents the recall of irrelevant batches or records. Real-time paper machine operating parameters and shift-based pulp blending adjustment records require retrieval to adapt to high-frequency update cycles. This avoids recalling expired production data. Multi-field parameters with clear units such as grammage, moisture content, and beating consistency require retrieval to match corresponding units. This prevents parameter confusion. Sheet break and product change records linked to specific production rolls require retrieval to incorporate process dimensions. This ensures recalled results match the current production process data, and supports accurate process adjustment.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.82 | Matches the precise matching requirements of pulp and paper manufacturing process parameters, filters low-relevance redundant results |
limit | Top 4–7 results | Adapts to the number of process records per production batch, avoids exceeding large model context length limits |
usingReRank | Enabled | Optimizes sorting of multi-source process data, prioritizes recalling records from core processes such as beating and papermaking |
| `datasetSearchUsingExtensionQuery | Enabled | Supplements extended queries for pulp and paper manufacturing professional terminology, covers more process documents |
collectionIds | By knowledge base collections for pulping, papermaking, and quality departments | Matches pulp and paper manufacturing departmental business scenarios, avoids recalling irrelevant cross-department data |
Three common mistakes
- Phenomenon: Retrieval results include numeric paragraph IDs from the knowledge base (such as 12356). Cause: The paragraph ID output switch is not disabled, so the numbers from process records are included in the large model context.
- Phenomenon: Newly uploaded pulp blending sheets, beating records, and other documents are all created in the root directory. Cause: The parent ID parameter format is incorrect, and the FastGPT required directory ID format is not matched, so the system cannot identify the parent directory.
- Phenomenon: After upgrading to FastGPT v4.17.0, the knowledge base can query data, but no results are output when passing to the large model. Cause: The context format of retrieval results is not adapted to the new output rules after the upgrade, so the large model cannot read the recalled content.
How to confirm the configuration is correct
- Process-related questions such as "Current batch beating consistency adjustment parameters" are submitted. Retrieval results are checked to confirm they include corresponding pulp blending sheets and beating records, and no paragraph ID information is present.
- The knowledge base directory management page is accessed. Newly uploaded documents are confirmed to be stored in corresponding subdirectories categorized by pulping, papermaking, and quality departments.
- A retrieval is triggered. The number of returned results is checked to confirm it falls within the 4–7 range, and the similarity value matches the preset range.
- Large model call logs are reviewed. Retrieval results are confirmed to include process parameters with units such as grammage, moisture content, and beating consistency, and no irrelevant cross-department data is present.
The values provided are common starting points and should be measured against the reader's own samples.