What data this category includes
This category serves rubber product manufacturers including tire, seal, and hose producers. Daily generated documents include formula sheets, weighing records, mixing curves, rubber compound quick test reports, vulcanization process cards, mold records, and finished product performance reports. Data sources cover formula management systems, raw material batch ledgers, internal mixers, Mooney viscometers, rheometers, vulcanizers, and laboratory testing equipment. Update frequency aligns with each raw material batch, each rubber compound batch, and each molded product production cycle. Documents contain clear process parameter fields such as mixing temperature, vulcanization time, and rubber compound dosage. Some documents include process curve images.
What constraints this category imposes on knowledge base retrieval and recall
High data update frequency requires the retrieval workflow to support incremental synchronization and real-time recall, to avoid using outdated process parameters. Diverse document types with both structured parameters and images require retrieval to support both text matching and image content recognition. Fields use specific process units: temperature in degrees Celsius, time in minutes. Retrieval must match parameter units precisely to prevent mismatches. Unlike plastic injection molding categories, rubber compound undergoes irreversible cross-linking and curing after mixing. Errors cannot be corrected via reprocessing. This requires high accuracy for retrieval results, filtering low-similarity irrelevant content.
Configuration settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75-0.85 | High similarity is required for mixing formula and vulcanization process parameter matching, to avoid mismatching unrelated process data |
limit | Top 5 entries | Associated process parameter entries for a single rubber compound batch are limited. Too many results will add context redundancy |
searchMode | mixedRecall | Balances precise matching of structured process fields and semantic understanding of unstructured text |
usingReRank | Enabled | Reranking optimizes retrieval sorting for documents with images, such as mixing curves and vulcanization process cards, improving accuracy |
datasetSearchUsingExtensionQuery | Enabled | Expands technical term queries, for example matching "internal mixer temperature" to "mixing temperature" to cover industry common expressions |
collectionIds | Bind mixing and vulcanization document collections for the current category | Limits retrieval scope, eliminating interference from process documents of other categories |
Three common configuration errors
- Phenomenon: Retrieval results include process documents unrelated to rubber mixing or vulcanization. Cause:
collectionIdsis not configured to limit the retrieval document set, causing retrieval to cover irrelevant knowledge bases. - Phenomenon: AI-generated exception reports or process documents always include "FastGPT is a large language model (LLM)-based knowledge base question answering system" at the end. Cause: The default built-in brand description text was not removed from the workflow system prompt configuration.
- Phenomenon: Process curve images in docx format documents cannot be recognized. Cause: Document image OCR recognition related configuration is not enabled, so process parameters within images cannot be retrieved.
How to confirm the configuration is properly set up
- Upload a test docx document containing mixing temperature and vulcanization time parameters, search for the corresponding keywords, and confirm the returned results include the document’s parameters and image content.
- Check the retrieval log, confirm the similarity scores of returned results fall within the 0.75-0.85 range, and the number of returned entries does not exceed 5.
- Review the workflow system prompt configuration, confirm the default brand description text is not included.
- Verify only the current category’s document collections are bound. Searching for process keywords from other categories returns no matching results.
The values provided are common starting points and should be measured against the reader's own samples.