What This Category of Data Looks Like
Ready-mixed concrete production data originates from engineering orders, sand, gravel, cement and admixture inspection records, aggregate moisture content measurement ledgers, mixing station control system logs, and project laboratory archived documents. Data is updated per vehicle production record and per batch of incoming raw materials. Laboratory test data is aggregated daily. Daily documents include mix proportion notices, opening identification forms, production and transport orders, raw material inspection reports, test block curing records, and strength test reports. Core fields include mix proportion parameters, aggregate moisture content, slump requirements, and test block strength values, with corresponding units such as mm and MPa.
Constraints Imposed on Knowledge Base Retrieval and Recall
Ready-mixed concrete production data is updated per vehicle, per raw material batch, and daily. Individual production records are linked to engineering orders, raw material sources, and laboratory test results. Retrieval must accurately match production time, raw material batches, and engineering parameters. Daily documents contain numeric test indicators and procedural approval records. Clear association logic must be maintained between professional terminology and business expressions. The high-frequency update feature requires the knowledge base to support incremental synchronization of the latest test block strength and aggregate moisture content data, to avoid lagging retrieval results. Quality traceability must cover the entire mix proportion, production, and transport workflow. Retrieval and recall must link multi-dimensional documents to avoid information silos from single-document searches.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Semantic similarity threshold for matching concrete professional terminology and business descriptions, balances retrieval precision and coverage |
limit | Top 10 entries | Limits the number of documents recalled per request, avoids context overload, and adapts to multi-dimensional associated retrieval needs |
searchMode | fullTextRecall | Adapts to cross-field associated retrieval needs for mix proportions and test data, covers full-text keyword and semantic matching |
usingReRank | Enabled | Re-ranks recalled results by relevance, corrects semantic bias in initial retrieval, and improves ranking accuracy for professional documents |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands search terms for business questions, for example expanding "slump non-conformity" to "slump requirements + measured value deviation", to adapt to professional expression habits in the concrete industry |
collectionIds | Configured by mixing station/laboratory grouping | Restricts retrieval scope to exclusive documents of the corresponding department, avoids interference from irrelevant cross-project data |
Three Common Misconfigurations
- Symptom: Local files cannot be read during conversation, but file upload works normally in the knowledge base backend. Cause: Local file read permission mapping is not configured, or the uploaded file format exceeds supported ranges, such as non-standardized test record scan files.
- Symptom: Knowledge base retrieval occasionally returns empty results, but normal recall works in knowledge base backend testing. Cause: The
similaritythreshold is set too high, or real-time production data has not completed incremental synchronization, resulting in no matching latest documents during retrieval. - Symptom: Retrieval results are returned even when the question is unrelated to knowledge base content. Cause: The
limitparameter is not configured, or the threshold is set too low, recalling irrelevant document fragments, leading the model to generate unrelated content.
How to Verify Proper Configuration
- Upload one latest opening identification form to the corresponding knowledge base, initiate a question containing "opening identification requirements", and check whether the recalled results include the core content of this document.
- Adjust the
similaritythreshold, initiate a question containing professional terminology, and compare the number of recalled results and relevance before and after adjustment. - After enabling
usingReRank, check whether the ranking of retrieval results aligns with business relevance, for example, documents related to slump appear before unrelated transport records. - When verifying multi-account isolation, switch accounts and confirm whether the retrieval scope corresponds to the configured
collectionIds, ensuring documents from different departments do not interfere with each other.
The values provided are common starting points and should be measured against the reader's own samples.