What This Category of Data Looks Like
Precast concrete component data is sourced from BIM or detailed design documents, reinforcement and concrete batch records, mold table and curing kiln operation data, laboratory test reports, and QR code traceability systems. Daily documents include component detailed drawings, reinforcement cutting lists, mold inspection records, pouring and vibration records, curing records, demolding and factory inspection reports. Data is updated corresponding to each individual component, each batch of materials, and each curing kiln batch. Fields include component number, reinforcement specification, curing temperature, demolding strength, and others. Units include millimeters, megapascals, degrees Celsius, hours, and others.
Constraints Imposed on Knowledge Base Retrieval and Recall
Unique data characteristics of precast concrete components impose multiple constraints on the retrieval and recall process. First, high-frequency single-item and batch-level updates require retrieval indexes to support near-real-time updates, to prevent recalling outdated curing or inspection data. Second, document types that mix structured tables (such as reinforcement cutting lists) and unstructured drawings (such as detailed drawings) require dual retrieval capabilities for both keywords and semantics. Third, dense professional terminology (such as sleeve embedment position, vibration compactness) requires retrieval matching to have accurate professional semantic recognition capabilities. Fourth, the associated attribute where each component corresponds one-to-one with on-site lifting positions requires recall results to be bound to the corresponding component number, to avoid mixing irrelevant data across components.
Configuration Recommendations
This configuration is based on FastGPT v4.17.0, optimized for the business characteristics of precast concrete components:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Adapts to the matching accuracy of precast component professional terminology, avoids mixing irrelevant documents with low matching degrees, and covers semantic deviations of professional vocabulary |
limit | Top 6–10 entries | Covers the number of associated documents involved in single equipment inspection, avoids excessive context exceeding the model's processing limit |
searchMode | mixedRecall | Adapts to scenarios with mixed structured and unstructured documents such as detailed drawings and cutting lists, balances accurate keyword matching and semantic recall capabilities |
usingReRank | Enabled | Performs secondary ranking on initial recall results, corrects semantic matching deviations of professional terminology, and improves result relevance |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms related to inspection and maintenance, for example, expanding "mold table inspection" to "mold table check" to cover retrieval needs for more professional expressions |
collectionIds | Configured per factory-specific collection | Isolates precast component data from different factories, avoids recalling irrelevant data across tenants |
Three Common Configuration Errors
- After uploading table documents such as reinforcement cutting lists, no matching results are returned when searching for "reinforcement sleeve position". The cause is that multi-vector support is not enabled for table data, and only plain text content is vectorized, making it impossible to extract structured field information within the table.
- A parameter error is returned when adding curing records to the knowledge base via an interface. The cause is that required fields such as component number and curing kiln number are not passed in accordance with the metadata requirements of precast component documents, leading to interface verification failure.
- Recall results mix precast component data from other factories. The cause is that exclusive
collectionIdsare not specified, leading to cross-tenant knowledge base data being included in the recall range.
How to Verify Successful Configuration
- Upload a precast component reinforcement cutting list, search for "reinforcement sleeve position", and check whether retrieval results include the corresponding field content in the table, to confirm that multi-vector support is enabled.
- Call the knowledge base addition interface, pass in the metadata and text content of a curing record, and check whether the interface return status is successful, to confirm that parameter configuration meets requirements.
- Switch to different knowledge base collections, search for the same equipment inspection keyword, and check whether results only include precast component data from the current factory, to confirm that the
collectionIdsconfiguration is correct. - Initiate an equipment inspection query, check whether the number of recall results falls within the 6–10 range, to confirm that the
limitconfiguration meets expectations.
The configuration values listed here are common starting points, and should be evaluated against specific deployment samples.