What Data for This Category Looks Like
Data for this category covers the full machining process. Sources include product drawings and 3D models, CAPP/CAM systems, machine tool collection data, tool management systems, MES systems, and measurement and inspection tool records. Daily documents include process routes, operation cards, tool lists, CNC programs, first-piece inspection records, patrol inspection forms, and measuring tool calibration records. Data updates trigger per work order, individual workpiece, sampling inspection, or tool change. Document fields include operation number, cutting parameters, tool life, inspection tolerance values, and calibration cycles. Units include rotational speed, feed rate, life hours, tolerance microns, and others.
What Constraints These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
Data for this category is scattered and has diverse update triggers. Knowledge base retrieval must support precise filtering across multiple dimensions such as work order ID, operation number, and tool number to avoid retrieving irrelevant historical data. Documents contain structured parameters and unstructured CNC program text. Retrieval weight configuration must be adapted for different content types to ensure effective retrieval of process parameters, inspection records, and program code. Data updates trigger per work order, sampling inspection, or tool change. Retrieval results must be real-time associated with the latest operation adjustments, tool life, and calibration records to avoid using expired documents. Multiple field units exist. Parameter identification logic must be unified during retrieval matching to prevent retrieval failures caused by unit differences.
The following configurations are based on FastGPT v4.17.0.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for structured content such as metal cutting machining process parameters and inspection records, and avoids retrieving low-relevance historical data |
limit | Top 10 entries | Covers multiple types of documents such as tools, inspections, and programs that may be associated with a single operation, and avoids excessive results exceeding the context length limit |
searchMode | mixedRecall | Balances structured field matching and full-text semantic retrieval, and adapts to diverse document types such as process routes and CNC programs |
usingReRank | Enabled | Improves the sorting priority of highly relevant documents, especially for defect analysis scenarios requiring precise matching such as tool wear and dimensional drift |
datasetSearchUsingExtensionQuery | Calibrated based on actual measurements | Adapts to professional terminology in metal cutting machining, optimizes the expansion logic of retrieval terms, and improves retrieval accuracy |
collectionIds | Bind the dataset ID of the current workshop/work order | Narrows the retrieval scope and avoids retrieving irrelevant process documents from other workshops |
Three Common Misconfigurations
- Phenomenon: The number of documents returned during retrieval is fixed to the default value, and the upper limit cannot be modified through configuration. Cause: Advanced configuration mode is not enabled, and only the default
limitvalue from basic configuration is used. - Phenomenon: The reranking function can be enabled normally in the knowledge base test interface, but the reranking results do not take effect when calling the large model via API. Cause: The
usingReRankconfiguration item is not enabled synchronously during API calls, and the setting is only completed in the test interface. - Phenomenon: Multiple completely identical operation cards or CNC programs appear in the retrieval results. Cause: The file deduplication switch of the dataset is not enabled, or duplicate uploaded documents are not filtered by work order ID during import.
How to Verify Successful Configuration
- Enter the knowledge base retrieval configuration interface, check whether the values of configuration items such as
similarity,limit, andusingReRankmatch the preset settings. - Upload a test operation card and tool list, perform retrieval, and confirm that the number of returned documents does not exceed the
limitsetting. - After enabling the reranking function, retrieve the query "tool wear causes dimensional drift", and confirm that highly relevant documents have higher sorting priority than low-relevance content.
- Import two completely identical process route documents, perform retrieval, and confirm that only one result is returned to verify that the deduplication configuration takes effect.
The values provided are common starting points and should be measured against the reader's own samples.