What data for this category includes
This category of data comes primarily from daily business documents of flow, pressure, temperature, and analytical instrument manufacturers and metrology laboratories. Documents include assembly and commissioning sheets, calibration procedures, original calibration records, error curves, metrology certificates, and software version logs. Data sources cover design documents, standard instrument monitoring data, calibration software outputs, laboratory information systems, and customer range requirements. The update frequency is after each calibration for every device. Documents contain fields such as traceability number, calibration date, error value, and range parameters. Some data includes physical units, such as pressure unit Pa, temperature unit ℃, and flow unit m³/h.
Constraints on knowledge base retrieval and recall from this category’s characteristics
The high-frequency update nature of this category’s data requires the retrieval link to support real-time incremental synchronization. This prevents recalling expired calibration records or assembly documents. Fields with attached physical units require unit matching during retrieval. This avoids mixing instrument data from different ranges. Mixed storage of multiple document types requires the retrieval system to support filtering recall results by document type. The uniqueness of core fields such as traceability number and calibration date requires recalled results to be accurately bound to the corresponding device’s unique identifier. This prevents mixing data across devices.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.70–0.85 | This category’s data requires precise matching of error values and traceability numbers. A threshold that is too low will introduce irrelevant calibration records, while a threshold that is too high may miss valid results |
limit | Top 6–10 results | Single-batch recall results must cover multiple sets of data from multi-point calibration, while avoiding overly long contexts that exceed model limits |
searchMode | embedding | The core retrieval requirement for this category is precise matching of calibration parameters and error curve features. Vector retrieval is suitable for mixed high-dimensional numerical and text retrieval scenarios |
usingReRank | Enabled | Semantic similarity of data such as calibration records and error curves requires secondary verification to improve the accuracy of recalling core parameters |
datasetSearchUsingExtensionQuery | Enabled | Can automatically supplement associated query terms such as customer range requirements and device models, to cover cross-document associated retrieval needs |
collectionIds | Specified by calibration document collection | Precisely limit the retrieval scope, avoid recalling general non-calibration documents, and improve retrieval efficiency |
Three Common Configuration Mistakes
- Issue: When calling the API interface for creating a document collection, PDF-format calibration procedures cannot be parsed correctly. Cause: PDF parsing parameters were not configured in the API request, and the default parsing logic does not adapt to the complex table and chart layout of industrial instrument documents.
- Issue: Unable to switch the vector model used for retrieval, resulting in matching accuracy that does not meet the retrieval requirements of calibration data. Cause: The vector model was not switched according to the system configuration process, and the default model does not adapt to this category’s mixed numerical and text retrieval scenarios.
- Issue: Calling the knowledge base retrieval interface returns missing fields or errors, and the traceability number of calibration records cannot be obtained. Cause:
collectionIdswas not limited to the calibration document collection, and the retrieval scope includes unstructured software version logs, resulting in failure to extract core business fields.
How to Verify Configuration is Correct
- Run a retrieval test targeting calibration procedures. Check whether the document types of the recalled results match expectations, and confirm that the
collectionIdsconfiguration takes effect. - View the similarity scores of retrieval results. Confirm that the
similaritythreshold setting meets the precise matching requirements of this category, and adjust the value range based on actual recall results. - Compare recall results before and after enabling
usingReRank. Confirm that secondary reranking improves the recall priority of core parameters such as error values and traceability numbers. - Call the API interface for creating a document collection. Verify that the PDF parsing parameters are correctly configured, and ensure that PDF-format calibration documents can be parsed normally.
The values provided here are common starting points, and should be evaluated against your own specific samples.