What Data for This Category Looks Like
Daily documents for this category include winding and core assembly records, vacuum drying records, insulating oil reports, partial discharge and withstand voltage test reports, and nameplate parameter tables. Data sources include design BOMs, silicon steel sheet and insulating material data, vacuum and drying equipment logs, oil test reports, and test station test records. Data is updated on a per-single-product or per-batch-of-materials basis. Each document contains clear fields such as insulation parameters, test values, material batch numbers, and equipment operation records. Most units use industrial standard units including millimeters, megapascals, and microcoulombs.
How These Characteristics Impact Knowledge Base Retrieval and Recall
The requirement to update data per single product or batch means retrieval and recall must accurately match the current batch’s material parameters and test data, to avoid mixing across batches. Documents contain many professional test values and insulation parameters, so prioritize recalling content strongly related to the current process link and filter out generalized text. Data sources span multiple system logs and reports, so support for cross-data-source associated retrieval is needed. Because products require full life cycle traceability, recall results must link to the source batch and equipment information of the corresponding document. Document structures vary: some are tabular test data, others are process records, so retrieval must support content recall across different formats.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches professional content such as insulation parameters and partial discharge values, filters irrelevant generalized text, and retains highly relevant process and test data |
limit | Top 8–12 entries | The number of process documents for a single product is limited. This range can cover required test records and assembly data, avoiding redundant recall |
searchMode | mixedRecall | Combines keyword and semantic retrieval, adapts to document types that contain both professional terminology and structured test data, and improves retrieval accuracy |
usingReRank | Enabled | The knowledge base accumulates updates with each batch of materials. Reranking optimizes the sorting priority of abnormal reports and insulation data, aligning with the logic of process drafting |
datasetSearchUsingExtensionQuery | Enabled | Automatically optimizes query expressions related to process abnormalities, supplements professional terms such as insulation and partial discharge, and improves retrieval matching accuracy |
collectionIds | Bind the exclusive dataset ID of the current production line/project | Isolates documents from different batches and production lines, avoids interference from cross-production line data, and ensures the accuracy of single-product traceability |
This configuration is based on FastGPT v4.17.0.
Three Common Configuration Mistakes
- API call responses do not include knowledge base data. The symptom is that interface returns only generic replies, with no knowledge base content related to processes or tests. The cause is failure to correctly bind the
collectionIdsparameter, leading to a retrieval scope covering irrelevant datasets, or failure to enable relevant knowledge base retrieval configurations. - Retrieval response time exceeds 30 seconds. The symptom is a timeout prompt on the interface or during API calls, with noticeably delayed returned results. The cause is enabling
usingReRankwithout adjusting the number of recalled entries, and knowledge base data volume has grown too large with accumulated batches, leading to excessive reranking calculation load. - Recall results do not include the corresponding document file names. The symptom is that returned fragments are not labeled with source documents such as insulating oil reports or partial discharge test reports. The cause is failure to configure source field display for recall results, or failure to enable associated recall of document metadata during retrieval.
How to Verify Configuration Is Correct
- Check the dataset binding configuration in the FastGPT backend, confirm that the
collectionIdsparameter is bound to the exclusive dataset of the current project. - Submit a test query containing insulation parameters and partial discharge abnormalities, and check whether returned results include relevant fragments from process documents or test reports.
- Review the retrieval log to confirm that configurations such as
searchMode,usingReRank, anddatasetSearchUsingExtensionQueryhave taken effect as set. - Cross-check the metadata of returned results to confirm that source information for corresponding documents is included, and verify that the file name association function operates properly.
The configuration values provided here are common starting points. Adjust them based on your own specific samples and use cases.