What data for this category looks like
Data for this category comes from daily outputs and system entries from manufacturing enterprise equipment departments, maintenance teams, and public utilities departments. It covers equipment ledgers, inspection forms, maintenance plans, work orders, failure analysis, spare parts lists, and special equipment inspection records. Data sources include EAM/CMMS systems, equipment sensor data, operator inspection records, maintenance history archives, spare parts inventory data, and statutory inspection information. Update frequencies cover shift-based, daily, weekly, monthly cycles, and immediate triggers when faults occur. Document structures include fields such as unique equipment identification, inspection items, fault codes, spare part models, and inspection validity periods.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Multi-source heterogeneous document types require retrieval to support cross-document field association. For example, associate inspection records, work orders, and spare parts lists via unique equipment identification. Frequently updated inspection and fault data require retrieval results to prioritize recently entered valid content, to prevent outdated information from interfering with decision-making. The statutory attributes of special equipment inspection records require retrieval to identify and prioritize documents approaching validity periods, to support compliance checks. Additionally, maintenance scenarios require quick matching of precise fields such as equipment model and fault code, to avoid generic retrieval returning irrelevant content that reduces maintenance decision efficiency.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | The semantic similarity of text related to equipment inspection and faults must stay within a reasonable range. Too low a value returns irrelevant content, while too high a value misses valid maintenance records |
limit | Top 8–12 entries | Maintenance scenarios balance information completeness and response speed. Too many single-recall results increase context processing load, while too few fail to cover multi-dimensional maintenance reference information |
searchMode | mixedRecall | Equipment-related documents include standardized numbers such as equipment models and fault codes, plus natural language failure analysis. The hybrid mode matches both structured fields and semantic content |
usingReRank | Enabled | Semantic similarity judgments for equipment faults and inspection records are easily disrupted by technical terms. Reranking filters low-quality recall results to improve accuracy |
datasetSearchUsingExtensionQuery | Enabled | Maintenance scenarios recall more matching historical work orders and maintenance plans by expanding the original query, such as adding equipment model and fault type, improving retrieval coverage |
collectionIds | Bind the collection ID of the corresponding equipment management knowledge base | Limit the retrieval scope to documents exclusive to this category, to avoid mixing irrelevant data from other categories |
Three Common Misconfigurations
- Phenomenon: After uploading a knowledge base file with a Chinese filename via API, retrieval fails to match the corresponding document content. Cause: The filename encoding parameter is not configured correctly, causing a break in the association between file metadata and actual content.
- Phenomenon: The number of retrieval results is far lower than the set
limitvalue. Cause:datasetSearchUsingExtensionQueryis not enabled, and the original query does not include sufficient unique equipment identification fields, leading to an overly narrow recall scope. - Phenomenon: Single retrieval takes more than 10 seconds. Cause:
usingReRankis not enabled andcollectionIdsis not limited, leading the retrieval scope to cover the entire platform knowledge base and increasing computational load.
How to Confirm Proper Configuration
- Upload a test file with a Chinese filename, use the FastGPT v4.17.0 API to retrieve the filename keyword, confirm that the corresponding document content can be returned normally.
- Set
similarityto 0.8, retrieve a known equipment fault code, check whether the relevance of the returned results meets expectations. - View the retrieval log, confirm that the number of returned results matches the set interval of
limit, and only includes documents from the bound knowledge base. - Simulate a maintenance scenario query, check whether the returned results include multiple types of documents such as equipment ledgers, maintenance work orders, and spare parts lists.
The values given are common starting points and should be measured against the reader's own samples.