What data for this category consists of
This category of data comes from enterprise EAM/CMMS systems, equipment sensor collection terminals, operator daily point inspection records, maintenance work orders and fault analysis archives, spare parts inventory lists, and special equipment statutory inspection documents. Data update cycles cover shift-based inspection submissions, daily/weekly/monthly maintenance plan releases, and real-time archiving when faults occur. Daily documents include standardized equipment ledger entries, graded point inspection checklist templates, shutdown maintenance task sheets, and spare part replacement comparison tables. Fields include unique device identification numbers, point inspection parameter thresholds, maintenance step sequence numbers, spare part models and inventory surpluses. Some documents include statutory inspection expiration date fields.
Constraints on knowledge base retrieval and recall from these characteristics
Decentralized data sources and large differences in update frequencies require the knowledge base to support multi-format batch import and incremental updates. Retrieval must prioritize matching the latest point inspection and fault data. Equipment ledgers and inspection checklists have high standardization but fine-grained details. Retrieval and recall must accurately match exact fields such as device numbers and parameter thresholds to avoid generalized results. Spare parts lists and statutory inspection documents have strict timeliness requirements. Retrieval results must filter expired data, and associate the device’s affiliated workshop and production scheduling window to avoid recommending maintenance plans that conflict with current production. Maintenance work orders and fault analysis have a high proportion of unstructured text. Retrieval semantic matching capabilities must be optimized to ensure accurate association between fault descriptions and historical maintenance solutions.
Configuration settings
This configuration is based on the knowledge base retrieval module of FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precise thresholds for equipment point inspection parameters, filters low-relevance non-equipment content |
limit | Top 8 entries | Balances the number of retrieval results and context length, adapts to the needs of maintenance teams for quick reference |
searchMode | mixedRecall | Balances exact fields such as device numbers and semantic matching for fault descriptions |
usingReRank | Enabled | Re-ranks initial retrieval results, prioritizes returning the latest point inspection and maintenance data |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms for point inspection terminology, covers colloquial point inspection descriptions commonly used by operators |
collectionIds | Configured by workshop/device type grouping | Limits retrieval scope to the point inspection and maintenance knowledge base of the corresponding workshop, avoids irrelevant cross-workshop results |
Three common configuration errors
- When selecting a knowledge base in the interface, the variable reference dropdown has no selectable values, and fields display as empty. The cause is failure to bind target documents such as equipment ledgers and inspection checklists as datasets that can be called by variables, or failure to configure correct application access permissions for the dataset.
- Under the same configuration, online chat and API calls return significantly different results. No valid retrieval results are returned when stream is set to false and detail is set to true. The cause is that the API request does not correctly pass the
collectionIdsparameter, or fails to bind the same knowledge base grouping as the online chat. - Chinese garbled characters appear after importing CSV-format inspection checklists or spare parts lists, even after manually adjusting character encoding settings. The cause is that the CSV file does not have a UTF-8 BOM header, or the actual file encoding is GBK or another non-standard UTF-8 extended format, causing abnormal knowledge base parsing.
How to verify successful configuration
- Navigate to the knowledge base management page, check that the import format and encoding of the target dataset match, and verify that parsed results for single inspection checklist entries have no garbled characters.
- In the application configuration interface, select the corresponding knowledge base grouping, and test whether the variable reference dropdown displays available fields such as device numbers and inspection items.
- Enter the same fault description in both online chat and API debugging tools, and check that the number and content of returned retrieval results are consistent.
- Adjust the
similarityparameter threshold, test changes in retrieval result relevance, and confirm that the threshold meets the matching requirements of the current business.
The values provided are common starting points and should be measured against the reader's own samples.