What the Data for This Category Looks Like
Data for this category comes from EAM/CMMS systems, equipment sensor collection terminals, operator inspection records, maintenance work orders, spare parts lists, special equipment inspection records, and similar sources. Update cycles include shift-based inspection submissions, daily and weekly maintenance plan updates, and immediate work order entry when faults occur. Documents are split into two types: structured tables and text. Fields include equipment number, inspection point, spare part model, inspection validity period, maintenance duration, and more. Units use equipment operation and maintenance specific identifiers such as millimeters, hours, and counts.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link?
Multiple heterogeneous data sources and uneven update frequencies require the retrieval link to filter recalled results by update time, prioritizing the latest inspection and maintenance data. Structured table documents make up a large share of the dataset, so precise indexing and matching of table content must be supported. Retrieval demands for precise fields such as equipment number and spare part model require adjusting matching weights to prevent irrelevant data from being included via generic matching. Bidding scenarios require quick location of historical maintenance cases and spare part replacement solutions, which have clear requirements for the timeliness and relevance of recalled results.
How to Configure Settings
The following configurations are based on FastGPT v4.17.0
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Match precise fields for equipment inspection, to avoid irrelevant equipment data being included in bid documents |
limit | Top 8 entries | Control the number of retrieval results, to avoid citing too many redundant maintenance cases in bid documents |
searchMode | mixedRecall | Combine equipment data from local knowledge bases and online industry maintenance standards, to meet the multi-source information needs of bidding scenarios |
usingReRank | Enabled | Reorder recalled inspection sheets and work order content, to prioritize matching high-priority maintenance needs in bidding scenarios |
datasetSearchUsingExtensionQuery | Enabled | Expand search terms for spare part models, to match replacement spare part list data |
collectionIds | Specify the collection ID of the equipment inspection and maintenance knowledge base | Limit the retrieval scope, to exclude document interference from other business categories |
Three Common Configuration Mistakes
- Enabling
usingReRankcauses persistent video memory usage. The symptom is that the reranking model process does not exit automatically after retrieval finishes. The root cause is that the automatic uninstallation logic for the reranking model after the retrieval process ends is not configured. - Importing inspection sheet table documents results in empty retrieval results. The symptom is that no valid content is returned for matching fields. The root cause is that the chunking and indexing configuration for table content is not enabled.
- Enabling
datasetSearchUsingExtensionQuerythrows an Invalid array length error. The symptom is that a prompt for invalid array length appears in the interface during retrieval. The root cause is that the extended query vocabulary format does not comply with system verification rules.
How to Confirm the Configuration Is Correct
- Upload a single standardized inspection sheet document, run a retrieval test that includes inspection points, and check if the returned results contain specific content from the table.
- View the system video memory monitoring panel, confirm that the reranking model is automatically uninstalled after retrieval finishes, with no persistent video memory usage.
- Enter a search term that includes a specific equipment number, verify that the
similarityscore of the returned results falls within the preset 0.72–0.85 range. - Run a retrieval test related to spare part models, confirm that expanded search results and online supplemented industry standard content are matched.
The values provided here are common starting points, and should be evaluated against the reader's own samples.