What Data for This Category Looks Like
Daily documents for general mechanical equipment assembly include assembly process cards, BOMs, torque records, clearance adjustment records, commissioning reports, factory inspection sheets, and packing lists. Data sources cover PLM/ERP, part inspection data, assembly tool collections, test benches, MES, and customer technical agreements. Updates follow individual equipment or per order. Each document includes specific fields such as order number, assembly date, torque value (unit: N·m), clearance value (unit: mm), and coaxiality tolerance (unit: μm).
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
Multi-source and heterogeneous data formats require the retrieval system to support matching different document structures, to avoid recall deviations caused by inconsistent fields. High-frequency updates require the retrieval scope to cover the latest order and assembly data, with regular synchronization of incremental knowledge base content. Specific numerical fields and units require matching parameter precision during retrieval, to prevent issues like torque unit confusion and incorrect clearance value matching. The risk of mixed multi-category documents requires retrieval to be limited to the exclusive collection for general mechanical equipment assembly, to avoid mixing in parts machining documents.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | General mechanical equipment assembly has high parameter precision requirements, so low-similarity irrelevant results must be filtered |
limit | Top 10 entries | The number of assembly process-related document chunks for a single device is limited; 10 entries can cover complete assembly steps and BOM hierarchy |
searchMode | mixedRecall | Balances keyword matching and semantic matching, adapting to step descriptions in process cards and structured data in BOMs |
usingReRank | Enabled | Improves the sorting priority of highly relevant documents, avoiding interference from low-quality initial recall results |
datasetSearchUsingExtensionQuery | Enabled | Associates metadata from original documents, ensuring retrieval results match information such as order number and assembly date from the original document |
collectionIds | Exclusive collection ID for general mechanical equipment assembly | Limits retrieval scope to prevent mixing in documents from other categories |
Three Common Misconfigurations
- When uploading CSV-format assembly process documents, Chinese characters appear garbled while English content displays normally. This occurs when CSV files are uploaded directly via the text dataset upload function without using the template import feature, causing encoding parsing errors.
- Retrieval results include outdated versions of assembly process cards and fail to match the latest order data. This happens when incremental synchronization rules are not configured, so the knowledge base has not updated the latest assembly documents.
- Knowledge base retrieval results include a large number of irrelevant parts machining documents unrelated to general mechanical equipment assembly. This is caused by not specifying
collectionIds, leading the retrieval scope to cover the full knowledge base.
How to Verify Successful Configuration
- Upload a test CSV document containing torque values and clearance parameters via the template import function, and confirm that Chinese fields do not appear garbled.
- Submit a query for "Reducer assembly torque requirements", check that the
similarityscores of returned results fall between 0.75 and 0.85, and that the number of results does not exceed 10. - Access the knowledge base management page, confirm that the retrieval scope only includes the exclusive collection for general mechanical equipment assembly, with no documents from other categories mixed in.
- Submit a test query for multi-level BOM information, check that returned results cover relevant assembly steps and parts lists.
The values provided here are common starting points, and should be evaluated against your own sample data.