What data looks like for this category
Core entities in the general mechanical equipment assembly category are manufacturers of pumps, valves, compressors, reducers, machine tools and non-standard equipment. Daily documents include assembly process cards, BOMs, torque records, gap adjustment records, commissioning reports, factory inspection sheets and packing lists. Data sources include PLM/ERP systems, part inspection data, assembly tool collection equipment, test benches, MES systems and customer technical agreements. Update frequency is once per individual equipment or per order. Document fields include structured and semi-structured content: process number, torque value (unit N·m), gap value (unit mm), material code, temperature rise (unit ℃) and other related fields.
What constraints do these characteristics impose on vector models and indexing workflows
Multi-source heterogeneous data sources require index systems to support unified processing of mixed-format documents. This avoids vector generation deviations caused by format differences. High-frequency, small-batch update rhythms require adaptation to incremental indexing mechanisms. This ensures fast index synchronization for assembly documents of a single piece of equipment. Documents contain a large number of process parameters with precise units. This requires vector models to balance semantic association and numerical matching accuracy. It prevents retrieval errors caused by missing units. Multi-level BOM and process document structure features require indexes to retain hierarchical associations between nodes. This avoids loss of cross-process context information during retrieval. A reasonable maxToken setting is needed to accommodate long document context.
How to set configurations
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the semantic integrity of single segments of documents such as assembly process cards and BOMs. Avoids splitting that disrupts process or parameter associations |
| Overlap Length | 50–80 characters | Retains process connection information between adjacent chunks. Prevents loss of step association during cross-chunk retrieval |
| Similarity Threshold | 0.72–0.85 | Adapts to the precise matching requirements of process parameters. Filters low-correlation non-standard document fragments |
| Single Search Limit | Top 8 results | Covers the complete steps of multi-process assembly. Avoids missing key adjustment links |
maxToken | Set to the selected embedding model/service maximum supported input tokens | maxToken is the input-token maximum supported by the selected embedding model/service; verify it in that provider configuration. See the v4.17.2 model-field reference. It is not retrieval/LLM context or a FastGPT-wide fixed cap. |
defaultToken | Set from representative sample documents | defaultToken is the default token count for direct text chunking; tune it on representative source documents. See the v4.17.2 model-field reference. It is not a vector dimension, embedding-model input cap, or retrieval/LLM context. |
Three common misconfigurations
- When the Doubao-embedding-large index model is enabled, a custom request address and apikey are filled in, and a direct test is run, an error occurs. The cause is incorrect configuration of vector model request header parameters, or the custom address not pointing to a compliant vector service endpoint.
- Table data in the knowledge base does not have multi-vector support enabled. Numerical fields such as torque and gap fail to be retrieved correctly. The cause is failure to enable multi-vector splitting configuration for structured table data, with only overall text chunks used for vector generation.
- The single search limit is set too low. Retrieval results only return the top 3 entries, which cannot cover all links of a complete assembly process. The cause is failure to adjust the search limit threshold based on the multi-process, multi-parameter business characteristics of this category.
How to confirm configurations are properly set
- The knowledge base vector model configuration page in FastGPT v4.17.0 is accessed. The Doubao-embedding-large model is verified as selected, with the correct custom request address and apikey filled in. A test connection is executed without errors.
- A test document containing a BOM table and assembly process card is uploaded. Chunk preview results are viewed. Adjacent chunks are confirmed to have overlapping characters, with processes not forcibly split.
- A process parameter retrieval test is initiated. The query "reducer assembly torque requirements" is entered. Returned results are confirmed to include correct torque values and corresponding processes. The number of results matches the configured single search limit.
- A test assembly document is submitted. The index queue status is checked to confirm it shows completed. Incremental index updates are verified to work normally.
The values provided are common starting points and should be measured against the reader's own samples.