What the Data for This Category Looks Like
The main entities in the general mechanical equipment assembly category are manufacturers of pumps, valves, compressors, reducers, machine tools, and non-standard equipment. Daily generated documents include assembly process cards, BOMs, torque records, clearance adjustment records, commissioning reports, factory inspection sheets, and packing lists. Data sources cover PLM/ERP systems, part inspection data, assembly tool collection equipment, test benches, MES systems, and customer technical agreements. Update frequency is once per device or per order. Document fields include technical parameters with units, such as torque values (N·m), clearance values (mm), coaxiality (μm), and others.
What Constraints These Characteristics Impose on Vector Models and Indexes
This category’s data updates per device or per order. Indexes must support incremental updates to avoid resource consumption from full rebuilds. Document types include structured BOMs, semi-structured inspection records, and unstructured process files. A chunking strategy is required to adapt to the varying paragraph lengths of different documents. Fields have clear units. Vector models must retain unit information to improve matching accuracy for technical parameters. The multi-variant order-based assembly model requires indexes to support grouping retrieval by order dimension. This prevents mixing of assembly data across orders.
How to Set the Configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the typical paragraph length of documents such as assembly process cards and BOMs, avoids splitting critical technical descriptions |
| Overlap Length | 100–150 characters | Retains contextual connections between adjacent chunks, adapts to continuous operation steps in clearance adjustment records and torque records |
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. |
| Similarity Threshold | 0.72–0.80 | Distinguishes qualified and unqualified assembly inspection records, filters low-match irrelevant data |
| Single Search Limit | Top 8 entries | Covers all inspection records and process parameters required for single-device assembly, avoids interference from redundant data |
defaultToken | Calibrated based on actual measurements | Adapts to the total token count of documents uploaded in a single batch, balances index construction efficiency and content completeness |
Three Common Mistakes
- Retrieval results include assembly records across multiple orders. Cause: No grouping dimension is set for the index using order ID. This leads to mixing of assembly data from different batches, and prevents accurate matching of the target order’s audit requirements.
- Vector index construction fails, and the interface returns a token limit error. Cause: The
maxTokenparameter is not configured. The length of a single chunk of text exceeds the model’s supported upper limit, triggering format verification failure. - Insufficient matching accuracy for retrieved clearance adjustment records. Cause: The chunk length is set too small. This splits continuous clearance adjustment steps and loses contextual information from preceding and following processes, reducing vector matching accuracy.
How to Confirm Proper Configuration
- Upload the assembly process card, BOM, and torque records for a single reducer. Check the index generation panel in FastGPT v4.17.0, and confirm that the progress is normal with no error prompts.
- Input an audit question that includes specific parameters, such as "What are the coaxiality inspection standards for this batch of pumps". Verify that the retrieval results include the inspection data collected by the corresponding test bench, and that the fields match accurately.
- Upload the packing list and commissioning report for a new order. Verify that the incremental index update is completed, and the newly uploaded documents can be retrieved normally.
- View the index configuration panel, confirm that the values of all configuration items match the preset plan, and that there are no unsaved modifications.
The values provided here are common starting points, and should be evaluated against the reader’s own samples.