What the data for this category looks like
Daily documents for general mechanical equipment assembly include assembly process cards, multi-level BOMs, torque records, gap adjustment records, commissioning reports, factory inspection certificates, and packing lists. Data sources cover PLM/ERP systems, part inspection data, assembly tool collection data, test bench test data, MES systems, and customer technical agreements. Data is updated once per equipment unit or per order. Fields include equipment model, order number, torque value (unit: N·m), gap value (unit: mm), coaxiality parameter (unit: μm), and others. Field structures vary across different documents based on the assembly stage.
What constraints these characteristics impose on multi-turn dialogue and prompt engineering
Since data is updated per equipment unit or per order, multi-turn dialogue must bind a unique identifier to avoid mixing data across orders. The structure of multi-level BOMs and multiple types of assembly records requires prompts to explicitly categorize and call corresponding data sources, preventing confusion between document content from different assembly stages. The requirement that fields have specific units means prompts must guide the output of process and exception report content that matches those units. The multi-step workflow for a single assembly unit means conversation context must retain records of key stages, while also controlling context size to ensure response efficiency.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Bind the unique identifier of the current assembly order | Ensure conversation context only associates with assembly data for the corresponding order, avoiding cross-order confusion |
messages | Retain the most recent 20 rounds of conversation content | Balance context completeness and response speed, adapting to the multi-step workflow of single-unit assembly |
variables | Store key parameters such as equipment model, order number, torque threshold (e.g., 120 N·m) | Quickly call preset assembly standards and order information without repeated queries |
| Maximum context window | 8000–12000 token | Accommodate the text content of multi-level BOMs and multiple assembly records, avoiding truncation of critical information |
| Single tool call limit | Top 3 matching data entries | Focus on the most relevant assembly process and exception data, reducing interference from redundant information |
Three Common Configuration Mistakes
- Phenomenon: Database tool calls return a 400 status code (no body). Cause: No order-specific parameters for database connections are configured in
variables, resulting in missing identifiers for the corresponding data source during calls. - Phenomenon: Conversation context loses some assembly records. Cause:
chatIdis not bound, causing conversation contexts from different orders to be loaded together and key assembly data to be overwritten. - Phenomenon: Generated exception reports have mismatched units. Cause: The prompt does not explicitly require matching the units from assembly records such as N·m and mm, leading to output inconsistent with on-site data.
How to Confirm Proper Configuration
- Initiate a test conversation, enter the order number of a specific equipment unit, and check whether the corresponding equipment BOM and process card data are automatically loaded in
variables. - Conduct multiple consecutive rounds of questions about gap adjustment and torque verification, and confirm that previous conversation content is not lost from the context.
- Call the database tool to query assembly records, and check whether the returned results only include data related to the current order.
- Generate a draft exception report, and verify that the units in the report match those from on-site records such as N·m and mm.
The values provided here are common starting points and should be evaluated against the reader's own samples.