What data for this category consists of
Data for the general mechanical equipment assembly category mainly comes from PLM, ERP, and MES systems, as well as torque records, gap adjustment records, commissioning reports, and factory inspection sheets collected by assembly tools. Daily documents include assembly process cards, multi-level BOM, packing lists, and more. Fields include part numbers, torque values, gap parameters, coaxiality data, and more. Most units are newton meters, millimeters, and micrometers. Data updates upon completion of each device assembly and each order delivery. Each device has a dedicated set of assembly and inspection data.
Constraints imposed by these characteristics during model integration and configuration
The multi-level BOM and scattered heterogeneous data such as torque and gap records for this category require limiting total context length during model integration. This prevents context overflow caused by multi-document association. Per-device exclusive data updated per order requires configuring dynamic data loading logic. This avoids mixing data from different orders. Assembly parameters with precise units require the model to accurately identify and return results with attached units. Real-time Q&A needs for on-site inspection and maintenance require the retrieval process to support fast, concise question matching.
How to set the configuration
This configuration guide is written for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the total length of multi-level BOM and multiple assembly documents, prevents context overflow |
maxTokens | 2000–3000 token | Meets the need for detailed responses in maintenance Q&A, while controlling single-round response latency |
quoteMaxToken | 1500 characters | Limits the total length of referenced documents, ensures core assembly parameters are fully extracted |
temperature | 0.3–0.5 | Reduces response randomness, ensures accuracy of assembly parameters and inspection data |
AIPROXY_API_ENDPOINT | Fill in the exclusive address provided by the platform | Adapts to the interface configuration of the call chain, ensures successful model call connectivity |
| Workflow/API tool selection | auto | Allows the model to independently choose whether to call tools, adapts to fault troubleshooting needs during inspection |
Three common configuration mistakes
- Phenomenon: Large language model calls return connection timeout, or no clear error logs appear in the console. Cause:
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare not filled correctly, causing requests to fail to reach the specified interface. - Phenomenon: Retrieved assembly data does not match actual documents, and BOM information from other orders is mixed into results. Cause: The
maxContextlength is not limited, causing the context to load assembly data from multiple unrelated orders, mixing assembly parameters from different batches. - Phenomenon: Speech-to-text transcribed text cannot be accurately identified as industrial assembly terminology. For example, "preload force" is misrecognized as "preload lee", and "coaxiality" is misrecognized as "coaxial cross". Cause: A speech recognition model adapted for mechanical assembly scenarios is not selected, or access parameters for the corresponding model are not configured, resulting in insufficient accuracy for professional term recognition.
How to confirm the configuration is correct
- Initiate a single-round assembly parameter query, verify whether the model's returned results include the torque values and gap parameters specified in the document, and whether the reference source matches the actual document.
- Simulate the call chain, check whether the interface returns a normal status code, and confirm that
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare configured correctly. - Adjust the
temperatureparameter, compare response content across different values, and confirm that it meets the business requirements for result accuracy. - Import the complete assembly document for a single device, test whether a context overflow prompt is triggered, and confirm that the
maxContextparameter is configured appropriately.
The configuration values listed above are common starting points. Adjust them based on your own specific samples and use cases.