What data for this category looks like
Data for this category comes from material grade manuals, customer technical requirements, temperature control instruments, furnace temperature uniformity testing equipment, hardness testers, metallographic analyzers, and workshop MES systems. Daily documents include heat treatment process cards, furnace loading layout drawings, real-time furnace temperature curves, hardness test reports, metallographic structure analysis sheets, equipment inspection forms, and abnormal disposal sheets. Data is updated per furnace charge or per batch. Fields include furnace temperature set and measured values (unit: ℃), holding duration (unit: hours), part hardness values (unit: HV), metallographic structure type, and other relevant details. Corresponding batch production records must be associated with this data.
Constraints imposed by these characteristics on the tool calling and plugins workflow
This category’s data is updated per furnace charge and includes documents with both text and image content. Tool calling must support real-time retrieval of the latest batch-specific furnace temperature curves, inspection records, and other related data. The multi-type document structure requires tools to support both text report parsing and image visual analysis. Data field unit matching logic must be validated to prevent analysis errors caused by inconsistent units. Abnormal disposal sheets must be linked to corresponding equipment and production batches. Tool calling must bind the association rules between equipment ID and batch number to avoid calling historical data from unrelated batches. It is necessary to adapt to image download and parsing for multimodal tools, handle scenarios where image retrieval fails due to network issues or permission restrictions, and match the actual data transmission time requirements of the workshop intranet.
How to set the configuration
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | Auto mode | Adapt to dynamic inspection requirements per furnace charge, allowing the model to independently select tools for the corresponding furnace charge |
| Workflow/API function-call configuration | Enable multimodal calling | Support parsing image-based document data such as furnace loading drawings and furnace temperature curves |
| Tool call timeout threshold | 600 seconds | Cover the conventional workshop intranet transmission time for furnace temperature curve download and metallographic data retrieval |
| Multimodal tool call limit | First 3 entries | Match the number of core documents per furnace charge, avoid redundant calls and context overflow |
| Tool call retry count | 2 times | Address common scenarios such as image retrieval failure and temporary network jitter |
| DingTalk webhook configuration | Bind to workshop equipment alarm group | Push inspection abnormality notifications to the corresponding work group |
Three common errors
- When calling a multimodal tool, image parsing failure is returned, and the log shows download timeout. Cause: The timeout threshold for image retrieval is not configured, or the threshold is set too short to cover the image transmission time of workshop intranet equipment.
- The model does not automatically select the inspection tool for the corresponding furnace charge, and always calls a fixed general document tool. Cause:
toolChoiceis not set to auto mode, and the tool configuration for an old batch is fixedly bound. - The DingTalk webhook push content lacks the equipment batch number. Cause: The batch ID field of the MES system is not bound in the tool call parameters, resulting in incomplete pushed data.
How to confirm the configuration is complete
- Upload the furnace loading drawing and furnace temperature curve files for a single furnace charge, trigger tool calling, and check whether the loading position and temperature values in the images can be correctly parsed.
- Simulate an equipment inspection abnormality scenario, trigger tool calling, and check whether the abnormality disposal tool is automatically selected and corresponding disposal suggestions are generated.
- After configuring the DingTalk webhook, trigger a test call, and check whether the pushed message includes the equipment ID, batch number, and abnormality content.
- View the tool call log, confirm that
toolChoiceautomatically switches to the corresponding tool across different furnace charge scenarios.
The values given are common starting points and should be measured against the reader's own samples.