What the Data for This Category Looks Like
This category’s data typically originates from wafer incoming material ledgers, production MES systems, operation logs of diffusion furnaces, coating equipment, and printers, output from EL and IV test equipment, and laboratory quality inspection data. Data update rhythms cover each production batch, each finished individual product inspection, and daily summary links. Daily documents include process cards, diffusion and coating parameter configuration sheets, printing and sintering process records, EL defect detection images, IV performance grading data, and batch yield reports. Core fields include conversion efficiency values, crack levels, color difference parameters, sintering temperature ranges, and more. Units are mostly percentages, watts, and pixel-level defect identifiers.
What Constraints These Characteristics Impose on the "Tool Calling and Plugins" Link
The high-frequency data characteristics at the wafer and batch level require tool calling to support targeted pulling of multi-source heterogeneous data. This includes structured process parameters and unstructured EL detection images and IV test reports. The high-frequency update rhythm requires tool trigger logic to match production nodes, to avoid pulling expired uncompleted process data. The multi-dimensional core field requirements require plugin configuration to support custom field mapping. This maps raw data returned by tools to exclusive indicators for photovoltaic cell manufacturing. At the same time, the continuous process and microstructure-sensitive characteristics require tool calling to associate upstream and downstream process data, to avoid deviations caused by isolated analysis.
How to Set the Configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | Auto | Adapt to scenarios with parallel multi-tool calling, match the multi-source data needs of multi-process audits for photovoltaic manufacturing |
| Workflow/API function-call configuration | Enabled | Support structured return of process parameters and inspection data, facilitate subsequent quality analysis and traceability |
| Tool Trigger Timing | After production batch completion | Match the update rhythm of batch-level aggregated data for this category, avoid pulling uncompleted process data |
| Field Mapping Rules | Associate conversion efficiency, crack level, yield fields | Correspond to the core business audit indicators of this category, ensure returned data fits the business scenario |
| Single Call Data Volume | 100–500 records | Adapt to batch-level data batch pulling, avoid interface overload or data parsing timeout |
| Timeout Threshold | 300–600 seconds | Match the generation cycle of device logs and EL/IV inspection reports, adapt to long-duration unstructured data parsing |
Three Common Misconfigurations
- Interface calls return a 514 status code. The cause is that data permissions for the photovoltaic cell manufacturing scenario are not configured, so the tool cannot pull EL detection images or IV grading data in a targeted manner.
- No alternative API call entry for the networked plugin can be found. The cause is that the custom API configuration item is not enabled, and only the default model call interface is bound.
- Variables passed via API in the workflow cannot trigger the knowledge base search node. The cause is that the format of the passed variables does not match the batch number or wafer ID fields supported by the knowledge base.
How to Verify the Configuration is Correct
- Initiate an API call simulating a production batch, check if the returned data includes core business fields such as conversion efficiency and crack level.
- View the tool call log, confirm that the
toolChoiceparameter takes effect, and the required multi-tool calling logic is automatically matched. - Test the variable reference function, pass a preset wafer ID, check if the knowledge base search node returns the process card data of the corresponding batch.
- Simulate a long-duration EL detection image parsing request, verify that the timeout threshold configuration takes effect, and no timeout error is triggered.
The values provided are common starting points and should be measured against the reader's own samples.