What this category of data looks like
Teams in the process, equipment, quality, and production departments of crystalline silicon cell manufacturers generate most data related to inspection and process records. Daily documents include process cards, diffusion and coating parameters, printing and sintering records, EL inspection images, IV grading data, and yield reports. Data sources include silicon wafer incoming materials, MES systems, diffusion furnaces, coating equipment, printing presses, EL and IV testing equipment, and laboratory data. The system updates data synchronously for each batch and each inspected wafer. Some daily aggregated yield data updates once per day. Data fields include process parameter values, defect identifiers, electrical performance indicators, batch and wafer-level IDs, and other related fields.
What constraints these characteristics impose on tool calling and plugins
Multi-source heterogeneous data types and update rhythms create clear constraints for tool calling and plugin configuration. First, data comes in multiple formats: structured parameters, images, test reports, and more. Tool plugins that support multiple data source types are required. Second, data updates in real time for each wafer or batch. Tool calling must support high-frequency pulls of the latest data, while cache must be properly configured to avoid duplicate requests. Third, production traceability requires precise querying by specified ID, so tool calling must support tracing data by wafer or batch. Finally, real-time verification of process windows and defect recognition requires tool calling response times to match workshop production rhythms.
How to set the configurations
These configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | Auto | Supports parallel calling of multiple tools, covering three types of tools: process parameter query, EL image inspection, and IV data verification |
| Single tool call timeout | 30 seconds | Matches the average response time of photovoltaic production equipment data pulling, avoids blocking production processes |
| Workflow/API function-call configuration | Mandatory | Ensures the large model outputs in a preset structured format, adapting to the fielded query requirements of photovoltaic data |
| Tool call retry count | 2 times | Balances call success rate and process time, adapting to scenarios of local workshop network fluctuations |
| Tool return result cache duration | 15 minutes | Matches the production interval cycle of each batch of cells, avoids repeatedly pulling historical data of the same batch |
Three Common Mistakes
- The interface displays a failed connection to the tool service, and the log returns a connection timeout. The cause is that data source authentication rules adapted for photovoltaic production data have not been configured, preventing the tool from pulling real-time data from the MES or equipment end during calls.
- The large model output includes the phrase "Citation Marker: [1]". The cause is that the output configuration for knowledge base citation markers has not been disabled, causing the context returned by tool calls to be directly spliced into the final result.
- No non-tool session link can be established after a tool call terminates. The cause is that the context retention setting after tool calls has not been enabled in the configuration, causing the session context to be cleared early.
How to confirm the configuration is correct
- Initiate a query with a specified batch ID or wafer-level ID, then confirm the tool pulls corresponding process cards, EL inspection images, IV grading data, and other content.
- Review the tool call log, then confirm the
toolChoiceandfunctionCallparameters take effect as configured, with no abnormal error messages. - Simulate a local workshop network fluctuation scenario, then verify the tool call retry mechanism triggers normally and resumes connection after network recovery.
- Generate the output of the test query, then confirm no extra citation marker phrases appear in the content.
The values provided here are common starting points and should be evaluated against the reader's own samples.