What the data for this category looks like
Data for the photovoltaic module encapsulation process comes from cell grading records, glass adhesive film backplane batch ledgers, MES systems, string welding and lamination equipment, and EL/IV testers. Corresponding documents are generated for each module and each material batch upon data update. Documents include layout drawings, material batch records, lamination process records, EL crack images, IV test reports, insulation voltage resistance records, and packing lists. Document fields include cell current grade, lamination temperature and duration, EL crack location, IV power value, insulation voltage resistance value, and others, with corresponding units of ampere, degree Celsius, second, watt, megaohm, and additional standard units.
What constraints these characteristics impose on the model access and configuration link
Since data is updated frequently per individual module and batch, model access must support real-time synchronization of multi-source heterogeneous data. Documents include textual process records, numerical test reports, and visual EL images, so multi-modal data processing must be supported. Fields are closely correlated, for example cell grade directly affects IV power, so accuracy of context matching must be ensured. Data formats output by different devices vary, so compatible access rules must be configured to avoid data parsing errors.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Matches the total length of multiple documents associated with a single module, to avoid context overflow |
vision | Enabled | Supports parsing visual documents such as EL crack images and IV test report charts |
temperature | 0.1–0.3 | Ensures consistency of inspection results, reduces randomness in judgments of cracks and power grades |
| Workflow/API tool selection | auto | Automatically determines whether to call tools to process raw device data, adapts to multi-source data access requirements |
AIPROXY_API_ENDPOINT | Fill in according to actual deployment address | Connects to third-party proxy services, adapts to cross-region data transmission scenarios |
quoteMaxToken | 1500 token | Limits the length of a single cited document, focuses on core inspection-related content |
Three common misconfigurations
- Phenomenon: A context overflow error is returned when calling the model, with a status code of 413. Cause: The
maxContextparameter was not adjusted based on the total length of documents associated with the module, exceeding the maximum context length supported by the model. - Phenomenon: The application can only retrieve knowledge base content, and cannot call online tools to supplement the latest process parameters. Cause: Tool call permissions were not configured correctly, or the access switch for the corresponding tool was not enabled in the application settings.
- Phenomenon: EL crack images cannot be parsed correctly, and visual-related content is not mentioned in the output results. Cause: The
visionparameter was not enabled, or the access key for the visual model was not configured.
How to confirm the configuration is successful
- Upload the EL image and IV test report of a single photovoltaic module, verify that the model can correctly identify the crack location and match the corresponding current grade and power value.
- Check the application logs of FastGPT v4.17.0, confirm that the
maxContextparameter is in effect, and no context overflow errors occur. - Test calling online tools, verify that the latest lamination process parameters can be obtained and integrated into the inspection results.
- Set
temperatureto 0.1, input the same inspection data multiple times, confirm that the output results remain consistent.
The values provided are common starting points and should be measured against the reader's own samples.