What the data for this category looks like
Data for photovoltaic module encapsulation comes from the battery piece grading system, glass adhesive film backplane batch ledger, workshop MES system, string welding and lamination equipment logs, and EL/IV tester output files. Daily retained documents include layout drawings, material batch records, lamination process records, EL crack images, IV test reports, insulation voltage withstand records, and packing lists. Data updates follow this pattern: each component generates a matching test record, each batch of raw materials updates the batch ledger synchronously. Single documents contain fields such as material model, process parameters, and test values. Some EL images are in image format, and must be linked to the batch information of the corresponding component.
What constraints these characteristics impose on the "model access and configuration" link
This category of data includes structured process records and image-based test files. Model access must support both text and image parsing, and adapt to multimodal input. Data update frequency is high. Process and test data for a single batch of components must be indexed within a short time frame. Context length must be limited to avoid interference from redundant information. Some fields must be associated with battery piece grading results and lamination parameters. Model calls must ensure that the context includes complete material and equipment logs for the corresponding batch, to prevent parameter matching errors. The setup must also adapt to the workshop's locally deployed network environment. Local API endpoints and access tokens must be configured.
How to set the configuration
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to complete process and test data for a single component batch, avoids exceeding the model's context limit |
vision | Enabled | Required to parse image documents such as EL crack images and packing list illustrations |
quoteMaxToken | 1500–2000 token | Retains sufficient process parameter details to support parameter matching during Q&A |
temperature | 0.1–0.3 | Reduces generalization deviation, ensures matching accuracy between process parameters and standards |
AIPROXY_API_ENDPOINT | FastGPT proxy address deployed locally in the workshop | Adapts to workshop intranet access restrictions, avoids external network latency |
AIPROXY_API_TOKEN | Exclusive workshop access token | Restricts unauthorized access, protects production data security |
Three common errors
- Phenomenon: The model cannot parse image files such as EL images and packing list illustrations, and returns an error that image content cannot be recognized. Cause: The
visionconfiguration item is not enabled, and multimodal parsing capability is not activated. - Phenomenon: When there is no directly matching process clause in the knowledge base, the model directly refuses to answer, and cannot perform generalized reasoning. Cause:
maxContextandtemperatureare not properly configured, leading to excessive context truncation or parameter deviation, making it impossible to associate similar process parameters. - Phenomenon: The locally deployed version cannot call the specified model, and returns a 403 permission error. Cause:
AIPROXY_API_TOKENis not configured correctly, or a non-exclusive workshop access token is used.
How to confirm the configuration is correct
- Upload a single EL image and the corresponding lamination process record, initiate a process compliance Q&A, and confirm that the model can correctly identify the crack location in the image and associate it with lamination parameters.
- Input a battery piece grading matching question that is not directly included in the knowledge base, and confirm that the model can perform reasonable deduction based on existing batch data.
- Check the model call log to confirm that the configurations of
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare correctly loaded, and there are no access errors. - Adjust the
temperatureparameter and initiate a test, confirm that the rigor of the response changes as expected with the parameter adjustment.
The values provided here are common starting points, and should be evaluated against the user's own samples.