What this category of data looks like
This category of data originates from silicon wafer incoming material records, production MES systems, real-time collected data from diffusion furnaces, coating equipment, and printing presses, EL and IV test results, and laboratory process verification records. Update cycles cover each production batch, single-wafer test results, and daily summaries. Document formats include structured process cards, diffusion and coating parameter tables, printing and sintering operation records, EL defect detection images, IV performance grading data, and yield summary reports. Most fields use dedicated process units, and data granularity ranges from batch level to single-wafer level.
What constraints do these characteristics impose on the citation source and traceability link
The fine-grained multi-source data and dedicated process fields of this category require accurate association of the traceability link with specific production batch or single-wafer test records, to prevent traceability deviations caused by vague matching. The mixed format of structured parameter tables and unstructured test reports requires traceability to support associating source information for both text parameters and image-based test results. Frequently updated production data requires the knowledge base synchronization mechanism to match the node rhythm of the production process, to avoid referencing outdated process parameters. Fields with dedicated process units require the recall link to verify parameter unit consistency, ensuring traced content matches actual production data.
How to set the configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Photovoltaic process documents are parameter-dense, so sufficient traceability context must be retained while avoiding exceeding model context limits |
responseData | Enabled | Citation source document metadata and paragraph location must be output to meet the traceability verification needs of process personnel |
detail | Highest level | Process parameters and test results must fully display source fields to facilitate matching with actual production data |
| Single recall limit | Top 5 entries | A single batch has many process parameter entries; limiting the number of recalled entries avoids redundancy while covering core process links such as texturing, diffusion, and passivation |
| Knowledge base synchronization cycle | Every 8 hours | Photovoltaic production data updates with production shifts; synchronization frequency matches production node rhythm |
Three common configuration mistakes
- Phenomenon: Output content does not display the document name and paragraph location of the citation source. Cause: The
responseDataconfiguration item is not enabled, or the knowledge base metadata collection rules are not configured. - Phenomenon: Core process parameters are not recalled after multiple knowledge bases are associated. Cause: Recall matching rules are not adjusted for the dedicated fields of photovoltaic process documents, resulting in content with mismatched parameter units being filtered out.
- Phenomenon: Structured parameter tables uploaded to the knowledge base cannot be correctly traced to specific production batches. Cause: Complete metadata display for the
detailconfiguration is not enabled, or the batch number is not used as a document index field.
How to verify correct configuration
- A query containing specific process parameters can be submitted, and the output content can be checked to confirm whether it includes the document name and paragraph location of the citation source.
- The knowledge base management interface can be accessed to confirm that the synchronization cycle setting matches the update rhythm of production shifts.
- A mixed document of a photovoltaic process card and an EL test report can be uploaded, a query can be submitted, and the metadata of the corresponding document can be checked for association.
- The
quoteMaxTokenparameter can be adjusted, and the context length of the output content can be verified to meet the parameter display requirements of process documents.
The values provided are common starting points and should be measured against the reader's own samples.