What Data for This Category Looks Like
Data for this category primarily comes from silicon wafer incoming materials, MES systems, operation logs of diffusion furnaces, coating equipment, and printing presses, EL and IV testing equipment, and laboratory test data. Daily generated documents include process cards, diffusion and coating parameter tables, printing and sintering records, EL detection images, IV grading data, and yield reports. Data updates follow production batches, single-chip test results, and daily summary cycles. Data fields include process parameter values, equipment runtime, detection image numbers, electrical performance values, batch identifiers, and some fields include physical units.
Constraints Imposed by These Characteristics on the "Citation Sources and Traceability" Link
Photovoltaic cell manufacturing data has multi-source, high-frequency, and multi-modal characteristics. Process parameters and test results are strongly correlated. Precise matching of batches and processes is required. Traceability for multi-source data requires citation sources to clearly associate equipment, test links, and production batches to avoid cross-batch confusion. High-frequency updated single-chip test data requires fine-grained retrieval. Text descriptions of long documents such as EL detection images must accurately correspond to original files. Parameter differences across different processes require retrieval scopes to be limited to corresponding production links. Traceability logic for numerical and image data must be compatible. Citation sources must correspond one-to-one with actual production links.
Configuration Settings
These configurations are based on the interface logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the paragraph structure of long documents such as process cards and IV grading data, preventing key parameters from being truncated |
| Overlap Length | 100–200 characters | Retains process-related information across chunks, preventing process parameters and batch identifiers from being split and lost |
quoteMaxToken | 1500–2000 token | Accommodates descriptive text for EL detection images and batch traceability information, ensuring complete citation content |
| Single Search Limit | Top 8 entries | Matches multi-source data across multiple processes, covering production records from different equipment and test links |
| Reranking | Enabled | Prioritizes association with batch numbers and process fields, improving the accuracy of traceability results |
| Max Context | 4000–6000 characters | Integrates multiple related documents to fully present the traceability chain of the production process |
Three Common Misconfigurations
- Calling the interface after configuration does not return citation details. The cause is failing to enable the associated configuration for the
responseDataparameter, or failing to correctly bind the knowledge base variable to the retrieval link. - Embedded pages cannot display citation content. The cause is failing to turn off the default citation pop-up switch, or failing to configure allowed citation content rendering in iframe parameters.
- Process associations are lost during retrieval after long document splitting. The cause is using heading-level separators to split parameter tables across processes, causing chunks to lose association information between processes and batches.
How to Verify Correct Configuration
- Upload a complete process card document, initiate a retrieval, and check if the returned results include parameter records and batch identifiers for the corresponding process.
- Call the interface to obtain returned content, check if the
detailfield contains citation source information that matches the metadata of the uploaded document. - Dynamically pass a batch number as a retrieval variable, verify that the returned results only associate production data for the corresponding batch, with no cross-batch irrelevant records.
- Embed an iframe and load the question-and-answer page, confirm that the page can expand to view original document fragments and source information for citations.
The values provided are common starting points and should be measured against the reader's own samples.