This page is compatible with FastGPT v4.17.0.
What This Category of Data Entails
Production data for photovoltaic solar cell manufacturing typically comes from wafer incoming material records, workshop MES systems, operation logs from diffusion furnaces, coating equipment, and printing presses, outputs from EL and IV testing equipment, and laboratory test data. Data updates align with per-batch production cycles, individual wafer test results, and daily summaries. Daily documentation includes process cards, diffusion and coating parameter tables, printing and sintering records, EL defect detection images, IV grading data, and yield-related reports. Fields include batch identifiers, equipment operation parameters, test point coordinates, electrical performance test values, and timestamps.
Constraints for Multi-turn Dialogue and Prompt Engineering
This category’s data includes structured operation parameters, equipment logs, test charts, and grading data. It has high update frequencies and fine batch granularity, which create multiple constraints for multi-turn dialogue and prompt configuration.
First, clearly distinguish ledger data across different batches and equipment. Prompts must specify rules for associating batch identifiers and test types to avoid cross-batch data confusion.
Second, for fine-grained individual wafer test data, multi-turn dialogue must support gradual user guidance to add query dimensions. For example, drill down from a batch overview to EL defect information for a specific test point.
Third, integrate data from dispersed sources incrementally during dialogue. Prompts must preset extraction logic for different data sources to ensure summary results accurately align with production links.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chatId | Bind to production batches, use separate conversations per batch | Production batch data for photovoltaic solar cells is independent across batches, preventing cross-batch context confusion and ensuring accurate queries |
messages | Retain the first 3 rounds of production ledger-related context per conversation | Balances context completeness and information overload risk, covering necessary historical query logic |
variables | Preset three variables: batch number, equipment type, and test type | Quickly locate core dimensions of corresponding production ledgers, reducing the number of multi-round follow-up questions |
| Single search limit | Top 10 matching records | The core data volume of a single batch of photovoltaic production ledgers is moderate; too many results distract from analytical focus |
| Query optimization | Enable, optimize queries for production terminology | Workshop process terminology differs from standard retrieval terminology; optimization improves retrieval matching accuracy |
| Maximum context length | 800–1200 characters | Adapts to the core information scale of a single batch of photovoltaic production ledgers, avoiding errors caused by context overflow |
Three Common Misconfigurations
- Phenomenon: Dialogue returns ledger data including EL detection images and IV data from non-target batches. Cause:
chatIdis not bound to production batches, leading to incorrect context association and cross-batch data being included. - Phenomenon: When running long-text ledger summary tasks, the interface shows the task completed normally but throws the
The value of "offset" is out of rangeerror. Cause: Context length is not limited; long-text data exceeds the system's processing threshold, causing abnormal offset calculation. - Phenomenon: After uploading multiple process cards and test reports, the dialogue cannot accurately associate parameters matching the target batch. Cause: Preset batch dimension
variablesare not configured, so the system cannot match multiple files to their corresponding production batches.
How to Verify Correct Configuration
- Manually enter the ID of a known production batch, and confirm that the ledger data returned by the dialogue only includes parameters and test results for that batch.
- Upload multiple documents corresponding to different batches, select the target batch via
variables, and confirm that the dialogue only calls associated data for that batch. - Run a summary task involving multiple documents, check for offset-related errors, and adjust the context length based on actual conditions.
- After enabling query optimization, enter commonly used workshop process terminology, and confirm that retrieval results match parameters for the corresponding production link.
All recommended configuration values are common starting points. Adjust them based on your own specific samples and use cases.