This page is written for FastGPT v4.17.0.
What data looks like for this category
The inspection data for photovoltaic cell manufacturing comes from silicon wafer incoming material ledgers, workshop MES systems, real-time collection data from diffusion furnaces, coating equipment, and printing presses, outputs from EL and IV testing equipment, and laboratory quality inspection records. Data updates occur per batch, per wafer inspection node, and as daily summaries. Documents include process cards, diffusion and coating parameter tables, printing and sintering operation records, EL defect detection images, IV grading datasets, and yield statistics reports. Fields include batch numbers, equipment operating parameters, defect level identifiers, electrical performance test values, and grading results.
What constraints these characteristics impose on workflow orchestration
Multi-source heterogeneous data formats and update frequencies require workflows to adapt to access nodes for different data sources and unify data parsing standards. High-frequency inspection requirements for each wafer and each batch require workflow trigger conditions to align with production flow rhythms, avoiding delayed or duplicate execution. Unstructured EL detection images and IV grading data require dedicated parsing nodes to ensure data can be used by subsequent verification nodes. The core requirement of batch traceability requires binding global parameters to ensure consistent data association across nodes, avoiding cross-batch data confusion. Processing of daily aggregated bulk data requires setting reasonable concurrency limits to avoid resource overload.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Trigger Condition | Trigger after the batch's inspection nodes are completed | Match the production rhythm of photovoltaic cell batch flow |
| Global Variable Binding | Bind batch number and equipment ID from variables | Ensure consistent data association across nodes in the workflow, meeting traceability requirements |
| Node ID Binding | Bind nodeId to the current batch node | Ensure uniqueness of node calls in the workflow, avoiding cross-batch conflicts |
| Unstructured Data Parsing | Enable the EL image parsing plugin, configure single-image processing duration to no more than 120 seconds | Adapt to the parsing complexity of EL detection images, avoiding timeouts |
| Knowledge Base Recall Limit | Recall the top 6 matching documents | Cover core reference documents such as process cards and parameter tables, avoiding redundancy |
| Node Timeout Setting | Set single-node timeout to 300 seconds | Adapt to the average time required for equipment data collection and test result uploads |
| Context Window | Retain the last 10 rounds of interaction context | Meet the multi-round judgment requirements for process parameter verification |
Three Common Mistakes
- Phenomenon: After a workflow node executes, logs show empty return values, and downstream nodes cannot obtain equipment parameter results from MCP tool calls. Cause: No binding mapping for tool return values was configured in
variables, causing nodes to fail to capture output data. - Phenomenon: In the output of the AI conversation node, context and reply content are mixed in the same field and cannot be split for use. Cause: Port mapping between context storage and reply output was not configured correctly, causing downstream nodes to fail to obtain correct verification basis.
- Phenomenon: Knowledge base recall results include more than 6 documents, or include process parameter content from non-current batches. Cause: No batch number binding filter condition was configured in the recall settings, leading to redundant recall results that do not match the current inspection scenario.
How to confirm the configuration is complete
- Trigger a simulated batch inspection process, check whether the batch number in
variablesis correctly passed across all nodes. - Call the MCP tool to obtain simulated equipment parameters, verify whether workflow nodes can normally capture and pass return values.
- Configure a knowledge base recall test, confirm that only process documents related to the current batch are recalled, with no irrelevant content.
- Run the full workflow, check whether the AI node's context storage and reply output generate correct content separately.
The values provided are common starting points and should be measured against the reader's own samples.