What the Data for This Category Looks Like
Business data for photovoltaic cell manufacturing comes primarily from the process, equipment, quality control, and production departments of crystalline silicon cell enterprises. Daily documents include process cards, diffusion and coating parameters, printing and sintering records, EL detection images, IV grading data, and yield reports.
Data sources cover silicon wafer incoming materials, MES systems, diffusion furnaces, coating equipment, printing presses, EL and IV testing equipment, and laboratory data. Update frequencies include per production batch, per piece detection, and daily synchronization.
Documents include structured parameters, unstructured detection images, and statistical reports. Fields cover specific business dimensions such as process window values and electrical performance indicators.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Business data characteristics of photovoltaic cell manufacturing impose multiple constraints on knowledge base retrieval and recall. First, high-frequency real-time updated data requires the retrieval system to support incremental synchronization and low-latency recall, to avoid affecting the accuracy of tender quotations due to data lag. Second, the mixed scenario of multiple document types requires retrieval to cover both precise matching of structured parameters and semantic association of unstructured images. A single retrieval mode cannot meet full-scenario requirements. Third, business fields have high professionality, requiring the retrieval system to recognize exclusive terms such as process windows and electrical performance grading, to avoid recalling irrelevant content. Fourth, tender scenarios require association of multi-link data. Recalled content must cover parameters of multiple processes including texturing, diffusion, and passivation, to ensure the completeness of tender documents.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.82 | Photovoltaic cell process parameters have high precision requirements, and low-correlation recalled content must be filtered |
limit | Top 8 entries | Tender documents need to cover parameters of multiple process links, and excessive redundant content will interfere with large model output |
searchMode | mixedRecall | Balances precise matching of structured parameters and semantic association needs of business scenarios |
usingReRank | Enabled | Unstructured content such as EL detection images and IV grading data requires reranking to improve relevance |
datasetSearchUsingExtensionQuery | Enabled | Tender scenarios require extended professional term association, for example, associating "conversion efficiency" with "photoelectric conversion efficiency" |
collectionIds | Bind datasets of process cards, equipment logs, and yield reports | Covers core business document types required for tenders |
Three Common Misconfigurations
- Calling the knowledge base interface prompts unauthorized access. The cause is failure to bind the dataset ID corresponding to the photovoltaic manufacturing business, or the API key permission scope does not cover the current knowledge base.
- After adjusting the
limitparameter to a larger value, the large model does not reference any recalled content. The cause is that thesimilaritythreshold is set too low, and the relevance of the recalled content is insufficient, so the large model cannot identify valid references. - Photovoltaic industry documents within WeChat Official Accounts cannot be loaded. The cause is failure to configure the external data source synchronization plugin, or failure to complete crawling authorization for Official Account content.
How to Verify Successful Configuration
- Check whether the datasets bound to the knowledge base under FastGPT v4.17.0 include core documents such as process cards, equipment logs, and IV grading data.
- Test retrieval of specific process parameters, such as diffusion furnace temperature parameters, to confirm that recalled results contain accurate content for the corresponding fields.
- Submit test questions for the tender scenario to verify whether the relevance of retrieval results meets business requirements.
- After adjusting configuration parameters, confirm whether the large model output correctly references recalled content.
The values provided here are common starting points and should be measured against the reader's own samples.