What the data for this category looks like
Core data for photovoltaic module encapsulation comes from the string welding, lamination, framing, and testing departments. Daily documents include layout diagrams, material batch records, lamination process records, EL images, IV test reports, insulation voltage withstand records, and packing lists. Data sources include the wafer grading system, glass adhesive film backplane batch ledgers, MES production systems, string welding and lamination equipment, and EL/IV testers. Data updates follow one test record per module, and one batch ledger per material batch. Core fields include wafer current grade, lamination temperature and time, EL crack location, IV test power value, and other relevant metrics. Some documents include image data and structured tables.
Constraints on document parsing and chunking workflows
The high-frequency update nature of this category requires parsing and chunking workflows to have low latency. This avoids slowing knowledge base synchronization for production processes. Multiple document types (text, tables, images) coexist. Parsing modules must adapt to plain-text PDFs, test reports with embedded images, and batch records with non-standard layouts. Core fields are highly specialized. Chunking must retain contextual associations, and avoid splitting specialized terms and parameter combinations. Batch data processing is required. Parsing workflows must support batch recognition of multiple batch records in a single file. They must also ensure parsing accuracy to prevent deviations in subsequent process judgments caused by chunking errors.
Configuration guidelines
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of process parameter paragraphs and test data blocks in photovoltaic documents. It avoids splitting core professional fields such as wafer current grade and lamination parameters. |
chunkSplitter | Split by paragraph + heading | Retains the hierarchical structure of process documents. It aligns with the heading annotation logic of EL images and IV reports, and avoids splitting core data across headings. |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Addresses parsing time for documents with embedded images such as EL images and IV test reports. It prevents parsing interruptions due to timeouts. |
customPdfParse | Enabled | Adapts to format parsing for non-standard PDFs such as layout diagrams and batch records. It correctly identifies table and image embedded layouts. |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports batch uploads of complete material batch records. It meets the requirement for one-time parsing of multiple documents. |
chunkSettingMode | Manual configuration | Adjusts chunking rules for different document types (text-based process records / image-based test reports) to optimize parsing performance. |
Three common configuration errors
- API returns a successful status but no parsing results:
PARSE_FILE_TIMEOUT_SECONDSis not set correctly. Parsing time exceeds the configured threshold, and complete data is not returned. - Split-related errors appear in logs:
chunkSplitteris configured for fixed character splitting. This does not adapt to the embedded image and cross-page table structure of EL images and IV reports, causing abnormal splitting logic. - Excessively long document parsing time: Single file maximum size is not limited, or
customPdfParseis not enabled. This causes large-volume batch test report parsing to occupy excessive system resources.
How to verify correct configuration
- Upload a typical lamination process record PDF. Check that the parsed text blocks retain complete paragraphs of core fields such as lamination temperature and time.
- Call the FastGPT v4.17.0 file parsing API. Verify that the returned results contain complete text and metadata for the corresponding document, with no empty fields or truncated content.
- View the parsing module's running logs. Confirm there are no split-related errors, and that parsing time meets the configured
PARSE_FILE_TIMEOUT_SECONDSthreshold. - Upload different types of documents (layout diagrams, EL images, batch records). Confirm that with
customPdfParseenabled, the system correctly identifies the layout structure of non-standard PDFs.
The values provided here are common starting points, and should be evaluated against the user's own samples.