What the Data for This Category Looks Like
Precast concrete component daily documents include component detail drawings, steel bar cutting lists, mold inspection records, pouring and vibration records, curing records, demolding and factory inspection records. Data sources include BIM or detailed design files, steel bar and concrete batches, mold tables and curing kilns, laboratory records, and QR code traceability. Update frequencies follow per individual component, per batch, and per kiln run. Document fields include component number, steel bar specification, pouring time, curing temperature, demolding strength, and others. Units include millimeters, cubic meters, degrees Celsius, megapascals, and others.
What Constraints Do These Characteristics Impose on the Document Parsing and Chunking Link?
The document structures for this category vary significantly. The field density and page span of large BIM-exported drawings differ greatly from single-page record sheets. Per the high-frequency update requirements for individual components, batches, and kiln runs, associated information between batches and components must be retained, and cross-batch associated content must not be split during chunking. Scanned on-site record documents have watermarks and low-contrast backgrounds, which can interfere with general OCR parsing. Documents associated with QR code traceability require complete context of traceability fields to be retained; splitting such content would break the association.
Configuration Settings
This document applies to FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field density of precast component documents, avoids splitting associated entries in steel bar cutting lists, and ensures semantic completeness of individual chunks |
chunkSplitter | Split by paragraph + header and footer | Precast component inspection records have unified header identifiers; splitting by this rule preserves the complete content of a single inspection record |
customPdfParse | enabled | Adapts to the low-contrast backgrounds and watermarks on scanned component record sheets; custom OCR recognition areas improve parsing accuracy |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Matches the parsing time required for large BIM-exported PDF drawings, prevents parsing failures due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the size of large BIM drawings for precast components, prevents upload failures due to oversized files |
chunkSettingMode | Switch by document type | Flexibly adjusts chunking rules for different document types such as detail drawings and record sheets, adapting to their respective structural characteristics |
Three Common Configuration Errors
- Symptom: After uploading a scanned PDF of precast component curing records, the parsing result contains no valid text content. Cause: The custom OCR configuration for
customPdfParsewas not enabled, and the low-contrast background and watermarks on the scanned document were not accommodated. - Symptom: When parsing a steel bar cutting list Excel file, only the content of the first sheet is read, and the remaining sheets are not recognized. Cause:
chunkSplitterwas not configured to split documents by sheet, resulting in multi-sheet content being merged into a single chunk. - Symptom: When uploading a PDF containing mold inspection photos, the images are parsed into meaningless text by OCR. Cause: The default mixed-text-and-graphics OCR switch was not disabled, forcing image content to be converted to text instead of preserving the original image.
How to Verify Proper Configuration
- Upload a single scanned PDF of a precast component demolding inspection record, and confirm whether the parsing result splits chunks by individual record, with no cross-record content concatenation.
- Upload an Excel steel bar cutting list with multiple sheets, and verify that the parsing result includes content from all sheets, with each sheet corresponding to an independent chunk.
- Upload a PDF containing mold inspection photos, and confirm whether the parsing result preserves the original image links, with no garbled text generated solely by OCR.
- Upload a PDF drawing with more than 20 pages exported from a large BIM project, and confirm that the parsing process does not produce timeout errors, and that the elapsed time matches the value set for
PARSE_FILE_TIMEOUT_SECONDS.
The values provided here are common starting points, and should be measured against the reader's own samples.