What Data for This Category Looks Like
Typical data for this category comes from raw material quality certificates, heating furnace systems, press collection data, MES systems, and non-destructive testing instruments. Daily documents include forging process specifications, cutting cards, heating furnace logs, forging process cards, mold life records, heat treatment work orders, and ultrasonic testing reports. Data is updated per forging piece, batch, or furnace heat. Most documents contain standardized professional fields such as heat/lot number, forging ratio, heating temperature, and mold usage count. They also include specific parameters like physical dimensions and test values.
Constraints on Document Parsing and Chunking
Data for this category is updated per forging piece, batch, or furnace heat. It also includes standardized professional fields and structured tables. These traits create three constraints for document parsing and chunking. First, retain associated identifiers such as batch and furnace number to avoid incorrect splitting of cross-batch content. Second, accurately identify table structures in ultrasonic testing reports and heating furnace logs. This prevents splitting cross-row test data and corresponding descriptions across chunks. Third, preserve contextual associations for professional parameters like forging ratio and flow direction. This avoids separating parameters from process descriptions. A single document may contain data for multiple furnace heats. Chunk boundaries must be divided based on batch nodes within the document.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 1000–1500 characters | This category’s documents include long-form process specifications and multi-column table data. This range can hold complete single-step process descriptions or single test records, avoiding splitting of critical parameters |
chunkSplitter | Paragraph + table recognition | Adapts to the mixed text and structured table characteristics of documents in this category, ensuring table content is fully retained as a single chunk unit |
customPdfParse | Enable MinerU adaptation mode | PDF documents in this category often include industrial test tables and process flow diagrams. The adaptation mode improves parsing accuracy for complex layouts |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Single large heat treatment work orders or ultrasonic testing reports may contain large amounts of data. This duration covers parsing time for standard documents |
chunkSettingMode | Manual configuration | Allows adjustment of chunking parameters for different document types such as process specifications and test reports. This adapts to the diverse document structures of this category |
Three Common Configuration Mistakes
- Phenomenon: When uploading tabular documents such as ultrasonic testing reports and heating furnace logs, some table content is split into scattered chunks, or table fields cannot be fully extracted. Cause: The
chunkSplitteris not configured to the paragraph + table recognition mode, causing the chunking logic to incorrectly split cross-row table content. - Phenomenon: Timeout errors occur when parsing large heat treatment work orders, or insufficient video memory prompts appear when running in a GPU environment. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted to a reasonable duration, and dedicated resources are not specified for document parsing tasks. - Phenomenon: After importing Word-format forging process documents, embedded detection photo links are lost and cannot be displayed normally in conversations. Cause: The option to retain complete paths for embedded resources during document parsing is not configured, causing links to fail after conversion to Markdown.
How to Confirm Correct Configuration
- Upload an ultrasonic testing report with multi-column tables. Check the parsed chunk content to confirm that all rows and columns of a single table are merged into a single chunk unit.
- Upload a single large heat treatment document. Confirm that no timeout errors occur during parsing. Adjust the
PARSE_FILE_TIMEOUT_SECONDSparameter based on the actual size of the document. - Import a Word-format forging process document that contains embedded images. Confirm that the converted content retains the complete path information of the images.
- View the document parsing logs to confirm that the specified custom PDF parsing mode is enabled, and there are no unrecognized professional fields or typesetting errors.
The values provided here are common starting points. They should be evaluated against the reader’s own samples.