What the data for this category looks like
Data for this category comes from material grade archives, customer technical requirement documents, temperature control instrument collection data, furnace temperature uniformity test reports, hardness tester and metallographic equipment test results, and workshop MES systems. Documents including heat treatment process cards, furnace loading layout drawings, furnace temperature curves, hardness and metallographic test reports, equipment inspection forms, and exception handling sheets are generated daily, with an update frequency of per furnace charge or per production batch. Documents contain fields such as material grade, heating, holding and cooling parameters, furnace temperature curve nodes, hardness values, and metallographic structure types. Most parameter units are degrees Celsius, hours, HV or HRC hardness units.
What constraints do these characteristics impose on the "document parsing and chunking" link
The high-frequency batch updates and mixed multi-type document characteristics of this category impose multiple constraints on document parsing and chunking. The number of documents per furnace charge or production batch is large and updated frequently, so systems must adapt to timeout and capacity limits for batch parsing. Documents contain structured parameters, time-series curves, and image content, so chunking rules must be set by distinguishing different content types to avoid splitting that destroys the contextual association of curves or images. Documents from mixed sources include both standardized data automatically collected by equipment and manual exception descriptions, so field associations must be retained during chunking to ensure that complete process exception logic can be matched during subsequent retrieval.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 600–1200 seconds | A single batch includes multiple documents such as furnace temperature curve PDFs and metallographic reports, which take a long time to parse. This setting avoids mid-process interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | A single furnace loading drawing or furnace temperature curve PDF may contain multiple pages of time-series data, so this setting adapts to large-capacity single-file uploads |
chunkSplitter | Switch based on document type | Use fixed character chunking for process cards, time node chunking for furnace temperature curves, and test item chunking for metallographic reports |
chunkSize | 800–1200 characters | Adapts to the information density of process parameters and exception descriptions, avoiding overly fragmented chunks that lose associations or overly long chunks that affect retrieval efficiency |
customPdfParse | Enable structured extraction mode | Automatically extract temperature-time nodes of furnace temperature curves and structured values of hardness tests, replacing information loss from plain text parsing |
chunkSettingMode | Configure by document category | Set differentiated chunking parameters for heat treatment process cards and exception handling sheets respectively to adapt to their respective content structures |
Three common mistakes
- Phenomenon: Batch upload of heat treatment-related documents is interrupted mid-process, and parsing progress cannot be restored after restarting the Docker container. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not set reasonably, and the single-batch parsing duration exceeds the system limit, causing the parsing process to be forcibly terminated without retaining breakpoints. - Phenomenon: After uploading a single large-capacity furnace temperature curve PDF, no corresponding chunked content is generated in the knowledge base. Cause: The structured extraction mode of
customPdfParseis not enabled, and plain text parsing cannot extract the temperature-time node data of the curve, resulting in missing valid chunks. - Phenomenon: After uploading a file via the dialog box, there is no parsing progress feedback on the interface, and no error logs are generated in the background. Cause: The
UPLOAD_FILE_MAX_SIZEparameter is not adjusted, and the uploaded metallographic report image file exceeds the system default limit, and is silently intercepted by the system.
How to confirm the configuration is correct
- Use the FastGPT v4.17.0 system, upload a single furnace temperature curve PDF within 1000 MB, and check whether the parsing progress is completed within the duration set by
PARSE_FILE_TIMEOUT_SECONDS. - After enabling
customPdfParse, parse the heat treatment process card document, and check whether the chunked content includes structured fields such as material grade and heating parameters. - Batch upload 3 or more different types of heat treatment documents, and confirm that all documents generate corresponding chunks without mid-process interruptions.
- Call the corresponding API to test the chunk addition function, and confirm that the incoming document type parameters can be correctly recognized by the system and chunks can be generated.
The values provided here are common starting points and should be evaluated against the reader's own samples.