This document is based on the configuration instructions for FastGPT v4.17.0.
What Data Looks Like for This Category
Data sources for this category include product drawings and 3D models, CAPP/CAM systems, machine tool collection data, tool management systems, MES systems, and inspection tool inspection records. Daily documents cover process routes, operation cards, tool lists, CNC programs, first-piece inspection records, patrol inspection forms, and measuring tool calibration records. Data updates trigger with work orders, workpieces, sampling batches, or tool changes. Document fields include operation number, tool model, cutting parameters, inspection dimensional deviation, calibration cycle, and more. Common units are millimeters, revolutions per minute, meters per minute, and other general machining units.
Constraints Imposed on Document Parsing and Chunking
The multi-type document structure and high-frequency update characteristics of this category create multiple constraints for document parsing and chunking. Split linear documents such as process routes and CNC programs by operation or code segment to avoid cross-operation logical breaks. Retain column field associations for tabular documents such as multi-column patrol inspection forms and tool lists to prevent loss of column correspondence after splitting. Use moderate chunk sizes for high-frequency updated data. An overly large chunk increases retrieval matching redundancy. Tie processing data strongly to workpieces and work orders. Retain context associations during chunking to ensure complete operation or inspection link information can be matched during retrieval.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the content length of single-operation machining documents, retains complete associations between operation parameters and inspection results, and avoids losing context after splitting |
chunkSplitter | Split by paragraph + table block | Adapts to multi-column tables such as patrol inspection forms and tool lists, as well as process description documents, and separately retains table column correspondence and paragraph logic |
customPdfParse | Enable structured parsing | Adapts to process drawing PDFs with vector annotations, extracts key structured information such as dimensional tolerances and operation numbers |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Adapts to uploaded files containing large CNC programs or multi-page process manuals, avoids parsing interruptions due to timeout |
UPLOAD_FILE_MAX_SIZE | 500 MB | Adapts to machining document packages containing 3D model attachments or multi-operation collections, meets upload requirements for large work order documents |
chunkSettingMode | Manual configuration | Adjusts chunking rules separately for different document types such as CNC programs and patrol inspection forms, adapts to the diverse document structures of machining scenarios |
Three Common Configuration Mistakes
- After uploading a multi-column patrol inspection form Excel, retrieval results lack column correspondence information. The root cause is that the chunking rule for table block splitting is not enabled, causing the Excel to be split by rows or fixed character counts, and column field associations are lost.
- When enabling
customPdfParse, the mineru API key is not configured. This results in PDF parsing failure, with the interface displaying abnormal parsing status or returning empty content. - After configuring
chunkSizeto an excessively large value, the conditional judgment component encounters null values when processing parsed boolean tool wear parameters. The oversized chunk causes parameter truncation, making the parameter unreadable by the component.
How to Verify Correct Configuration
- Upload a single-page operation card PDF, check the parsed chunk content, and confirm each chunk contains complete operation parameters and inspection results, with no cross-operation content spliced together.
- Upload a multi-column patrol inspection form Excel, and verify during retrieval that a single field (such as tool number) can match corresponding inspection sizes and deviation values, confirming column associations are not lost.
- Upload a 100 MB CNC program package, check the parsing progress, and confirm no timeout errors occur. After parsing completes, code segment content can be retrieved normally.
- Configure
customPdfParseand bind the mineru API, upload a process drawing PDF with dimension annotations, and check if the parsing result includes structured information such as tolerance values and operation numbers.
The values given are common starting points and should be measured against the reader's own samples.