What this category of data looks like
Daily documents for this category include foundry process cards, batching lists, heat records, mold resumes, first-piece inspection records, material test reports, and nonconforming product review forms. Data sources cover ERP material master data, melting furnaces and thermometers, spectrometers, MES heat records, mold management systems, and customer drawings. Update frequency is per heat, per batch, or daily. Documents include fields such as alloy batching ratios, melting temperatures, pouring parameters, defect types and corresponding heat and mold numbers. Units include weight percentage, degrees Celsius, millimeters, and similar units.
What Constraints These Characteristics Impose on Document Parsing and Chunking
Documents such as heat records and batching lists, updated per heat or per batch, are updated frequently. This requires chunking tools to support high-frequency incremental parsing. Multiple document types contain both structured fields and unstructured descriptions. Chunking must preserve field association to avoid breaking the logic linking batching ratios and defect traceability. Documents including mold resumes and material test reports have fixed-format headers and detail rows. Structured parsing is needed to accurately extract key fields. Data formats vary widely across sources. Flexible chunking rules are required to adapt to different document types, while ensuring that associated information between defects, heats, and mold numbers is not separated.
How to Set the Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field length and association logic of foundry documents, avoiding splitting that breaks complete information for batching ratios and defect traceability |
chunkSplitter | Semantic chunking | Adapts to documents with mixed structured and unstructured content such as process cards and heat records, preserving content relevance |
customPdfParse | Enable structured parsing | Adapts to fixed-format documents such as mold resumes and material test reports, accurately extracting header and detail fields |
UPLOAD_FILE_MAX_SIZE | 100 MB | Covers the size limit for batch-uploaded documents such as single-batch heat records and mold resumes |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing time required for large log files exported from melting data and spectrometers |
chunkSettingMode | Manual configuration | Adjust chunking rules for different document types such as batching lists and defect review forms to meet business requirements |
Three Common Mistakes
- Uploading audio or video files causes parsing failure with an "unsupported file format" prompt. The cause is missing configuration of corresponding multimedia parsing rules and disabled parsing plugins for non-document files.
- Calling the workflow API to upload images returns a parameter error, and the image field cannot be recognized. The cause is failure to pass image binary data or file paths in the parameter format required by the platform.
- Chunking results lose association information between heats and defects. The cause is that
chunkSizeis set too small, splitting the same associated record across multiple chunks and breaking business logic relevance.
How to Confirm Configuration Is Properly Set
This verification process applies to FastGPT v4.17.0.
- Upload a single foundry batching list, and check whether chunking results retain associated content between alloy batching ratios and heat numbers.
- Upload a batch of heat record files, and verify whether parsing time matches the preset
PARSE_FILE_TIMEOUT_SECONDSsetting. - Call the API to upload test documents, and check whether returned chunked data contains complete field information with no missing or disconnected content.
- Check chunked data stored in the knowledge base, and confirm that key fields of structured documents are not omitted.
The values provided are common starting points and should be measured against the reader's own samples.