What this category of data looks like
The primary entities for this category are foundries, die casting workshops, and casting process and quality departments of equipment parts suppliers. Daily generated documents include casting process cards, batching lists, heat records, mold histories, first-piece inspection records, material reports, and nonconforming product review forms. Data sources include ERP material master data, melting furnaces and thermometers, spectrometers, MES heat records, mold management systems, customer drawings and technical requirements. Update frequencies are per heat, per batch, or daily. Documents contain structured process parameters and quality inspection fields, covering alloy ratios, melting temperatures, pouring parameters, defect types, and related content.
Constraints on Document Parsing and Chunking
Diverse document types include large volumes of structured tables. Parsing tools must accurately identify table structures and preserve field relationships. Frequently updated documents, such as heat records generated per heat, require chunking rules adapted to rapidly changing content. This prevents critical process data from being split by improper chunking. Documents from different sources have encoding differences. Some older archived documents use non-standard encodings, making parsing fault tolerance more difficult. Individual document lengths vary widely, ranging from single-page batching lists to multi-page mold histories. Chunking configurations must handle both short and long document processing needs.
How to Configure Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field length of process parameters and defect records in foundry documents, avoiding splitting critical parameter combinations |
chunkSplitter | Split by table + paragraph | Adapts to the multi-row table structure of batching lists and heat records, preserving data relationships |
customPdfParse | Enabled | Adapts to complex layouts of casting process cards and material reports, extracting structured fields |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Covers parsing time for multi-page documents such as melting records and mold histories |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports batch upload of multiple archived documents per heat |
chunkSettingMode | Manual configuration | Adjust chunking rules for different document types (such as XLSX tables, PDF process cards) |
Three Common Configuration Errors
- Phenomenon: After uploading an XLSX batching list, table fields are empty or spliced chaotically. Cause: No table-based chunking rule is configured. The default chunking splits the row and column relationships of the table.
- Phenomenon: Parsing a PDF furnace temperature record throws the error "the argument ‘windows-1252’ is invalid encoding". Cause: No encoding format adapted to foundry documents is specified. The default encoding cannot recognize special process symbols in the document.
- Phenomenon: Parsing times out when batch uploading multiple heat records. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted. The default timeout duration is insufficient for processing multi-page archived documents.
How to Confirm Correct Configuration
- Upload a standard foundry batching list XLSX. Check that parsed chunks retain the complete parameter combination for each row of batching data.
- Upload a PDF furnace temperature record with special characters. Confirm there are no encoding errors and all content is fully extracted.
- Upload a mold history document exceeding the default upload size. Confirm the upload succeeds and parsing completes.
- Adjust the
chunkSizeparameter. Compare chunking results across different chunk lengths to confirm alignment with business field splitting logic.
The values provided here are common starting points and should be evaluated against the reader's own samples.