What the data for this category looks like
Daily documents for this category include foundry process cards, batching lists, heat records, mold history logs, 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, customer drawings, and technical specifications.
Data updates occur per heat, per batch, or daily. Most documents use structured or semi-structured formats. They include fixed fields such as alloy batching ratios, furnace temperature, pouring time, mold cycle counts, and defect types. Some documents include industrial-specific embedded tables and parameter columns.
Constraints on document parsing and chunking
Multi-source heterogeneous industrial data formats require parsing tools to support embedded tables and fixed field separators. Frequently updated batch documents need support for batch parsing and timeout fault tolerance.
Fixed field units and linked data between heat numbers and mold cycles require retaining contextual relevance during chunking. This avoids splitting content that spans related fields. Long documents such as mold history logs and melting logs need appropriate chunk lengths. They must also avoid splitting continuous paragraphs that contain defect descriptions and their corresponding heat numbers.
Configuration Setup
For FastGPT v4.17.0, the following configuration items can be used to adapt document parsing and chunking for foundry scenarios:
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the average length of single-field sections in foundry documents. It avoids splitting continuous paragraphs that contain heat numbers and defect descriptions |
chunkSplitter | Use specified separators + semantic segmentation | Matches the fixed field separator format used in foundry documents, such as "Heat No.:", "Pouring Temperature:". It retains field relevance |
customPdfParse | Enabled | Adapts to structured table data embedded in industrial PDFs. It accurately extracts professional fields such as alloy batching ratios and furnace temperature |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Matches the parsing duration of single large melting logs or mold history logs. It prevents parsing failures due to timeout |
UPLOAD_FILE_MAX_SIZE | 500 MB | Allows uploading batch document archive packages produced daily in workshops. It covers the upper limit of daily document scale |
Three Common Configuration Mistakes
- Scenario: Parsing result is empty or core fields are missing after uploading a foundry process PDF. Cause:
customPdfParseis not enabled, and the default parsing logic does not support industrial embedded table formats. - Scenario: Timeout error occurs during batch parsing of heat records. Cause:
PARSE_FILE_TIMEOUT_SECONDSis set below 300 seconds, preventing complete parsing of long documents. - Scenario: Association information between heat numbers and defects is lost after chunking. Cause:
chunkSizeis set too small, splitting continuous paragraphs that contain heat numbers and defect descriptions.
How to Verify Correct Configuration
- Upload a single heat record PDF, and confirm the parsed result includes complete alloy batching ratios, furnace temperature values and their associated units.
- Upload a batch archive document, confirm no timeout errors occur for parsing tasks, and that
PARSE_FILE_TIMEOUT_SECONDSmatches actual task durations. - Adjust the
chunkSizeparameter, compare chunking results, and confirm no continuous content containing heat numbers and defect descriptions is split. - Verify that
UPLOAD_FILE_MAX_SIZEallows uploading the largest daily document package produced in the workshop, with no upload interception prompts.
The values provided are common starting points and should be measured against the reader's own samples.