What this type of data looks like
Daily documents in heat treatment scenarios include heat treatment process cards, furnace loading diagrams, furnace temperature curves, hardness and metallography reports, equipment inspection forms, and exception handling forms. Data sources cover material grades, customer technical requirements, temperature control instruments, furnace temperature uniformity testing equipment, hardness testers and metallography equipment, and manufacturing execution systems. The update frequency is per furnace run and per batch. Each individual document has clear fields and units. For example, heating temperature on process cards uses degrees Celsius as the unit, holding time uses hours or minutes, and hardness reports specify Rockwell or Vickers hardness values.
Constraints for document parsing and chunking
The high-frequency updates and multi-batch nature of documents require support for batch parallel parsing to prevent processing delays for single furnace run documents. Mixed-format documents contain both text and charts. The parsing process must handle both plain text extraction and recognition of annotation information from charts. Key parameters in documents are tightly bound to associated descriptions. For example, heating temperature and its corresponding holding time cannot be split. The chunking logic must retain contextual relevance. Exception handling forms have a strong link between problem descriptions and rectification measures. After chunking, complete problem and solution content must be available during review.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Heat treatment documents contain key values such as temperature and time. The chunk length covers a single process parameter and its associated description to avoid splitting critical linked information |
chunkSplitter | Split by paragraph + heading | Heat treatment documents have clear process chapters and report chapters. Splitting by paragraph preserves the integrity of individual review logic |
customPdfParse | Enabled | Heat treatment documents include furnace temperature curve charts and metallography images. Enabling custom parsing extracts text annotations and numerical values attached to charts |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | A single furnace temperature curve PDF may contain multiple sets of time-temperature data. Parsing takes a long time; this avoids timeout truncation |
UPLOAD_FILE_MAX_SIZE | 1000 MB | When uploading multiple furnace run documents in batches, support for large single-batch file submissions is required |
Three Common Configuration Mistakes
- Phenomenon: Parsed document content is missing numerical annotations for furnace temperature curves. Cause: The
customPdfParseconfiguration is not enabled. Default parsing only extracts plain text and ignores key parameters attached to charts. - Phenomenon: Chunking results split heating temperature and its corresponding holding time into different chunks. Cause: The
chunkSizesetting is too small, and does not cover the complete associated description of a single process parameter. - Phenomenon: Some files fail to parse when uploading multiple documents in batches. Cause: The
PARSE_FILE_TIMEOUT_SECONDSsetting is too short. Parsing of large files such as furnace temperature curves is terminated before completion.
How to Verify Correct Configuration
- Upload a single heat treatment process card PDF, and check that the parsed result includes all process parameters and associated descriptions with no missing content.
- Review the chunking results to confirm that linked parameters such as heating temperature and holding time are located in the same text chunk.
- Upload a batch of furnace temperature curve files, and check that there are no timeout errors in parsing tasks and all files have completed parsing.
- Call the text stream interface to confirm that the returned content retains the original document's heading hierarchy and paragraph breaks, and conforms to text typesetting specifications.
The values provided are common starting points and should be measured against the reader's own samples.