What the Data for This Category Looks Like
It typically involves injection molding, stamping, die-casting mold enterprises and mold centers of manufacturing enterprises. Daily documents include mold design task sheets, DFM reports, part drawings, processing process cards, mold trial reports, mold modification notices, mold records, etc. Data sources include customer product models, mold flow or forming simulation files, CAD/CAM exported files, machining and measurement data, records related to mold trial samples. Update frequency is triggered by project phases, each mold trial or mold modification. Most documents mix structured tables and parameter fields with units, and some embed CAD drawings or simulation result attachments.
What Constraints These Characteristics Impose on the Document Parsing and Chunking Link
Mold documents mix structured tables, parameter fields with units, and non-text attachments, requiring precise retention of parameter context during parsing to avoid splitting that breaks the associated logic of process parameters; frequent updates from mold trials and modifications require chunking by project modules to prevent mixing of cross-project content; embedded CAD drawings and simulation files require parsing to adapt to non-pure text content, avoiding loss of process instructions bound to drawings; document updates have no fixed cycle, requiring incremental parsing by project dimension to avoid repeated processing of full historical data.
How to Configure the Settings
These configurations are adapted for FastGPT v4.17.0:
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Mold design documents contain structured parameters with units and long-text process descriptions. This range balances parameter integrity and retrieval granularity |
chunkSplitter | Split by paragraph + title | Mold documents are often organized by design phase and process link. Splitting by title preserves module integrity |
customPdfParse | Enabled | Some documents contain CAD-exported PDF drawings. Custom parsing supports extraction of non-pure text embedded content |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Large CAD simulation PDF files take longer to parse. This range avoids timeout errors |
chunkSettingMode | Manual configuration | Adjusting chunking rules for structured tables and drawing attachments in mold documents requires more flexibility from manual configuration |
Three Common Misconfigurations
- Phenomenon: A timeout error is triggered when uploading large CAD simulation PDF files, with a request timeout prompt. Cause: The
PARSE_FILE_TIMEOUT_SECONDSvalue is not adjusted to the range suitable for large files. The default timeout duration is insufficient to complete parsing. - Phenomenon: When parsing a scanned mold trial report, the extracted text content is empty or garbled. Cause: The parsing mode adapted to scanned documents is not enabled. The default PDF parsing only supports text-based PDFs.
- Phenomenon: Cross-project mold modification notice content is retrieved after chunking, resulting in mixed review results. Cause: The
chunkSplitteris not configured by project module, and the chunking process does not associate document metadata belonging to the project, leading to confused content boundaries.
How to Confirm the Configuration Is Correct
- A single mold design task sheet PDF is uploaded, and the parsed text is verified to have completely extracted fields with units such as the number of cavities and material shrinkage rate.
PARSE_FILE_TIMEOUT_SECONDSis adjusted to 600 seconds, a simulation report PDF larger than 100MB is uploaded, and no timeout errors are confirmed to occur.- After
chunkSplitteris configured to split by title, the chunking results are checked to confirm independent division into modules such as DFM reports and mold trial reports. - After
customPdfParseis enabled, the process instructions embedded in CAD drawings are verified to be correctly extracted without content truncation.
The values provided are common starting points and should be evaluated against relevant samples.