What this category's data looks like
Core data for this category originates from resin batch property sheets, injection molding machine operating parameters, mold temperature controller records, production MES systems, and dimensional and appearance inspection reports. Daily documents include injection molding process cards, raw material drying records, mold histories, first-piece inspection forms, patrol inspection forms, and material change and machine cleaning records. These documents contain fields such as injection pressure, holding time, mold temperature, raw material moisture content, and recycled material ratio, with units including MPa, seconds, ℃, %, and others. Data is updated once per production batch, each mold change or material change, and each shift production cycle.
What constraints these characteristics impose on the "document parsing and chunking" link
Documents in this category have both structured table and semi-structured paragraph characteristics. They have high update frequency and strong timeliness. Parsing and chunking must accurately match production scenario logic. Large batches of uploaded mold histories and patrol inspection form collections have significant volume, so support for large file chunked parsing is required. Documents from different batches have field differences, so unrelated data across batches must not be merged into chunks. Key parameters such as filling, holding, and cooling windows in process cards must not be truncated during splitting. Truncation would negatively impact subsequent question-and-answer accuracy. Associated parameters such as raw material drying and recycled material ratio must retain contextual association, and cannot be split into independent chunks.
How to set the configurations
These configurations apply to FastGPT v4.17.0
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of single-segment process parameters in injection molding process cards, avoiding splitting critical descriptions across filling, holding, and cooling windows |
chunkSplitter | Paragraph + custom delimiter | Adapts to fixed title structures such as "Process Parameters", "Drying Records", and "Mold Histories" in injection molding documents, retaining contextual association of parameters |
customPdfParse | Enabled | Recognizes structured tables and fixed headers in injection molding documents, extracts dedicated fields such as "raw material moisture content" and "recycled material ratio" |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the conventional volume of batch-uploaded mold history and patrol inspection form collections |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Covers the conventional time required for large batch file parsing, avoiding mid-parsing timeout interruptions |
chunkSettingMode | Manual mode | Adjusts chunking rules for specific fields in injection molding documents to ensure key parameters are not split |
Three common mistakes
- Symptom: Large batch document uploads are interrupted mid-process, and parsing cannot resume after restarting the service. Cause: Reasonable values for
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDSare not configured. Batch files may exceed storage limits, or parsing time may exceed thresholds, triggering interruptions. - Symptom: Parsed documents are visible, but the model does not reference process parameter content from the documents. Cause: The
chunkSplitterdoes not split content according to document structure. Critical content such as filling, holding, and cooling windows and raw material drying parameters may be split into different chunks, making contextual association impossible. - Symptom: No parsing progress prompt appears after uploading a file via the dialog box, and no background error messages are displayed. Cause:
customPdfParseis not enabled. General PDF parsing cannot recognize structured tables and dedicated fields in injection molding documents, resulting in silent parsing failure.
How to confirm configurations are set correctly
- Upload a single injection molding process card, and check if the parsed text retains dedicated fields such as filling pressure, holding time, and mold temperature.
- Upload a batch collection of material change and machine cleaning records, and verify that the parsing task completes within 600 seconds with no interruptions or timeout prompts.
- Search for questions related to a batch's raw material moisture content, and confirm that returned results include complete parameters and context from the corresponding chunk.
- Adjust
chunkSizeto the 600–1500 character range, and verify that split paragraphs do not span descriptions of critical process windows.
The values given are common starting points and should be measured against the reader's own samples.