What the Data for This Category Looks Like
Daily documents for general mechanical equipment assembly include assembly process cards, BOMs, torque records, gap adjustment records, commissioning reports, factory inspection sheets, and packing lists. Data sources cover PLM/ERP, part inspection data, assembly tool collection, test benches, MES, and customer technical agreements. Documents are updated per single device or order. Each document contains structured or semi-structured content such as device model, order number, process parameters with precise units (e.g., torque value N·m, gap value mm), and process steps. Document structures differ across different orders.
How These Characteristics Create Constraints for Document Parsing and Chunking
The document structure for this category changes dynamically based on orders and device models. Non-standard parsing logic is required, and fixed templates cannot be used for chunking. Documents contain a large number of process parameters with precise units, plus hierarchical process and BOM structures. When chunking, the association between fields and parameters must be retained to avoid disconnected content after parsing. Frequently updated single-device or single-order documents need support for batch parallel upload parsing to match business processing rhythms. Time-series documents such as torque records and gap adjustment records must retain context in operation order. This prevents chunking from splitting complete process flows.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Range/Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the paragraph and field lengths of documents such as assembly process cards and BOMs, retains complete context of process parameters and their corresponding descriptions |
chunkSplitter | Split by Markdown headings + paragraphs | Adapts to the hierarchical structure of assembly documents, such as process step titles and corresponding operation instructions in process cards, retains hierarchical associations of BOMs |
customPdfParse | Enabled | Preserves embedded table and parameter column formatting in PDFs, avoids formatting errors and field loss in torque records and gap adjustment records |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses batch parsing time for multiple documents supporting a single device (e.g., BOM + inspection sheet + commissioning report), prevents mid-task interruptions |
chunkSettingMode | Custom mode | Adjusts chunking rules for different document types (BOM, process card, inspection sheet) to adapt to structural differences in order-based documents across multiple product varieties |
Three Common Mistakes
- Hierarchical fields in parsed BOMs are empty. The cause is that
customPdfParseis not enabled. The embedded BOM table structure in the PDF is flattened and loses its hierarchy, making it impossible to restore the hierarchical association. - Parsing times out when uploading multiple documents in batch. The cause is that
PARSE_FILE_TIMEOUT_SECONDSis not adjusted to a duration suitable for multi-document batch processing. Tasks are terminated after exceeding the default threshold. - Adjacent adjustment steps in torque records are split after chunking. The cause is that
chunkSizeis set too small. A single complete torque adjustment process is split into multiple fragmented blocks, which cannot support context association for subsequent audits.
How to Verify Proper Configuration
- A typical assembly process card is uploaded, and parsed chunks are checked for complete association between process step titles and their corresponding operation instructions.
- Multiple different types of documents (BOM, torque records, commissioning reports) are uploaded to confirm parallel parsing tasks run normally.
- Parsed text blocks are reviewed to confirm that process parameters with units are not lost or altered.
- The value range of
chunkSizeis adjusted to verify that chunk length matches the document paragraph characteristics of this category.
The values provided here are common starting points and should be measured against the reader's own samples.