What Data for This Category Looks Like
Typical entities in the general mechanical equipment assembly category are manufacturers of pumps, valves, compressors, reducers, machine tools and non-standard equipment. Daily documents include assembly process cards, BOMs, torque records, clearance 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. Update frequency is per unit or per order. Documents include structured fields such as material codes, torque values (unit: N·m), clearance values (unit: mm), and coaxiality parameters. Some documents use a mixed format of text and tables.
Constraints on Document Parsing and Chunking
This category organizes production per order. Document structures vary significantly across different orders. The document set for a single device includes multiple files in different formats. Chunking must retain field associations within the same document. For example, the material code and quantity in a BOM table must stay in the same chunk. Structured numerical fields in documents have high parsing accuracy requirements. Chunking that is too fine splits field associations. Chunking that is too coarse causes redundant context. Documents from different systems have inconsistent formats. Native PDF parsing often results in table misalignment, so multi-format parsing adaptation is required. In addition, tender documents require rapid extraction of BOM data related to quotations. Parsing and chunking must prioritize the integrity of core business fields.
Configuration Settings
This document applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category’s documents mix text paragraphs and structured tables. This range balances context integrity and retrieval efficiency, and avoids splitting associated fields such as BOMs and torque records |
chunkSplitter | Separate by paragraph + table | Native character-based chunking will destroy the structured content of tables. Separating by paragraph and table preserves the integrity of individual document units |
customPdfParse | Enabled | PDFs for this category are mostly exported from PLM/ERP. Native parsing often causes table misalignment. A custom parsing service can adapt to structured document formats |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | The document set for a single device may include multi-page commissioning reports and multiple inspection sheets. This duration covers parsing needs for most documents, and avoids timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports uploading full assembly documents for a single order, and meets document volume requirements for most batch assembly scenarios |
chunkSettingMode | Manual mode | Document structures vary significantly across different orders. Manual mode allows adjusting chunking parameters for individual documents to optimize retrieval accuracy |
Three Common Configuration Mistakes
- Tables appear misaligned or have missing fields after uploading a PDF. The cause is failing to enable
customPdfParse. The native parsing service has insufficient ability to adapt to structured tables. - File parsing fails and returns a timeout error during conversation. The cause is
PARSE_FILE_TIMEOUT_SECONDSbeing set too short, or not configuring a custom parsing service, leading to parsing time exceeding the default limit. - Retrieval fails to associate BOMs with corresponding torque records. The cause is
chunkSizebeing set too small, splitting a single associated document into multiple chunks and breaking data context associations.
How to Verify Correct Configuration
- Upload a BOM PDF from a PLM system, and check that the parsed table is complete with no misalignment or missing columns.
- Upload a document containing multi-page assembly records, and check that the chunking result retains the full context of a single record, without splitting a single record into multiple chunks.
- Upload a document set under 500 MB, and confirm that the upload and parsing process has no timeout interruptions, and that chunking results can be generated normally.
- Adjust the
chunkSizeparameter to 600 characters and 1400 characters, compare the field association integrity of the chunking results, and confirm that the current setting is suitable for the current business documents.
The values listed above are common starting points, and should be evaluated against in-house samples.