What this category's data looks like
Data for inspection and process record audit in electric motor and generator manufacturing comes primarily from electromagnetic design data, copper wire and insulation material batch records, winding equipment logs, dynamic balancing machine test data, electrical test bench reports, and batch information from ERP/MES systems.
Daily documents include winding process cards, insulation treatment records, dynamic balance reports, dielectric strength test reports, and type and factory inspection test reports. Data is updated per product unit or batch, and synchronized after each design change. Documents have fixed fields and units, such as winding turns, insulation thickness, vibration amplitude, and insulation resistance value.
What constraints these characteristics impose on document parsing and chunking
Documents in this category often contain complex nested tables, formal reports with signatures, multi-chapter process descriptions, and closely linked fields. For example, winding turns must correspond to material batches and equipment parameters. A single test report may include multiple sets of test data. Random chunking will break the logical association between parameters.
Digitally signed PDFs cannot be read by conventional parsing tools. Merged cells in complex tables and cross-page distribution can also cause field misalignment after parsing. In addition, single documents have large file sizes, requiring sufficient parsing time. Timeouts will cause task failure.
How to configure the settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of individual parameter blocks or process paragraphs in documents of this category, avoiding splitting of key associated fields |
chunkSplitter | By title + paragraph | Preserves the chapter structure of documents, adapting to hierarchical content in winding process cards and test reports |
customPdfParse | Enabled | Adapts to digitally signed PDF documents, resolving unrecognizable encrypted content |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Meets the parsing time requirement for large single test reports, avoiding timeout failures |
UPLOAD_FILE_MAX_SIZE | 100 MB | Adapts to the maximum file size limit for uploading multiple combined test reports for a single product batch |
chunkSettingMode | Manual | Allows separate adjustment of chunking rules for complex table documents, avoiding damage to parameter associations |
Three common configuration mistakes
- Uploading a digitally signed PDF results in empty content. The cause is that the
customPdfParseparameter is not enabled, and the parsing logic for encrypted signed documents is not adapted. - An error occurs when calling the parsing tool, indicating file read failure. The cause is that the
UPLOAD_FILE_MAX_SIZEsetting is too small, exceeding the upload file size limit. - Fields are misaligned after parsing complex tables. The cause is that
chunkSplitteris set to split by character, instead of splitting by table blocks or paragraph titles, which damages the structural integrity of the table.
How to confirm the configuration is correct
- Upload a digitally signed dielectric strength test report, and check whether core parameter content such as insulation resistance and test voltage is fully displayed after parsing.
- Upload a single type test report with a volume of approximately 80 MB, and check whether the parsing task is completed within 300 seconds without timeout errors.
- Upload a winding process card, and check whether the chunked content retains complete paragraphs of key parameters such as winding turns and wiring method, without field splitting or misalignment.
- Check the knowledge base parsing logs of FastGPT v4.17.0 to confirm that there are no prompt records of file read failure, format error, or timeout.
The values listed above are common starting points and should be evaluated against the reader's own samples.