What the data for this category looks like
Production data for SMT mounting and electronic assembly comes from customer BOMs, Gerber coordinate files, MES systems, solder paste and component batch records, placement machines, SPI/AOI inspection equipment, and reflow oven temperature data. Data is generated in real time per work order and each line change. Daily documents include BOM tables, line change records, stencil logs, furnace temperature logs, first article confirmation sheets, SPI/AOI defect reports, maintenance records, and traceability labels. Document fields include part numbers, reference designators, component batches, temperature parameters, defect types, and others. Coordinate files use millimeters as the unit. Furnace temperature records use degrees Celsius as the unit.
What constraints these characteristics impose on document parsing and chunking
The high-frequency real-time update rhythm requires document parsing to support fast batch processing, to avoid delays that reduce ledger summary efficiency. The mixed multi-type document structure, which includes both structured tables and long text paragraphs, requires chunking logic to adapt to different formats, to avoid breaking field association relationships. Specific business fields such as part numbers and reference designators must be fully retained. Chunking must not split cross-field association relationships. The requirement for combined upload of multiple documents per work order requires the parsing process to adapt to large file capacities, while controlling chunk size to avoid context overflow.
Configuration settings
This article applies to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | SMT documents include both BOM tables with dense fields and long-text furnace temperature logs. This range retains complete business fields while avoiding chunk breakage |
chunkSplitter | Mixed table + paragraph splitting | Adapts to the characteristics of SMT documents mixing structured tables and unstructured paragraphs, ensuring all types of data can be fully extracted |
customPdfParse | Enabled | Accurately extract tabular fields from SPI/AOI reports and first article confirmation sheets, avoiding missing key business data with the default parsing logic |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | The parsing duration for a single work order's multiple document collections is usually long. This range covers the complete parsing process and avoids timeout failures |
UPLOAD_FILE_MAX_SIZE | 2000 MB | Meets the common capacity requirements for batch uploading work order documents, with a reasonable upper limit calibrated through actual testing |
chunkSettingMode | Manual mode | Allows separate adjustment of chunking parameters for different document types such as BOM tables and furnace temperature logs, adapting to multi-structure documents in SMT scenarios |
Three common misconfigurations
- After uploading a SPI/AOI report PDF that includes component appearance images, the parsed result does not extract the defect description associated with the images. The cause is that the corresponding parsing configuration is not enabled; the default logic only extracts text content.
- When batch uploading 3 or more SPI/AOI report PDFs, the parsing task returns a timeout error. The cause is that
PARSE_FILE_TIMEOUT_SECONDSis not adjusted; the default duration is insufficient for processing multiple frequently updated document collections. - After exporting an old server knowledge base as a single Excel file and importing it to a new server, some component batch data in line change records is lost. The cause is that documents are not split by work order; chunking merges cross-work order associated data, leading to broken field associations.
How to confirm the configuration is properly set
- Upload a standard SMT BOM table, check if the parsed chunks retain complete part numbers, reference designators, and alternative part fields.
- Upload a PDF document containing SPI/AOI defect data, check if the parsed result includes tabular defect type and quantity fields.
- Upload a 500 MB work order document collection, confirm that the parsing task completes within the preset
PARSE_FILE_TIMEOUT_SECONDSduration. - Switch to
chunkSettingModemanual mode, adjustchunkSize, then upload the test document again, confirm that the chunk length matches expectations.
The values given are common starting points and should be measured against the reader's own samples.