What This Category of Data Looks Like
Data for this category originates from engineering, production, quality control, and laboratory processes at printed circuit board factories. It covers engineering production instructions, Gerber parsing records, drilling and lamination parameter tables, chemical solution analysis reports, AOI defect reports, electrical test reports, slice analysis reports, and more. Data sources include Gerber, drill tape, and BOM files delivered by customers, CAM output files, equipment operating parameters, bath solution test results, data collected by AOI and electrical test equipment, and laboratory test records. Update frequencies range from per production batch, per shift, to continuous real-time collection. Documents include multiple structures: plain text, multi-header tables, embedded defect images, and others. Fields include specialized parameters with specific units, such as drill hole diameter, lamination temperature, defect coordinates, and on-resistance values.
Constraints Imposed by This Category’s Characteristics on Document Parsing and Chunking
The multi-type document structures, specialized fields, and high-frequency update characteristics of this category impose multiple constraints on the document parsing and chunking process. First, AOI reports and slice reports include embedded defect images. Both text and image content must be identified to avoid only recalling text information. Second, engineering parameter tables and electrical test reports include multiple sets of related parameters. Chunking must retain the contextual association between parameters and their corresponding batch and equipment numbers, to avoid splitting that breaks business logic. Third, when uploading multiple production batch documents in bulk, single-file parsing duration and overall processing efficiency must be controlled to adapt to high-frequency update data processing needs. Finally, non-standard text such as Gerber parsing records requires accurate identification of specialized fields, to avoid incorrect splitting of specialized parameter groups.
How to Set the Configuration
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the length of parameter entries and associated context in PCB specialized documents, avoiding splitting of drilling parameters and their corresponding batch information |
chunkSplitter | Split by paragraph + table boundaries | Adapts to the mixed text and table structure of AOI reports and electrical test reports, retaining complete data within tables |
customPdfParse | Enable image OCR parsing | Adapts to AOI and slice reports with embedded defect images, enabling simultaneous recall of text and images |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Supports bulk processing of large-volume Gerber parsing records and multi-batch reports, avoiding single-file parsing timeouts |
chunkSettingMode | Manual configuration | Adjusts chunking rules for specialized engineering documents, avoiding automatic splitting that breaks parameter contextual association |
Three Common Misconfigurations
- Symptom: Only text descriptions are recalled after uploading an AOI report, with no annotated content from embedded defect images. Cause: The OCR parsing switch for
customPdfParseis not enabled, and embedded image information in the document is not identified. - Symptom: After importing a drilling parameter Excel spreadsheet, aperture and rotation speed parameters for the same batch are split into different chunk segments. Cause:
chunkSizeis set too small, orchunkSplitterdoes not trigger splitting by table boundaries, breaking the contextual association of parameters. - Symptom: Parsing timeout errors occur when uploading more than 5 electrical test reports in bulk. Cause:
PARSE_FILE_TIMEOUT_SECONDSis set lower than the actual parsing duration of a single report, and does not accommodate the cumulative duration of bulk processing.
How to Confirm the Configuration Is Correct
When an AOI report with embedded defect images is uploaded and a retrieval test is initiated, both image annotations and corresponding text descriptions are confirmed to be recalled. When a drilling parameter Excel spreadsheet with multiple sets of related parameters is imported and chunking results are reviewed, aperture and rotation speed parameters for the same batch are confirmed to be retained within the same segment. When a single large-volume Gerber parsing record is uploaded, no parsing timeout error is confirmed to be triggered, and the file status is confirmed to show completed. When the knowledge base configuration interface is accessed and chunking settings are reviewed, chunkSettingMode is confirmed to be set to manual configuration, and chunkSize is confirmed to fall within the preset range.
The configuration values listed above are common starting points. Adjust them based on specific samples and use cases relevant to the deployment environment.