What this category’s data looks like
PCB manufacturing-related data comes from Gerber files, drill files, and BOM lists delivered by customers. It also includes outputs from factory CAM systems, production equipment, bath solution testing, AOI inspection equipment, electrical testing equipment, and laboratories. Daily documents cover engineering production instructions, Gerber parsing records, drilling and lamination parameters, chemical analysis reports, AOI inspection reports, electrical testing reports, and sectioning reports. Data update frequency is adjusted continuously based on production batches and shifts. A single production batch is associated with multiple types of professionally formatted documents. Some documents include continuous process parameter and test result fields.
What constraints these characteristics impose on document parsing and chunking
PCB manufacturing documents come from both customer-delivered materials and outputs from every stage of production. Supported formats include professional CAD files, custom reports, and common office documents. Parsing tools must support multi-format adaptation. Data update frequency is adjusted in real time per production batch. Documents per batch have large total file sizes, so the system must support batch uploads and long document parsing. Documents contain closely linked process parameters and test results. Chunking must preserve semantic relationships between fields, and avoid splitting critical paired information. Parsing errors can directly cause production batches to be scrapped. As a result, parsing and chunking accuracy requirements are high. Key parameters or test conclusions must not be omitted.
Configuration Settings
The following configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Covers parsing time for 100MB-level PCB process reports and sectioning reports, prevents long document parsing timeouts |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Meets total upload requirements for multiple documents linked to a single production batch, supports daily batch file upload scenarios |
chunkSize | 800–1200 characters | Preserves complete semantics of a single set of linked process parameters or test results, avoids splitting critical paired fields |
chunkSplitter | By professional fields and paragraph boundaries | Adapts to structured layouts of PCB documents such as engineering instructions and inspection reports, preserves associations between paragraphs and fields |
customPdfParse | Enabled | Adapts to some custom-format PCB inspection report PDFs, optimizes parsing results for professional layouts |
Three Common Configuration Errors
- Symptom: A timeout error appears when uploading PCB process reports larger than 10 MB. Cause: The
PARSE_FILE_TIMEOUT_SECONDSconfiguration value is lower than the actual parsing time required for this document type. - Symptom: Some drill parameter documents in DOCX format fail to parse normally. Cause: Custom parsing rules have not been configured for PCB-specific office documents, so embedded charts or professional fields cannot be recognized.
- Symptom: Tables in AOI inspection reports in the knowledge base are not chunked correctly. Cause: The
chunkSplitterconfiguration is set to split by fixed characters, and the row and column association structure of the table is not preserved, leading to incorrect splitting of table data.
How to Verify Configurations Are Set Correctly
- Upload a single 100 MB Gerber parsing report, and confirm that the parsing task completes within the time limit set by
PARSE_FILE_TIMEOUT_SECONDS. - Upload a DOCX-format AOI inspection report that includes embedded tables, and confirm that the chunking result preserves the complete row and column data of the table.
- Import preset PCB process question-and-answer pairs, and verify that returned results only reference original text from the knowledge base, with no additional content generated.
- Check the
UPLOAD_FILE_MAX_SIZEconfiguration value, and confirm that it exceeds the maximum size of daily uploaded single documents to prevent upload failures.
The values provided are common starting points, and should be evaluated against the reader's own samples.