Data Overview for This Category
This category of data comes from engineering, production, quality control, and laboratory teams at printed circuit board factories. It covers engineering work instructions, Gerber parsing records, drilling and lamination parameters, chemical analysis reports, AOI and electrical test reports, slice reports, and more. Data sources include customer-provided Gerber files, drill tapes, BOMs, CAM output files, device operating parameters, bath test results, and data from all inspection links. The system collects updates continuously per production batch and shift. Each document includes fields such as process parameters, test results, batch numbers, and device numbers. Units include professional process units such as millimeters, ohms, and ppm.
Constraints on Document Parsing and Chunking
PCB process documents contain large volumes of professional parameters and batch-related information. Chunking must retain the binding relationship between parameters and their corresponding batches to avoid losing context after splitting. High-frequency continuously collected data generates many small files, so the system must adapt to the efficiency requirements of batch parsing multiple files. Parsing specialized format files such as Gerber and drill tapes requires dedicated rules, and generic document parsing logic cannot be used directly. Fields vary significantly across different document types: AOI reports focus on defect coordinates and quantity, while chemical analysis reports focus on concentration parameters. The system must support adjusting chunking rules by document type to avoid retrieval errors caused by mixing content across document types.
Configuration Settings
These configurations are optimized for FastGPT v4.17.0 for PCB manufacturing use cases:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | PCB process documents include multiple sets of continuous parameters. This range retains complete process step parameters and batch-related information, preventing parameter breaks after splitting. |
chunkSplitter | Split by professional process nodes | Documents are segmented by processes such as drilling, lamination, and inspection. This matches business users' retrieval habits and fits the structural characteristics of professional documents. |
customPdfParse | Enabled | Adapts to professionally formatted PDF files such as Gerber parsing records and AOI reports, preserving the parameter alignment from the original layout. |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Large Gerber files or batch production ledger parsing require longer processing time. This prevents parsing failures due to timeout. |
UPLOAD_FILE_MAX_SIZE | 2000 MB | Supports batch file uploads for single production batches, adapting to the business need for importing multiple factory documents at once. |
chunkSettingMode | Customize by document type | Configure differentiated chunking rules for different process documents (such as parameter-type and inspection-type documents) to match the retrieval priority of each document type. |
Three Common Configuration Mistakes
- Symptom: After importing Excel-format production ledgers into the table dataset, the correspondence between process parameters and batch numbers is lost. Cause: The column-associated chunking rule is not enabled. The default splitting logic does not retain the row-column binding relationship of the table.
- Symptom: When retrieving AOI report table content stored in the knowledge base, only partial cell data is returned. Cause: The
chunkSizevalue is too small, splitting complete table rows into multiple chunks. This results in only partial fragments being returned during retrieval. - Symptom: PCB process documents chunked with different vector models locally cannot match expected retrieval results after being uploaded to the server. Cause: Unified chunking rules and parsing configurations are not used. Different splitting logic breaks the contextual association of documents.
How to Verify Correct Configuration
- Upload a single typical process document (such as a drilling parameter ledger) and check if the parsed chunks contain complete process parameters and batch-related information.
- Upload multiple small files (such as daily AOI reports) and verify that the parsing completion time aligns with business processing rhythms, with no timeout errors.
- Import an Excel-format BOM file and confirm that the imported data fields exactly match the original table's column names and content.
- Switch between different document types (such as chemical analysis reports and slice reports) and check if the chunking rules automatically adapt to the document structure.
The values provided here are common starting points and should be measured against the reader's own samples.