What this category of data looks like
Daily documents generated by engineering, production, quality control, and laboratory teams at printed circuit board factories. These include engineering production instructions, Gerber parsing records, drilling and lamination parameters, chemical solution analysis reports, AOI reports, electrical test reports, and slice inspection reports.
Data sources include customer-provided Gerber files, drill tapes, BOMs, CAM data, device parameters, bath solution test results, AOI data, electrical test data, and lab data. Update frequency is per production batch, per shift, and continuous collection.
Documents contain three types of content: process steps, parameter tables, and test data. Most fields include professional units, such as mm or mil for drill hole diameters, ℃ for lamination temperatures, and g/L for chemical solution concentrations.
Constraints on document parsing and chunking
PCB manufacturing documents come from scattered sources and update frequently. New process and test records are created for each production batch or shift. This requires the parsing system to support incremental triggering.
Different document types have large structural differences. Gerber parsing records are structured parameter lists. AOI reports are test data with embedded charts. Engineering production instructions are chaptered process descriptions. This requires adapting chunking logic to each document type.
Professional units are mixed within documents. Parsing must unify unit labels to avoid information confusion after chunking.
Single documents may be large in size. Parsing errors can lead to scrapping of entire production batches. This imposes strict requirements for parsing timeout limits and chunking accuracy.
How to configure settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | PCB documents include process steps and parameter tables. This length fits the complete semantics of a single process description or set of test data |
chunkSplitter | Split by paragraph + heading | Most PCB process documents have clear chapter headings and paragraph divisions. This chunking method preserves the logical structure of the document |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Parsing large batches of Gerber file packages or long slice reports takes significant time. This range covers processing durations for most standard documents |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Fits the common sizes of batch-exported Gerber compressed packages and multi-batch AOI report collections |
customPdfParse | Enabled | PCB process documents often include professional charts and parameter tables. Custom parsing preserves table structures and OCR-recognized image content |
chunkSettingMode | Single document priority | Prevents mixing content from other documents during chunking. This fits the business scenario of single-document queries |
Three common configuration errors
- Phenomenon: Parsing loads historical documents together for processing. Returned results include irrelevant historical process data. Cause:
chunkSettingModeis not set to single document priority. The default configuration calls the chunking context of the global knowledge base. - Phenomenon: After uploading image attachments of AOI reports, the model returns empty parsing results. Or chemical solution analysis documents show OCR garbled text. Cause: The image parsing switch for
customPdfParseis not enabled. OCR language adaptation rules are not configured. - Phenomenon: After uploading a single drilling parameter PDF, queries return content from the entire library of process documents. It is not possible to limit results to the currently uploaded file. Cause: The uploaded file ID is not associated with the chunking task. The single document binding logic is not enabled during chunking.
How to verify correct configuration
- Upload a single drilling parameter PDF. Check the parsed chunk list in the FastGPT backend. Confirm that only the chapter and parameter content of this document are included, with no entries from other historical documents.
- Upload an AOI report that includes images. Confirm that the parsed result includes test data text from the images, with no garbled text or empty fields.
- Upload a Gerber compressed package under 1000 MB. Wait for parsing to complete. Confirm that no timeout error is triggered, and backend logs show parsing succeeded.
- Initiate a query based on the currently uploaded document. Confirm that returned answers are only based on this document content, with no reference to other knowledge base entries.
The values provided are common starting points. They should be evaluated against relevant samples.