What the data for this category looks like
Data for this category comes from end-to-end production workflows at professional heat treatment plants and heat treatment workshops in automotive and bearing component factories. It covers documents including heat treatment process cards, furnace loading diagrams, furnace temperature curves, hardness and metallographic reports, equipment inspection forms, and exception handling records. Data is collected from material grades, customer technical requirements, temperature control instruments, furnace temperature uniformity testing equipment, hardness testers and metallographic equipment, and MES systems. Data is updated per furnace load and per batch. Documents include fixed fields such as heating temperature (℃), holding time, hardness value (HV), and furnace temperature fluctuation range. Document structures vary significantly by type. Furnace temperature curves are mostly time-series charts, while process cards use structured text.
Constraints on document parsing and chunking
This category’s diverse document types and major structural differences create parsing challenges. Plain text parsing alone cannot extract valid information from time-series furnace temperature curves and metallographic reports with charts. Additional processing links charts and their associated text. The high-frequency update cadence of per furnace load and per batch requires the knowledge base to support fast batch parsing. Single-document parsing duration must align with production cycles. Accurate extraction of fixed fields such as temperature and hardness values directly impacts subsequent review work. Chunking must retain the binding relationship between fields and their context, to avoid losing associated information after splitting. Business logic-driven splitting of documents with mixed multi-source data prevents field matching failures during review. Indiscriminate character-based cutting causes these failures, so targeted splitting is required.
How to configure settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category of documents includes time-series curves, structured fields, and long-form process descriptions. A chunk length of 800–1200 characters balances field relevance and retrieval accuracy |
chunkSplitter | Customized by document type | Split furnace temperature curve documents first by timestamp, and process card documents by process paragraphs. This avoids indiscriminate cutting that disrupts business logic |
customPdfParse | Enable OCR mode | Some documents include scanned furnace loading diagrams and metallographic report images. Enabling OCR extracts text and parameter information from images |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Documents associated with a single furnace load may include multiple time-series curve charts. A 600-second timeout setting covers the full parsing process |
UPLOAD_FILE_MAX_SIZE | 500 MB | When batch uploading multiple furnace load documents, a 500 MB upper limit adapts to production data files exported in bulk from workshops |
Three common mistakes
- After uploading a PDF document with scanned images, the parsed result has no embedded text from images. The cause is failing to enable OCR mode for
customPdfParse. Basic text parsing alone cannot extract parameter text from images. - The knowledge base prompts a timeout when parsing a single furnace temperature curve document. The cause is the single-document parsing duration exceeding the value set for
PARSE_FILE_TIMEOUT_SECONDS. Point-by-point parsing of time-series curves requires longer processing time. - When building a knowledge base, table datasets cannot be selected for parsing. The cause is failing to enable the permission switch for the corresponding dataset, or the current filtering logic does not cover production table documents.
How to confirm configurations are set correctly
- Upload a test PDF containing a scanned furnace loading diagram, and check if the parsed result includes text information such as furnace loading position and number from the image.
- Upload a single complete furnace temperature curve document, and verify that the parsing process has no timeout prompts and the parsing duration is lower than the value set for
PARSE_FILE_TIMEOUT_SECONDS. - Select a heat treatment process card document, and check that the split chunks retain continuously associated content for processes, heating temperature, and holding time, with no field splitting breaks.
- Batch upload 5 single furnace load documents, and verify that the total file size does not exceed the value set for
UPLOAD_FILE_MAX_SIZE, with no file interception prompts.
The values provided are common starting points and should be measured against applicable local samples.