What this category’s data looks like
This category covers the full forging workflow. Data sources include raw material quality certificates, furnace heating systems, press collection data, MES systems, non-destructive testing instruments, and customer forging drawings. Daily generated documents include forging process specifications, cutting lists, furnace heating logs, forging process cards, mold life records, heat treatment work orders, and ultrasonic testing reports. Data is updated per individual forging, batch of materials, or furnace heat. Included fields are furnace batch number, forging ratio, fiber direction, and defect type. Some data uses units such as MPa and hours.
Constraints on Vector Models and Indexes
Data sources for this category are scattered, and update frequency fluctuates with production batches. Indexes must support fast incremental updates by furnace batch number or forging batch, to avoid excessive time spent on full index rebuilding. Specialized fields such as forging ratio, fiber direction, and defect type carry industry-specific semantics. A vector model adapted to the mechanical manufacturing field must be selected to ensure proper semantic alignment. Text lengths vary widely across document types, from short texts like cutting lists to long texts like ultrasonic testing reports. Configuration settings for defaultToken and maxToken must be adjusted to cover industrial documents of all lengths. Structured collection data and unstructured documents are stored together. Clear field mapping rules for vector indexes must be defined to prevent vector confusion across data from different sources.
How to Configure Settings
All configurations below are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk length | 800–1200 characters | Ensures semantic integrity for long texts such as ultrasonic testing reports and forging process cards, and avoids unnecessary splitting of specialized terms |
| Overlap length | 100–150 characters | Compensates for semantic gaps at chunk boundaries, and covers key specialized fields such as forging ratio and fiber direction that span multiple chunks |
| Similarity threshold | Calibrated based on actual testing | Must match business accuracy requirements for forging defect detection and process matching, to avoid false recalls or missed recalls |
| Single search limit | Top 10 results | Balances recall efficiency and result coverage, and adapts to rapid retrieval needs for batch forging audits |
defaultToken | 8192 | Adapts to the upper limit of vector generation for long documents, and covers the text length of complete process specifications |
| Maximum context | 600 seconds | Adapts to response time requirements for batch retrieval of industrial data, and avoids timeout interruptions |
Three Common Configuration Mistakes
- Scenario: Enabling the Doubao-embedding-large index model, entering a custom request address and APIKey, and testing directly results in an error. Cause: The request header format or request path for this model was not matched, and mapping rules for specialized fields were not configured per vector generation requirements for industrial scenarios.
- Scenario: Table data in the knowledge base (such as pressure value tables from furnace logs) fails to generate valid vectors. Cause: Multi-vector support configuration for table data was not enabled, and vector generation mapping rules for table columns were not specified.
- Scenario: The number of retrieved results does not match expectations, and forging data from non-current batches is returned. Cause: Index filtering rules were not configured based on furnace batch number or forging batch, leading to accidental recall of cross-batch data.
How to Verify Proper Configuration
- Run a vector generation test, upload a single ultrasonic testing report, and verify that the generated vector dimensions match the standard dimensions of the selected model.
- Trigger an incremental index update, upload a new furnace log, and check that the index update log only displays processing records for the new batch, with no full rebuild prompts.
- Search for forging data associated with a specified furnace batch number, and confirm that returned results only include documents corresponding to that furnace batch number, with no cross-batch data included.
- Test long text chunking performance, import a complete forging process specification, and confirm that no specialized terms are split unnecessarily.
The values provided here are common starting points, and should be measured against the reader's own samples.