What the Data for This Category Looks Like
Data for this category comes from daily production and customer communication documents of professional heat treatment plants and heat treatment workshops in automotive and bearing parts factories. Documents include heat treatment process cards, furnace loading diagrams, furnace temperature curves, hardness and metallographic reports, equipment inspection forms, abnormal disposal records, and more.
Data sources cover material grade standards, customer technical requirements, temperature control instruments, furnace temperature uniformity test results, outputs from hardness testers and metallographic equipment, and MES system records. Updates occur per furnace load and per batch.
Single documents contain clear fields. Process parameters include heating temperature, holding duration, and cooling rate. Furnace temperature curves mark real-time temperature and time points. Hardness reports list material grade, test load, and hardness value unit.
Constraints Imposed on Knowledge Base Retrieval and Recall
The high-frequency update requirement per furnace load and per batch means retrieval must quickly match the latest process parameters. This prevents outdated data from being used, which would reduce bidding accuracy.
Multiple types of structured and unstructured documents coexist. The retrieval module must support precise matching of tables, curve data, and text simultaneously.
Data fields include precise parameters with units such as temperature, duration, and hardness value. Unit association must be retained during retrieval to avoid matching deviations.
Dual sources of customer technical specifications and on-site equipment data require retrieval to distinguish general processes from customized requirements. This avoids confusion between standard processes and special working conditions.
Historical data from abnormal disposal records must be recalled for bidding reference on similar issues. Retrieval must cover associated documents for abnormal scenarios.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the precision requirement for matching heat treatment process parameters, avoids too low a value leading to irrelevant results, and avoids too high a value missing customized processes |
limit | Top 8 entries | The process combinations involved in a single furnace load bidding are limited, and too many results will increase the verification cost of quotation preparation |
searchMode | embedding | Balances text semantics and parameter value matching, adapts to multiple types of documents such as process cards and furnace temperature curves |
usingReRank | Enabled | Performs secondary sorting of initially recalled results by relevance, correcting deviations from single vector matching |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms, supplements associated keywords such as material grades and equipment models, adapting to the multi-dimensional requirements of customer technical specifications |
collectionIds | Fill in according to the actually bound exclusive knowledge base ID for heat treatment | Isolates knowledge base data of different categories, avoiding mixing irrelevant documents such as surface treatment and forging |
Three Common Configuration Mistakes
- Phenomenon: Retrieval results fail to correctly reference material grades and process requirement variables provided by customers, leading to matching deviations. Cause: The variable reference assignment rules of the knowledge base search module are not correctly configured, and fields in customer technical specifications are not bound to retrieval parameters.
- Phenomenon: After importing CSV-formatted heat treatment process card data, retrieval cannot recognize structured table fields. Cause: The template import function is not used to upload the CSV file, and direct upload using the text dataset causes the table structure to be lost, preventing precise encoding by the vector model.
- Phenomenon: Retrieval returns a large number of irrelevant surface treatment and forging documents. Cause: The
collectionIdsparameter is not correctly configured, and the exclusive heat treatment knowledge base is not isolated from datasets of other categories.
How to Verify Correct Configuration
- Enter the debugging interface of the knowledge base retrieval module, enter the material grade and process requirements of a target customer, and check whether recalled results only include bound exclusive heat treatment documents.
- Adjust the
similarityparameter, observe changes in the number of recalled results, and confirm that results meet business expectations after adjustment. - Upload a test abnormal disposal document, run a query containing the keywords "crack" and "deformation", and confirm whether the reranking function adjusts the priority of relevant results.
- View the retrieval log, confirm that extended query terms automatically associate fields such as material grades and equipment models, meeting business requirements.
The values provided are common starting points and should be measured against the user's own samples.