What the Data for This Category Looks Like
The data for this category originates from material grade standards, customer technical requirements, temperature control instruments, furnace temperature uniformity test reports, hardness tester and metallographic equipment test results, and workshop MES systems. Daily documents include heat treatment process cards, furnace loading diagrams, furnace temperature curves, hardness and metallographic reports, equipment inspection sheets, and exception handling sheets. Data is updated per furnace charge and per batch. Fields include furnace temperature range (unit: ℃), furnace loading weight (unit: kg), hardness value (unit: HV or HRC), holding time (unit: minute), and others. The structure varies significantly by document type. Process cards consist mostly of procedural entries, while furnace temperature curves are time-series data.
Constraints Imposed on Knowledge Base Retrieval and Recall by These Characteristics
This category of data is updated in batches per furnace charge. The volume of data associated with a single furnace charge is small, but the update frequency is high. Fast support is required for locating target documents quickly by furnace charge and batch. Diverse document types require different retrieval logic: structured process cards and time-series furnace temperature curves each need tailored matching. Numeric fields such as hardness and furnace temperature must support exact matching and range retrieval. Customer technical conditions are closely tied to material grades, so cross-knowledge base associated retrieval must be supported. Professional term density in individual data entries is high, so retrieval semantic matching accuracy must be optimized to prevent irrelevant results from being included. Exception handling sheets have strong timeliness, so recent exception handling records must be prioritized for recall.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the semantic similarity of heat treatment professional terms, filters low-relevance results, and prevents accidental recall of process data from non-target furnace charges |
limit | Top 8–12 entries | The number of documents associated with a single furnace charge is limited. Controlling the number of recalled entries avoids context overload and adapts to the quick review needs of workshop audits |
searchMode | mixedRecall | Balances structured keyword matching for process cards and furnace loading diagrams, and semantic time-series matching for furnace temperature curves, covering multiple document types |
usingReRank | Enabled | A large number of professional terms appear in heat treatment data. Reranking optimizes the sorting accuracy of recalled results, and prioritizes returning furnace temperature and hardness data most relevant to the current audit question |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands keywords such as material grades and process parameters entered by users, and matches associated customer technical conditions and equipment inspection records |
collectionIds | Configured by workshop/production line grouping | Differentiates knowledge bases for different heat treatment production lines, prevents irrelevant cross-production line data from being recalled, and adapts to parallel production scenarios across multiple workshops |
Three Common Misconfigurations
- Retrieval returns no results without an error prompt. The cause is that
datasetSearchUsingExtensionQueryis not enabled. The system cannot expand keywords such as the entered material grade, leading to failure to match corresponding process card data. - Knowledge base retrieval is triggered intermittently during a conversation. The cause is that the grouping range of
collectionIdsis not fixed, or a forced retrieval node is not added to the workflow. This allows the system to automatically determine whether to trigger retrieval based on context. - Retrieval latency becomes excessive after multiple questions in a single conversation round. The cause is that the number of recalled entries limited by
limitis not set, and the cache mechanism for reranked results is not enabled. Repeated loading of large volumes of time-series furnace temperature curve data causes delays.
How to Verify Correct Configuration
- Enter the material grade of a specific furnace charge, and check whether the recalled results include the corresponding process card and customer technical condition documents.
- Adjust the value of
similarity, observe changes in the relevance of recalled results, and verify whether the threshold setting adapts to professional term matching. - Add a forced knowledge base retrieval node to the workflow, initiate multiple consecutive questions, and check whether retrieval latency stays within a reasonable range.
- Switch the
collectionIdsconfiguration for different production lines, confirm that only heat treatment data from the corresponding production line is recalled, and no irrelevant cross-production line results are included.
The configuration values listed above are common starting points. Adjust them based on your own specific samples and use cases.
FastGPT v4.17.0