What this category of data looks like
Data for this category comes from formula systems, raw material batch management modules, internal mixers, Mooney viscometers, vulcanizing machines, and laboratory testing systems. Daily documents include formula sheets, weighing records, mixing curves, rubber quick-test reports, vulcanization process cards, mold usage records, and finished product performance reports. Data updates align with each raw material batch shipment, each batch of rubber mixing, and each vulcanization operation. Document fields include formula number, raw material batch number, mixing temperature range, vulcanization time, Shore hardness, tensile strength, and supporting units such as degrees Celsius, minutes, and MPa.
What constraints these characteristics impose on the knowledge base retrieval and recall workflow
High-frequency incremental updates require the retrieval system to support near-real-time synchronization of new batch and process data. This prevents retrieval of expired formulas or operation records. Documents with continuous parameters and table structures such as mixing curves and vulcanization process cards require chunking that preserves individual record integrity. This avoids splitting key information like weighing sequence and temperature curve inflection points. Document structures with multiple fields and units require matching fields to their corresponding units during retrieval. This prevents invalid results with mismatched parameters. Each vulcanization operation ties to a specific batch and mold. Retrieval results must accurately bind context for the corresponding production link, to avoid incorrect cross-batch matches.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precision requirements for mixing processes and vulcanization parameters, to avoid retrieving irrelevant formulas |
limit | Top 6 entries | Covers full-process parameters for a single batch’s mixing and vulcanization, to control total context length |
usingReRank | Enabled | Differentiates between batch data under similar process parameters, to reduce false retrievals |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval scope for referential fields including formula numbers, batch numbers, and mold numbers |
searchMode | embedding | Adapts to unstructured text retrieval needs for mixing curves and process cards |
This page is configured based on FastGPT v4.17.0
Three Common Configuration Errors
- Phenomenon: Weighing records imported from Excel are automatically merged into a single text block, with multiple records not split. Cause: No custom delimiter is configured to match the end marker of Excel rows, and the default splitting logic does not adapt to table row structures.
- Phenomenon: Retrieval results fail to distinguish batch differences between similar formulas, leading to confusion between different batches with identical parameters. Cause:
usingReRankis not enabled, and results are sorted only by basic semantic similarity, with no secondary filtering for results with matching semantics but different batches. - Phenomenon: Queries containing "this batch" or "this mold" do not return corresponding documents. Cause:
datasetSearchUsingExtensionQueryis not configured, and referential content in queries is not expanded for matching, resulting in limited retrieval scope.
How to Verify Configuration Success
- Upload a single Excel weighing document with multiple rows, confirm the knowledge base chunking result shows one data entry per row, to verify the custom delimiter configuration takes effect.
- Enter a query containing a clear batch number, confirm retrieval results only include documents for the corresponding batch, to verify the
similaritythreshold and incremental synchronization rules take effect. - Enter a query containing similar process parameters, confirm returned results are sorted secondarily by batch and time, to verify the
usingReRankfunction is enabled. - Enter a query containing referential expressions, confirm returned results match documents for the corresponding production link, to verify
datasetSearchUsingExtensionQuerytakes effect.
The values provided are common starting points and should be evaluated against the reader’s own samples.