What This Category’s Data Looks Like
This category covers the entire production process for air conditioners, refrigerators, washing machines, and small household appliances. Core documents include bill of materials (BOM) configurations, assembly work instructions, refrigerant charging records, electrical safety test reports, performance spot check reports, and repair records. Data sources cover ERP/MES systems, component batch ledgers, tightening and charging equipment logs, safety tester data, performance laboratory reports, and after-sales fault databases. Updates are performed per single product, per shift output batch, or per spot check plan. Each data entry includes fields such as equipment number, batch number, test parameters, and result judgment. Units include process parameter units such as pressure, current, and noise decibels.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
Mixed-model mixed-line assembly requires retrieval to accurately match product models and process steps, to avoid cross-model data confusion. High-frequency updates for single-product or batch data require the retrieval system to support real-time or near-real-time recall. Balance between data freshness and retrieval efficiency is necessary. Scenarios with mixed storage of multiple document types require retrieval to filter recall scope by document type. This ensures assembly links prioritize calling work instructions, and safety compliance links prioritize calling test reports. Consumer-facing product characteristics require retrieval results to comply with safety compliance and energy efficiency requirements, to avoid compliance risks caused by recalling abnormal data. Structured data generated by equipment and laboratories requires retrieval to support filtering results by parameter thresholds, to accurately locate abnormal records.
Configuration Settings
This configuration is designed based on the knowledge base function of FastGPT v4.17.0. Specific parameters are as follows:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the precision requirements for home appliance manufacturing process parameters, to avoid low-similarity irrelevant data being included in ledger summaries |
limit | Top 8–12 entries | Single-product ledger summaries need to cover multiple links such as assembly, testing, and repair. Controls context length to avoid exceeding model limits |
searchMode | embedding | Prioritizes recall based on semantic and numerical features of process parameters, adapting to retrieval needs for structured production data |
usingReRank | Enabled | Mixed-model data from mixed-line production can easily produce results with similar semantics but incorrect model numbers. Reranking improves precision |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms related to process parameters, covering recall for fields such as equipment numbers and batch numbers |
collectionIds | Specified per production ledger-specific collection | Isolates data from other categories, to avoid calling irrelevant documents during mixed-line retrieval |
Three Common Configuration Errors
- Symptom: Index configurations cannot be edited directly from the retrieval results page. Adjustments must be made within the corresponding knowledge base collection. Cause: After platform interface updates, the quick edit entry on the retrieval results page has been removed. Index configurations must be completed via the collection management page specified by
collectionIds. - Symptom: Retrieval results returned by platform testing and API calls are inconsistent. Cause: Knowledge base index configurations for test and production environments are not synchronized, or the
datasetSearchUsingExtensionQueryparameter is only enabled in the test environment. - Symptom: Retrieval takes longer than 40 seconds, or returns no results. Cause: The
similaritythreshold is set too low, resulting in excessive recall of low-relevance data, orusingReRankis not enabled, leading to an overly large candidate set that increases retrieval and reranking time.
How to Verify Successful Configuration
- The
collectionIdsconfiguration for the knowledge base collection should be checked, and confirmation made that only document collections related to production ledgers are included, with no irrelevant data mixed in. - A test query including specific product models and process parameters can be submitted, and the
similarityscores of returned results verified to fall within the preset range. - Retrieval results from platform testing and API calls should be compared, and confirmation made that the number of recalled entries and content are consistent.
- The same query can be executed multiple times, and retrieval latency verified to stay within a reasonable range with no significant fluctuations.
The values provided here are common starting points, and should be evaluated against appropriate samples.