What data for this category looks like
The data for this category primarily comes from ERP orders and inventory, BOMs and process routes, equipment capacity, staff shifts, work-in-progress, and supplier delivery systems. Daily documents include sales orders, master production schedules, material requirements plans, weekly and daily scheduling tables, work orders, material shortage lists, and delivery change notices. Fields include order numbers, process step numbers, equipment models, shift time periods, material codes, delivery dates, and others. Units include pieces, hours, days, sets, and similar units. Data updates occur daily, weekly, or when triggered by orders or abnormal events.
What constraints these characteristics impose on knowledge base retrieval and recall
Data sources are scattered and have large format differences. Retrieval must support cross-document field matching to ensure coverage of different types of production documents such as BOMs, scheduling tables, and work orders. Update frequency is unstable and triggered by abnormal events. The retrieval system must support incremental synchronization or on-demand refresh of knowledge base indexes to avoid returning outdated scheduling rules. Document fields contain specialized production terminology and units. Retrieval must accurately match specific fields in process specifications and scheduling tables to ensure returned content meets professional production planning requirements. Data covers all products and processes. The retrieval scope must be limited to production planning-specific documents to avoid mixing irrelevant content.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Match specialized fields in process specifications and scheduling tables, filter low-relevance general content |
limit | Top 8 entries | Cover multi-process data for a single work order while controlling total context length |
searchMode | fullTextRecall | Adapt to cross-field matching for multi-format documents, support accurate field retrieval for BOMs and scheduling tables |
usingReRank | Enabled | Reorder recall results by order priority, prioritize matching the process requirements of urgent work orders |
datasetSearchUsingExtensionQuery | Enabled | Supplement extended scheduling-related keywords such as "emergency order rescheduling" and "material shortage adjustment" to match hidden process standards |
collectionIds | Specify by business module | Only bind knowledge base collections related to PMC and shop scheduling, exclude irrelevant administrative and financial documents |
This configuration is based on FastGPT v4.17.0.
Three common configuration mistakes
- Symptom: Slow knowledge base import speed, unable to match daily updated scheduling tables and work order data. Cause: Batch parallel import configuration is not used, only single-file sequential upload workflows are employed.
- Symptom: Non-production scheduling documents appear in retrieval results, unable to accurately match process specification requirements. Cause:
collectionIdsis not specified to bind dedicated production planning knowledge base collections, so the retrieval scope covers irrelevant administrative and financial documents. - Symptom: Retrieval results are delayed in emergency order insertion scenarios, unable to return the latest rescheduling rules in a timely manner. Cause: A knowledge base synchronization mechanism triggered by abnormal events is not set, only fixed-cycle full updates are relied on.
How to confirm the configuration is successful
- Upload a batch of test scheduling tables and work order documents, run the retrieval query "2024 October Week 3 machining work order process requirements", confirm whether returned results only include production planning-related documents.
- Simulate an emergency order insertion scenario, upload a temporary delivery change notice, run the retrieval query "Emergency order insertion rescheduling rules", confirm whether results include the latest adjusted content.
- View retrieval logs to confirm that
usingReRankanddatasetSearchUsingExtensionQueryhave taken effect, and that returned results are sorted according to the logic of prioritizing urgent work orders. - Test batch import of multiple docx and pdf format scheduling documents, confirm whether import completion time meets business update rhythm requirements.
The values provided here are common starting points and should be evaluated against the reader's own samples.