What data for this category looks like
Core data for this category comes from daily outputs and system entries submitted by manufacturing enterprise equipment departments and maintenance teams. This includes equipment ledgers, point inspection forms, maintenance plans, work orders, failure analysis reports, spare parts lists, and special equipment inspection records. Data sources cover EAM/CMMS systems, device sensor collections, operator manual point inspections, historical maintenance archives, spare parts inventory ledgers, and statutory inspection information. Update cycles include shift-based point inspection submissions, weekly or monthly maintenance plan distribution, and real-time records when failures occur. Document fields include equipment numbers, point inspection parameter thresholds, failure codes, spare part models, inspection validity periods, and other fields. The length of individual documents varies widely.
What constraints these characteristics impose on source citation and traceability
Multi-source and heterogeneous data sources require citation traceability to accurately match exclusive fields such as equipment numbers and failure codes. This prevents irrelevant production batch documents from being retrieved. The high-frequency update business rhythm requires filtering expired data, and only retaining valid point inspection and maintenance records from recent cycles. The wide variation in individual document length requires controlling the total token usage of citations. This avoids exceeding the context window. Maintenance decisions only require core failure analysis and spare parts information, and do not need to retrieve all historical archives. This requires limiting the number of single retrieval results.
How to set the configuration
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 8000–12000 token | The length of individual documents such as equipment point inspection forms and maintenance work orders varies widely. This range covers complete citations of most core fields and avoids truncating key parameters |
| Single search limit | Top 5 entries | Team point inspection and maintenance data is scattered across multiple work orders. Too many results will exceed the context window and affect the coherence of maintenance decisions |
detail | Enabled | Required fields such as equipment number, point inspection time, and spare part model of cited documents must be returned to meet traceability requirements |
responseData | Return top 3 citation metadata | Maintenance decisions only require core failure analysis and spare parts lists. Too many citations will distract from decision-making |
| Data source filtering rules | Updated within the last 30 days | Point inspection and maintenance plans are updated weekly or monthly. Expired data has no practical reference value |
Three Common Misconfigurations
- Symptom: Generated responses exceed the preset token limit, and logs show the number of citation sources is far higher than expected. Cause: The
quoteMaxTokensetting exceeds the average length of equipment-related documents, and the number of single search return results is not limited. - Symptom: Citation prompt text in the interface does not switch to English as required. Cause: No English prompt template is configured in the system prompt associated with the knowledge base, and only the default Chinese configuration is retained.
- Symptom: Returned citation results do not match the equipment number of the current point inspection. Cause: No filtering rule based on the equipment number field is configured, and equipment maintenance documents from other workshops are retrieved.
How to Confirm the Configuration Is Correct
- Upload a single equipment point inspection form and maintenance work order, trigger the citation traceability process, and check if the returned citation metadata includes exclusive fields such as equipment number and point inspection time.
- Adjust
quoteMaxTokento the preset value, generate a response containing multiple citations, and verify that the token count does not exceed the preset limit. - Enter the failure code of a specific device, check whether all retrieved citation documents are associated with the device’s number, and there is no irrelevant production batch data.
- Switch the system prompt to English, and check whether the citation prompt text is synchronously changed to the English format.
The values provided are common starting points and should be validated against in-house deployment samples.