What This Category's Data Looks Like
The core data for this category covers the entire production and testing lifecycle of power battery packs, energy storage cabinets, and BMS systems. Daily documents include grouping rules, BOM lists, welding torque records, BMS flashing records, insulation and voltage withstand reports, EOL test data, and system configuration lists. Data sources include cell capacity grading databases, factory MES systems, welding equipment logs, torque tool records, BMS software version repositories, and EOL test bench data. The system updates data once per manufactured pack or energy storage cabinet, and synchronizes updates each time the BMS software version changes. Each document field uses dedicated units matching its corresponding process step, such as newton meters for torque values and volts for test voltages.
What Constraints These Characteristics Impose on Multi-turn Dialogue and Prompt Engineering
Multi-source data requirements mean the dialogue system must correlate variables from multiple independent data sources, to avoid missing information from a single source. The high update frequency requires multi-turn dialogue to call the latest production and testing data in real time, to ensure the timeliness of process documents and exception reports. The multi-field structure with dedicated units requires prompt engineering to clearly specify field ranges and matching rules, to reduce parameter confusion. The unique data characteristics of individual packs or cabinets require each dialogue session to bind a unique work order ID, to ensure generated content fully matches the corresponding product. Severe consequences such as thermal management failure and BMS configuration errors require multi-turn dialogue to include a parameter verification step, to ensure the accuracy of key data such as high-voltage connections and insulation tests.
How to Set Configurations
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Bind the unique identifier of the current production work order | Correlate the exclusive BOM, welding records, and EOL test data for a single pack or cabinet, to avoid information confusion across work orders |
messages | Retain the previous 15–20 turns of dialogue content | Cover full-process historical information including grouping rules, welding parameters, and BMS configurations |
variables | Import three specified datasets: "cell capacity grading data", "welding torque records", and "EOL test data" | Directly call the core production and testing data sources for this category, to reduce manual data entry |
| Maximum context length | 8000–12000 characters | Adapt to the total length of multi-stage documents, to avoid truncation of critical information |
| Question optimization switch | Enabled | Simplify the expression of professional process terms, to match the input habits of on-site engineers |
| Variable reference temperature | 0.2–0.4 | Ensure accuracy of process parameters and exception descriptions, reduce subjective bias |
Three Common Mistakes
- After configuring the database connection plugin, users see an error prompt, and the log shows a data source connection timeout. The cause is that the communication ports of the factory MES system and EOL test bench have not been added to the FastGPT outbound access whitelist.
- The temperature setting button disappears when using variable reference mode, making parameter adjustment impossible. The cause is that in FastGPT v4.17.0 and earlier versions, variable reference mode locks temperature parameters by default. Users must specify temperature parameters via custom prompt engineering.
- When users click the dialogue copy button, the generated Markdown content has no carriage returns, leading to messy formatting after pasting. The cause is that the automatic line break rendering switch for dialogue content has not been enabled, or the Markdown format line break configuration is incorrect.
How to Confirm Correct Configuration
- Enter known cell capacity grading parameters, check whether the dialogue can call the bound
variablesdata to generate corresponding grouping rules, and confirm that the data source association is working properly. - Check whether the
chatIdbound to the session matches the unique identifier of the current production work order, and confirm the matching between the session context and the corresponding product. - Test copying dialogue content, check whether the generated Markdown format includes correct line breaks, and ensure that the layout is normal after pasting.
- Adjust the temperature parameter in the custom prompt engineering, check whether the generated exception report maintains accurate professional term expressions, and has no subjective bias.
Note: The configuration values listed here are common starting points. Adjust them based on your own specific samples and use cases.