What the data for this category looks like
The data sources for this category include registered process documents, raw material batch ledgers, DCS production control systems, weighing systems, LIMS laboratory information management systems, chromatography data systems, and QMS quality management systems. Data updates occur once per production batch or per process step. Daily generated documents include batch production records, material balance ledgers, reaction and purification process records, cleaning validation records, impurity profile reports, deviation and change documents. Fields include reaction parameters, material identifiers, impurity content, deviation details, and others. Units involve degrees Celsius, hours, milligrams per kilogram, and similar measurements.
What constraints do these characteristics impose on the "workflow orchestration" link
Data for this category comes from multiple sources and updates per batch. This requires workflows to support triggering execution based on production batches to avoid repeated or missed processing. The need to connect multiple systems requires workflow nodes to support docking with different types of data source nodes, and configure nodeId to accurately identify the return data of each node. The complex document structure and multi-dimensional fields require workflows to configure variables to split and extract different business fields, avoiding context overload. At the same time, process abnormalities may lead to batch scrapping. Workflows must embed pre-check links to intercept non-compliant input data.
How to set the configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Variable Definition | Split variables by production batch, including material batch number, reaction temperature, impurity test value | Documents for this category must be bound to batch information. Splitting variables allows accurate matching of corresponding process data |
| Node ID Configuration | Assign a unique nodeId to each data source node | Accurately identify return data from different nodes when connecting multiple systems, avoiding field confusion |
| Single Search Limit | Top 5 entries | The top 5 entries in the associated dataset of process documents cover core process and exception information |
| Context Length Limit | 800–1200 characters | Avoid context overflow caused by long text, and adapt to the single-step record length of process documents |
| Exception Check Node | Calibrate based on actual testing | Must match business rules such as impurity thresholds and material balance deviation ranges for this category |
| Retry Count | 3 times | Address temporary network fluctuations when connecting multiple systems, avoiding process interruptions caused by a single failed call |
Three common mistakes
- Phenomenon: Workflows cannot support users choosing whether to skip knowledge base retrieval. Every workflow run triggers retrieval, consuming additional computing resources. Cause: No optional retrieval switch variable configured in
variables, and no conditional branch logic added to workflow nodes. - Phenomenon: Field empty errors occur after workflow execution, and the generated exception reports lack impurity profile data. Cause: Failed to split
variablesby batch to extract impurity test fields from the LIMS system, leading to incorrect data source field mapping. - Phenomenon: Workflow execution times out, failing to complete process document generation within the required time frame. Cause: No single search limit configured, leading to retrieval of too many irrelevant documents and excessive processing time.
How to confirm the configuration is correct
- Manually trigger a test batch workflow to check if the process starts according to the configured trigger rules.
- View the workflow variable configuration to confirm that it includes the material, reaction, and impurity-related fields required for this category.
- Simulate triggering an exception scenario to verify that the conditional branch correctly executes skip or retrieval logic.
- View workflow execution logs to confirm that node call counts and retry counts match the configured requirements.
The values provided are common starting points and should be measured against the reader's own samples.