✍️ Simulation Assessment

The Simulation Assessment section allows you to select the soft-sensor model you want to evaluate and define the time range using a slider, based on the data from the experiment ID selected in the Data Source section. This functionality enables you to inspect model predictions through an interactive visualization and track the behavior of one or more model variables alongside the predicted outputs.

Additionally, the section provides tools to search for specific values within the monitored variables. This feature makes it easy to identify anomalies, document these unusual values, and generate reports summarizing the anomalous behavior of the monitored model.

The Simulation Assessment section provides a comprehensive interface for evaluating soft sensor performance under various scenarios.

1. Initial panel

Upon launching the STAMM dashboard, you are presented with a form that allows you to:

Select a monitoring soft sensor from a dropdown menu listing all models available in the Model Repository.
Specify a time window to define the range of data points considered for the experiment currently selected in the Data Source section.
Add variables to monitor, enabling you to track additional inputs that the selected model considers, alongside its predictions.

This section is designed to give users flexibility in configuring simulations, ensuring that the evaluation reflects the conditions and variables most relevant to your analysis.

Once the information from the previous panel has been entered, a chart will be displayed that allows you to compare the behavior of selected variables against the model's predictions over the chosen time window.

2. Data table: filtering and anomaly reporting

Upon entering the information in the initial panel, a table is displayed showing the data used in the chart. This includes the values of the model's monitored variables, the model's prediction values, and their corresponding timestamps.

The table offers two key functionalities:

Value Filtering — filter the data based on specific criteria to focus on relevant subsets.
Anomaly Reporting — identify and report anomalies in the data, supporting deeper analysis and quality assurance.

Value Filtering — helps you locate specific values within the dataset used by the chart. You can search by variable, timestamp, or model value, making it easier to focus on relevant data points.

The filter supports a variety of operators such as >, <, >=, <=, as well as partial text searches, including dates. This functionality allows for precise and flexible data exploration, enabling you to quickly identify patterns or anomalies within the dataset.

Anomaly Reporting — the table includes a button that allows you to report anomalies detected in either the model's prediction values or the monitored variables. When you click this button, a popup form appears, similar to the example shown below:

After completing and submitting the form, a confirmation is displayed:

3. Reports

Once anomalies have been recorded, STAMM can produce multi-sheet reports that bring together the chart snapshot, the flagged entries, and metadata about the selected model and time window.