The Virtual Sensor API is used to detect faults in a variable of importance. This could be for instance to detect overheating in electrical assets or to detect thermal anomalies in HVAC process.

It leverages supervised machine learning techniques to learn the relationship between the target variable, a set of heterogeneous available variables, and to identify drift between the predicted value of the target and its actual variable.

This API should be used when the target variable is supposed to have an underlying pattern and when measurements of the target variable are always available.

This document provides a general tutorial for users who want to consume the Virtual Sensor Fault Detection API.

The Virtual Sensor Fault Detection API includes 4 features:

• Learn the model (the relationship between the input data and the target on which to detect faults). It requires data with measurement of the target. (createModel endpoint).

• Detect fault by comparing the target variable value and its estimated value by the model. It requires the model ID of the corresponding model and the measured value of the target (applyModelAndDetectFaults endpoint).

• Update an existing model by feeding it with new data (updateModel endpoint).

• Get information on an existing model (getModelInformation endpoint).

Take the tour and experiment on your own data prior to any development through our hands-on application !

try-analytics-se.azurewebsites.net

The createModel endpoint creates a virtual sensor model. The created virtual sensor model uses information available from driver measurements and process parameters to model the relationship between the drivers and the target.

The principle of this service is the following:

• Historical training data depicting how a system works is available.
• A target variable, i.e., data of special importance for which we want to build a virtual sensor model, is known.
• Some pattern in the data is suspected, i.e. some relationship between some of the available data and the target can be observed.
• This pattern can be used to create a virtual sensor model to generate predictions.

The model can be later on applied to predict the target. Comparing the real measurements to the predicted target, we can identify drift from normal behavior.

The applyModelAndDetectFaults endpoint applies the model over test data containing the target variable, and uses the result as a reference to detect anomalies in the input signal.

The updateModel endpoint updates the model created with the createModel service. The input data should be the same variables as the ones used to create the initial model.

This service returns information on an existing prediction model that is securely stored on the Schneider-Electric platform.

Amount of calls to the API in the SANDBOX are limited.

To get a full experience and extend the thresholds, please subscribe to the API and use the production environment.

Run in Postman collection

First, you need to download the free Postman application for your operating system. Postman currently supports Mac, Windows, and Linux. Once you have Postman installed, you can import the Forecasting API Postman collection by clicking the orange “Run in Postman” button. At this point, you should see the Forecasting API folder in the left-hand column under the Collections tab.

Step 1: Setup environment

Step 2: Prepare Data Step 3: Create a new energy model Step 4: Train model Step 5: Apply an existing forecasting model to new data Step 6: Get model information on an existing prediction model

This API uses API keys to authenticate and allow access to the API.

Your API keys carry many privileges, so be sure to keep them secure!

The API key should be included in all API requests in the Authorization header that looks like the following:

| Authorization: your_subscription_key

For steps on how to generate your_subscription_key please refer to the Features tab.

We follow the error response format proposed in RFC 7807also known as Problem Details for HTTP APIs. As with our normal API responses, your client must be prepared to gracefully handle additional members of the response.

Contact the SE Exchange support team at exchange.support@se.com and include:

• Endpoint URL
• Request/Response
• Any additional information like Screenshots , Postman collections.