As we learned in Chapter 13, Understanding Diagnostics, Maintenance, and Predictive Analytics, to develop an analytics model, we can follow five main steps:

1. Defining a problem statement and preparing the data
2. Exploratory Data Analysis (EDA)
3. Building the model—either physics-based or data-driven

4. Packaging and deployment
5. Monitoring

In the previous example, we covered all of these steps in our creation of a physics-based digital twins model. When we work with a data-driven approach, however, we need an additional sub-step to train and test the model using real data. We can use the machine learning abilities and high computational abilities (both GPU-and CPU-based) provided by Azure ML to accomplish this goal.

To demonstrate this, we will develop a simple logistic regression model to identify the anomalies of our wind turbine based on experimental data.

We will use the wind turbine digital twins to produce simulated data to infer a surrogate model. The purpose of our exercise is not to find the best algorithm, but simply to use the computational capabilities of the Azure ML service.

We will use the Python ML library scikit-learn.