Page History

Work with script tasks and classes to create machine learning models.

How-to → Examples → Feature → Application → Tasks and Classes

Summary

This solution example demonstrates the basic concepts involved in using machine learning models. The algorithms presented include:

• Regression

• Binary Classification

• Anomaly Detection

It runs machine learning models to predict the qualityIndex (Regression), determine whether the current process variable values pass quality control (Binary Classification), and detect whether any anomalies are present (Anomaly Detection).

Technical Information

Requirements

• .NET 8.0

• Multiplatform FrameworX solution

• Understanding of machine learning concepts.

• In Scripts / Classes, the class that contains the machine learning model must include the ML.NET dependencies. To simplify this, when creating a new class, select “Add ML.NET namespaces.” (it was already done in this example)

Solution explained using The Four Pillars Methodology

The explanation below focuses only on the Regression model; however, the same logic applies to the other two models. Only the most relevant methods are described. Elements such as variable declarations are not explained in detail but can be reviewed directly in the example. In this example, the models were pre-created in a Microsoft Visual Studio environment. The models were exported as .zip files and used directly within FrameworX. Although it is also possible to train the models inside FrameworX, reinforcing that all dependencies used must support .NET 8.0.

Pillar 1: Data Foundation (Unified Namespace)

Under Unified Namespace / Tags, you will find the Motor PVs, as well as the input and output tags used by the models. All members of these tags can also be reviewed under Unified Namespace / UserTypes.

The data is described in the table below:

Pillar 2: Industrial Operations (Process Modules)

Under Devices, the ValueSimulator protocol is configured to simulate field data. In Devices / Points, you can view the range of random values generated for each variable in the Address column. In Devices / AccessTypes, the Read AccessType being used has a polling rate of 5 seconds; in this specific driver, this means that new values are generated every 5 seconds.

Pillar 3: Business Operations (Application Modules)

This is where the ML.NET code is implemented. In Scripts / Tasks, there is a RunMLTask, which is responsible only for calling the models and running predictions automatically.

1. In Scripts / Classes, there are three machine learning model implementations. There is also a class named RunML, which is responsible solely for invoking these models.

float quality = await @Script.Class.Regression.PreviewQuality((float)temperature, (float)pressure);

2. Load the model:

var pathModel = @"<modelPath>";
trainedModel = mlContext.Model.Load(path, out modelSchema);

3. Instance the class with the inputs:

var newManufacturingData = new ManufacturingQualityInputData()
{
	Temperature = temperature,
	Pressure = pressure
};

4. Do the prediction calling the ML.NET methods:

var predEngine = mlContext.Model.CreatePredictionEngine<ManufacturingQualityInputData, ManufacturingPredictionResult>(trainedModel);

return predEngine.Predict(newManufacturing);

Pillar 4: User Interaction (Operator UI Modules)

On the main page in Displays / Draw, you can see three tables displaying the outputs of each machine learning model, along with motor symbols below them to indicate whether the model output is considered good (green) or not (red).

• Regression:
  The output is Quality. A value above 80 is considered good.

• Binary Classification:
  The outputs are PassQualityControl and Probability. A value of 1 for PassQualityControl is considered good, and a higher Probability indicates better confidence.

• Anomaly Detection:
  The outputs are IsAbnormal and Score. A result of Normal is considered good. The higher the Score, the more certain the model is that the condition is abnormal.

More details about The Four Pillars Methodology.

In this section...